While vision-language models (VLMs) have demonstrated remarkable performance across various tasks combining textual and visual information, they continue to struggle with fine-grained visual perception tasks that require detailed pixel-level analysis. Effectively eliciting comprehensive reasoning from VLMs on such intricate visual elements remains an open challenge. In this paper, we present VipAct, an agent framework that enhances VLMs by integrating multi-agent collaboration and vision expert models, enabling more precise visual understanding and comprehensive reasoning. VipAct consists of an orchestrator agent, which manages task requirement analysis, planning, and coordination, along with specialized agents that handle specific tasks such as image captioning and vision expert models that provide high-precision perceptual information. This multi-agent approach allows VLMs to better perform fine-grained visual perception tasks by synergizing planning, reasoning, and tool use. We evaluate VipAct on benchmarks featuring a diverse set of visual perception tasks, with experimental results demonstrating significant performance improvements over state-of-the-art baselines across all tasks. Furthermore, comprehensive ablation studies reveal the critical role of multi-agent collaboration in eliciting more detailed System-2 reasoning and highlight the importance of image input for task planning. Additionally, our error analysis identifies patterns of VLMs' inherent limitations in visual perception, providing insights into potential future improvements. VipAct offers a flexible and extensible framework, paving the way for more advanced visual perception systems across various real-world applications.

Rui Wang, Junda Wu, Yu Xia, Tong Yu, Ruiyi Zhang, Ryan A. Rossi, Subrata Mitra, Lina Yao, Julian McAuley. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Recent advancements in diffusion models have shown promising results in sequential recommendation (SR). Existing approaches predominantly rely on implicit conditional diffusion models, which compress user behaviors into a single representation during the forward diffusion process. While effective to some extent, this oversimplification often leads to the loss of sequential and contextual information, which is critical for understanding user behavior. Moreover, explicit information, such as user-item interactions or sequential patterns, remains underutilized, despite its potential to directly guide the recommendation process and improve precision. However, combining implicit and explicit information is non-trivial, as it requires dynamically integrating these complementary signals while avoiding noise and irrelevant patterns within user behaviors. To address these challenges, we propose Dual Conditional Diffusion Models for Sequential Recommendation (DCRec), which effectively integrates implicit and explicit information by embedding dual conditions into both the forward and reverse diffusion processes. This allows the model to retain valuable sequential and contextual information while leveraging explicit user-item interactions to guide the recommendation process. Specifically, we introduce the Dual Conditional Diffusion Transformer (DCDT), which dynamically integrate both implicit and explicit signals throughout the diffusion stages, ensuring contextual understanding and minimizing the influence of irrelevant patterns. Extensive experiments on public benchmark datasets demonstrate that DCRec significantly outperforms state-of-the-art methods.

Agentic recommendations cast recommenders as large language model (LLM) agents that can plan, reason, use tools, and interact with users of varying preferences in web applications. However, most existing agentic recommender systems focus on generic single-agent plan-execute workflows or multi-agent task decomposition pipelines. Without recommendation-oriented design, they often underuse the collaborative signals in the user–item interaction history, leading to unsatisfying recommendation results. To address this, we propose the Multi-Agent Collaborative Filtering (MACF) framework for agentic recommendations, drawing an analogy between traditional collaborative filtering algorithms and LLM-based multi-agent collaboration. Specifically, given a target user and query, we instantiate similar users and relevant items as LLM agents with unique profiles. Each agent is able to call retrieval tools, suggest candidate items, and interact with other agents. Different from the static preference aggregation in traditional collaborative filtering, MACF employs a central orchestrator agent to adaptively manage the collaboration between user and item agents via dynamic agent recruitment and personalized collaboration instruction. Experimental results on datasets from three different domains show the advantages of our MACF framework compared to strong agentic recommendation baselines.

Abstract Since the SciCap dataset’s launch in 2021, the research community has made significant progress in generating captions for scientific figures in scholarly articles. In 2023, the first SciCap Challenge took place, inviting global teams to use an expanded SciCap dataset to develop models for captioning diverse figure types across various academic fields. At the same time, text generation models advanced quickly, with many powerful pre-trained large multimodal models (LMMs) emerging that showed impressive capabilities in various vision-and-language tasks. This paper presents an overview of the first SciCap Challenge and details the performance of various models on its data, capturing a snapshot of the field’s state. We found that professional editors overwhelmingly preferred figure captions generated by GPT-4V over those from all other models and even the original captions written by authors. Following this key finding, we conducted detailed analyses to answer this question: Have advanced LMMs solved the task of generating captions for scientific figures?

Navigation is a fundamental task in the research of Embodied AI, and recent advances in machine learning algorithms have garnered growing interest in developing versatile Embodied AI systems. However, current research in this domain reveals opportunities for improvement. First, the direct application of RNNs and Transformers often overlooks the distinct characteristics of navigation tasks compared to traditional sequential data modeling. These methods are inherently designed to capture long-term dependencies, which are relatively weak in navigation scenarios, potentially limiting their performance in such tasks. Second, the reliance on task-specific configurations, such as pre-trained modules and dataset-specific logic, compromises the generalizability of these methods. We address these constraints by initially exploring the unique differences between Navigation tasks and other sequential data tasks through the lens of Causality, presenting a causal framework to elucidate the inadequacies of conventional sequential methods for Navigation. By leveraging this causal perspective, we propose Causality-Aware Transformer (CAT) Networks for Navigation, featuring a Causal Understanding Module to enhance the model’s Environmental Understanding capability. Meanwhile, our method is devoid of task-specific inductive biases and can be trained in an End-to-End manner, which enhances the method’s generalizability across various contexts. Empirical evaluations demonstrate that our methodology consistently surpasses benchmark performances across a spectrum of settings, tasks, and simulation environments, specifically, in Object Navigation within RoboTHOR, Objective Navigation, Point Navigation in Habitat, and R2R Navigation. Extensive ablation studies reveal that the performance gains can be attributed to the Causal Understanding Module, which demonstrates effectiveness and efficiency in both Reinforcement Learning and Supervised Learning settings. Additionally, further analysis highlights the robustness of our method, demonstrating its capacity to consistently perform well across diverse experimental settings and varying conditions. This robustness underscores the adaptability and generalizability of our approach, reinforcing its potential for application across a wide range of tasks.

Transformers have emerged as the leading architecture in deep learning, proving to be versatile and highly effective across diverse domains beyond language and image processing. However, their impressive performance often incurs high computational costs due to their substantial model size. This paper focuses on compressing decoder-only transformer-based autoregressive models through structural weight pruning to improve the model efficiency while preserving performance for both language and image generation tasks. Specifically, we propose a training-free pruning method that calculates a numerical score with Newton's method for the Attention and MLP modules, respectively. Besides, we further propose another compensation algorithm to recover the pruned model for better performance. To verify the effectiveness of our method, we provide both theoretical support and extensive experiments. Our experiments show that our method achieves state-of-the-art performance with reduced memory usage and faster generation speeds on GPUs.

Yu Xia, Subhojyoti Mukherjee, Zhouhang Xie, Junda Wu, Xintong Li, Ryan Aponte, Hanjia Lyu, Joe Barrow, Hongjie Chen, Franck Dernoncourt, Branislav Kveton, Tong Yu, Ruiyi Zhang, Jiuxiang Gu, Nesreen K. Ahmed, Yu Wang, Xiang Chen, Hanieh Deilamsalehy, Sungchul Kim, Zhengmian Hu, Yue Zhao, Nedim Lipka, Seunghyun Yoon, Ting-Hao Kenneth Huang, Zichao Wang, Puneet Mathur, Soumyabrata Pal, Koyel Mukherjee, Zhehao Zhang, Namyong Park, Thien Huu Nguyen, Jiebo Luo, Ryan A. Rossi, Julian McAuley. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Change point detection in time series aims to identify moments when the probability distribution of time series changes. It is widely applied in many areas, such as human activity sensing and medical science. In the context of multivariate time series, this typically involves examining the joint distribution of multiple variables: If the distribution of any one variable changes, the entire time series undergoes a distribution shift. However, in practical applications, we may be interested only in certain components of the time series, exploring abrupt changes in their distributions while accounting for the presence of other components. Here, assuming an underlying structural causal model that governs the time-series data generation, we address this task by proposing a two-stage non-parametric algorithm that first learns parts of the causal structure through constraint-based discovery methods, and then employs conditional relative Pearson divergence estimation to identify the change points. The conditional relative Pearson divergence quantifies the distribution difference between consecutive segments in the time series, while the causal discovery method allows a focus on the causal mechanism, facilitating access to independent and identically distributed (IID) samples. Theoretically, the typical assumption of samples being IID in conventional change point detection methods can be relaxed based on the Causal Markov Condition. Through experiments on both synthetic and real-world datasets, we validate the correctness and utility of our approach.

Retrieval-augmented large language models (LLMs) have been remarkably competent in various NLP tasks. However, it was observed by previous works that retrieval is not always helpful, especially when the LLM is already knowledgable on the query to answer. Motivated by this, Adaptive Retrieval-Augmented Generation (ARAG) studies retrieving only when the knowledge asked by the query is absent in the LLM. Previous works of ARAG either require accessing the pre-training corpus or prompting with additional model inferences. Aiming to avoid such drawbacks, we propose to determine whether the model is knowledgeable on a query via inspecting the (contextualized) pre-trained token embeddings of LLMs. We hypothesize that such embeddings capture rich information on the model’s intrinsic knowledge base, which enables an efficient way of judging the necessity to retrieve from an external corpus. Extensive experiments demonstrate our ARAG approach’s superior performance across various benchmarks.

Chain-of-Thought (CoT) has been a widely adopted prompting method, eliciting impressive reasoning abilities of Large Language Models (LLMs). Inspired by the sequential thought structure of CoT, a number of Chain-of-X (CoX) methods have been developed to address challenges across diverse domains and tasks. In this paper, we provide a comprehensive survey of Chain-of-X methods for LLMs in different contexts. Specifically, we categorize them by taxonomies of nodes, i.e., the X in CoX, and application tasks. We also discuss the findings and implications of existing CoX methods, as well as potential future directions. Our survey aims to serve as a detailed and up-to-date resource for researchers seeking to apply the idea of CoT to broader scenarios.

Image Difference Captioning (IDC) aims to generate natural language descriptions that highlight subtle differences between two visually similar images.While recent advances leverage pre-trained vision-language models to align fine-grained visual differences with textual semantics, existing supervised approaches often overly focus on dataset-specific language patterns and fail to capture fine-grained and context-aware preferences on IDC, due to limited annotation diversity and a lack of semantically informative negative examples during training, To address these limitations, we propose an adversarial direct preference optimization (ADPO) framework for IDC, which formulates IDC as a preference optimization problem under the Bradley-Terry-Luce model, directly aligning the captioning policy with pairwise difference preferences via Direct Preference Optimization (DPO).To model more accurate and diverse IDC preferences, we introduce an adversarially trained hard negative retriever that selects counterfactual captions, This results in a minimax optimization problem, which we solve via policy-gradient reinforcement learning, enabling the policy and retriever to improve jointly.By dynamically generating semantically challenging negatives, our method reduces reliance on dataset-specific patterns.Experiments on benchmark IDC datasets show that our approach outperforms existing baselines, especially in generating fine-grained and accurate difference descriptions.

Xintong Li, Junda Wu, Tong Yu, Rui Wang, Yu Wang, Xiang Chen, Jiuxiang Gu, Lina Yao, Julian McAuley, Jingbo Shang. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Large Language Model (LLM) agents are commonly tuned with supervised finetuning on ReAct-style expert trajectories or preference optimization over pairwise rollouts.Most of these methods focus on imitating specific expert behaviors or promoting chosen reasoning thoughts and actions over rejected ones.However, without reasoning and comparing over alternative actions, LLM agents finetuned with these methods may over-commit towards seemingly plausible but suboptimal actions due to limited action space exploration.To address this, in this paper we propose Self-taught Ac-tioN Deliberation (SAND) framework, enabling LLM agents to explicitly deliberate over candidate actions before committing to one.To tackle the challenges of when and what to deliberate given large action space and step-level action evaluation, we incorporate self-consistency action sampling and execution-guided action critique to help synthesize step-wise action deliberation thoughts using the base model of the LLM agent.In an iterative manner, the deliberation trajectories are then used to finetune the LLM agent itself.Evaluating on two representative interactive agent tasks, SAND achieves an average 20% improvement over supervised finetuning on initial expert data and also outperforms state-of-the-art agent tuning approaches.

Multimodal large language models (MLLMs) have recently shown great progress in text-rich image understanding, yet they still struggle with complex, multi-page visually-rich documents. Traditional methods using document parsers for retrieval-augmented generation suffer from performance and efficiency limitations, while directly presenting all pages to MLLMs leads to inefficiencies, especially with lengthy ones. In this work, we present a novel framework named **S**elf-**V**isual **R**etrieval-**A**ugmented **G**eneration (SV-RAG), which can broaden horizons of *any* MLLM to support long-document understanding. We demonstrate that **MLLMs themselves can be an effective multimodal retriever** to fetch relevant pages and then answer user questions based on these pages. SV-RAG is implemented with two specific MLLM adapters, one for evidence page retrieval and the other for question answering. Empirical results show state-of-the-art performance on public benchmarks, demonstrating the effectiveness of SV-RAG.

Offline evaluation of LLMs is crucial in understanding their capacities, though current methods remain underexplored in existing research. In this work, we focus on the offline evaluation of the chain-of-thought capabilities and show how to optimize LLMs based on the proposed evaluation method. To enable offline feedback with rich knowledge and reasoning paths, we use knowledge graphs (KGs) (e.g., Wikidata5M) to provide feedback on the generated chain of thoughts. Due to the heterogeneity between LLM reasoning and KG structures, direct interaction and feedback from knowledge graphs on LLM behavior are challenging, as they require accurate entity linking and grounding of LLM-generated chains of thought in the KG. To address the above challenge, we propose an offline chain-of-thought evaluation framework, OCEAN, which models chain-of-thought reasoning in LLMs as a Markov Decision Process (MDP), and evaluate the policy’s alignment with KG preference modeling. To overcome the reasoning heterogeneity and grounding problems, we leverage on-policy KG exploration and reinforcement learning to model a KG policy that generates token-level likelihood distributions for LLM-generated chain-of-thought reasoning paths, simulating KG reasoning preference. Then we incorporate the knowledge-graph feedback on the validity and alignment of the generated reasoning paths into inverse propensity scores and propose KG-IPS estimator. Theoretically, we prove the unbiasedness of the proposed KG-IPS estimator and provide a lower bound on its variance. With the off-policy evaluated value function, we can directly enable off-policy optimization to further enhance chain-of-thought alignment. Our empirical study shows that OCEAN can be efficiently optimized for generating chain-of-thought reasoning paths with higher estimated values without affecting LLMs’ general abilities in downstream tasks or their internal knowledge.

Offline-to-online Reinforcement Learning (O2O RL) aims to perform online fine-tuning on an offline pre-trained policy to minimize costly online interactions. Existing work used offline datasets to generate data that conform to the online data distribution for data augmentation. However, generated data still exhibits a gap with the online data, limiting overall performance. To address this, we propose a new data augmentation approach, Classifier-Free Diffusion Generation (CFDG). Without introducing additional classifier training overhead, CFDG leverages classifier-free guidance diffusion to significantly enhance the generation quality of offline and online data with different distributions. Additionally, it employs a reweighting method to enable more generated data to align with the online data, enhancing performance while maintaining the agent's stability. Experimental results show that CFDG outperforms replaying the two data types or using a standard diffusion model to generate new data. Our method is versatile and can be integrated with existing offline-to-online RL algorithms. By implementing CFDG to popular methods IQL, PEX and APL, we achieve a notable 15\% average improvement in empirical performance on the D4RL benchmark such as MuJoCo and AntMaze.

For question-answering (QA) tasks, in-context learning (ICL) enables language models (LMs) to generate responses without modifying their parameters by leveraging examples provided in the input. However, the effectiveness of ICL heavily depends on the availability of high-quality examples, which are often scarce due to data privacy constraints, annotation costs, and distribution disparities. A natural solution is to utilize examples stored on client devices, but existing approaches either require transmitting model parameters—incurring significant communication overhead—or fail to fully exploit local datasets, limiting their effectiveness. To address these challenges, we propose Federated In-Context Learning (Fed-ICL), a general framework that enhances ICL through an iterative, collaborative process. Fed-ICL progressively refines responses by leveraging multi-round interactions between clients and a central server, improving answer quality without the need to transmit model parameters. We establish theoretical guarantees for the convergence of Fed-ICL and conduct extensive experiments on standard QA benchmarks, demonstrating that our proposed approach achieves strong performance while maintaining low communication costs.

Offline reinforcement learning (RL) presents distinct challenges as it relies solely on observational data. A central concern in this context is ensuring the safety of the learned policy by quantifying uncertainties associated with various actions and environmental stochasticity. Traditional approaches primarily emphasize mitigating epistemic uncertainty by learning risk-averse policies, often overlooking environmental stochasticity. In this study, we propose an uncertainty-aware distributional offline RL method to simultaneously address both epistemic uncertainty and environmental stochasticity. We propose a model-free offline RL algorithm capable of learning risk-averse policies and characterizing the entire distribution of discounted cumulative rewards, as opposed to merely maximizing the expected value of accumulated discounted returns. Our method is rigorously evaluated through comprehensive experiments in both risk-sensitive and risk-neutral benchmarks, demonstrating its superior performance.

Visual Language Navigation (VLN) is a fundamental task within the field of Embodied AI, focusing on the ability of agents to navigate complex environments based on natural language instructions. Despite the progress made by existing methods, these methods often present some common challenges. First, they rely on pre-trained backbone models for visual perception, which struggle with the dynamic viewpoints in VLN scenarios. Second, the performance is limited when using pre-trained LLMs or VLMs without fine-tuning, due to the absence of VLN domain knowledge. Third, while fine-tuning LLMs and VLMs can improve results, their computational costs are higher than those without fine-tuning. To address these limitations, we propose Weakly-supervised Partial Contrastive Learning (WPCL), a method that enhances an agent’s ability to identify objects from dynamic viewpoints in VLN scenarios by effectively integrating pre-trained VLM knowledge into the perception process, without requiring VLM fine-tuning. Our method enhances the agent’s ability to interpret and respond to environmental cues while ensuring computational efficiency. Experimental results have shown that our method outperforms the baseline methods on multiple benchmarks, which validates the effectiveness, robustness, and generalizability of our method.

Yeonjun In, Sungchul Kim, Ryan A. Rossi, Mehrab Tanjim, Tong Yu, Ritwik Sinha, Chanyoung Park. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Yu Xia, Junda Wu, Sungchul Kim, Tong Yu, Ryan A. Rossi, Haoliang Wang, Julian McAuley. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Forecasting multi-step user behavior trajectories requires reasoning over structured preferences across future actions, a challenge overlooked by traditional sequential recommendation. This problem is critical for applications such as personalized commerce and adaptive content delivery, where anticipating a user’s complete action sequence enhances both satisfaction and business outcomes. We identify an essential limitation of existing paradigms: their inability to capture global, listwise dependencies among sequence items. To address this, we formulate User Behavior Trajectory Prediction (UBTP) as a new task setting that explicitly models longterm user preferences. We introduce Listwise Preference Diffusion Optimization (LPDO), a diffusion-based training framework that directly optimizes structured preferences over entire item sequences. LPDO incorporates a Plackett–Luce supervision signal and derives a tight variational lower bound aligned with listwise ranking likelihoods, enabling coherent preference generation across denoising steps and overcoming the independent-token assumption of prior diffusion methods. To rigorously evaluate multi-step prediction quality, we propose the task-specific metric: Sequential Match (SeqMatch), which measures exact trajectory agreement, and adopt Perplexity (PPL), which assesses probabilistic fidelity. Extensive experiments on real-world user behavior benchmarks demonstrate that LPDO consistently outperforms state-of-the-art baselines, establishing a new benchmark for structured preference learning with diffusion models.

Diffusion models have demonstrated exceptional capabilities in generating a broad spectrum of visual content, yet their proficiency in rendering text is still limited: they often generate inaccurate characters or words that fail to blend well with the underlying image. To address these shortcomings, we introduce a novel framework named ARTIST, which incorporates a dedicated textual diffusion model to focus on the learning of text structures specifically. Initially, we pretrain this textual model to capture the intricacies of text representation. Subsequently, we finetune a visual diffusion model, enabling it to assimilate textual structure information from the pretrained textual model. This disentangled architecture design and training strategy significantly enhance the text rendering ability of the diffusion models for text-rich image generation. Additionally, we leverage the capabilities of pretrained large language models to interpret user intentions better, contributing to improved generation quality. Empirical results on the MARIO-Eval benchmark underscore the effectiveness of the proposed method, showing an improvement of up to 15% in various metries.

Visualization recommendation aims to enable rapid visual analysis of massive datasets. In real-world scenarios, it is essential to quickly gather and comprehend user preferences to cover users from diverse backgrounds, including varying skill levels and analytical tasks. Previous approaches to personalized visualization recommendations are non-interactive and rely on initial user data for new users. As a result, these models cannot effectively explore options or adapt to real-time feedback. To address this limitation, we propose an interactive personalized visualization recommendation (PVisRec) system that learns on user feedback from previous interactions. For more interactive and accurate recommendations, we propose Hier-SUCB, a contextual combinatorial semi-bandit in the PVisRec setting. Theoretically, we show an improved overall regret bound with the same rank of time but an improved rank of action space. We further demonstrate the effectiveness of Hier-SUCB through extensive experiments where it is comparable to offline methods and outperforms other bandit algorithms in the setting of visualization recommendation.

Retrieval-Augmented Generation (RAG) is often used with Large Language Models (LLMs) to infuse domain knowledge or user-specific information. In RAG, given a user query, a retriever extracts chunks of relevant text from a knowledge base. These chunks are sent to an LLM as part of the input prompt. Typically, any given chunk is repeatedly retrieved across user questions. However, currently, for every question, attention layers in LLMs fully compute the Keys and Values (KVs) repeatedly for the input chunks, as state-of-the-art methods cannot reuse KV-caches when chunks appear at arbitrary locations or with arbitrary contexts. Naive reuse leads to output quality degradation. This leads to potentially redundant computations on expensive GPUs and increases latency. In this work, we propose Cache-Craft , a system for managing and reusing precomputed KVs corresponding to the text chunks (which we call chunk-caches ) in RAG-based systems. We present how to identify chunk-caches that are reusable, how to efficiently perform a small fraction of recomputation to fix the cache and maintain output quality, and how to efficiently store and evict chunk-caches in the hardware for maximizing reuse while masking any overheads. With real production workloads as well as synthetic datasets, we show that Cache-Craft reduces redundant computation by 51% over SOTA prefix-caching and 75% over full recomputation. Additionally, with continuous batching on a real production workload, we get a 1.6× speed up in throughput for both the LLama-3-8B and 70B models and a 2.1× and 2× reduction in end-to-end response latency respectively, compared to prefix-caching, while maintaining generation quality.

Junda Wu, Tong Yu, Xiang Chen, Haoliang Wang, Ryan Rossi, Sungchul Kim, Anup Rao, Julian McAuley. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Kaige Xie, Tong Yu, Haoliang Wang, Junda Wu, Handong Zhao, Ruiyi Zhang, Kanak Mahadik, Ani Nenkova, Mark Riedl. Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2024.

Few-shot continual learning (FSCL) is a challenging setting as it requires models to learn new knowledge with a few examples over time, and fast adapt to new tasks without forgetting previous knowledge. Prompt-tuning, as an efficient learning approach for language models, has shown competitive performance in data-efficient learning for various NLP tasks, motivating us to explore how to effectively perform prompt-tuning in FSCL for text classification. In this work, we focus on studying prompt-tuning for continual classification, aiming to alleviate catastrophic forgetting and improve knowledge transfer with few-shot data in FSCL. After carefully analyzing the limited representation capability of existing soft-prompt initialization methods, we propose Task-Aware Initialization (TAI), a novel initialization approach that can combine the information from both context and label space. Extensive experiments with different language models including recent instruction-finetuned LLM in two FSCL settings (shot-invariant and shot-variant) demonstrate the superiority of TAI over current approaches.

While "instruction-tuned" generative large language models (LLMs) have demonstrated an impressive ability to generalize to new tasks, the training phases heavily rely on large amounts of diverse and high-quality instruction data (such as ChatGPT and GPT-4). Unfortunately, acquiring high-quality data, especially when it comes to human-written data, can pose significant challenges both in terms of cost and accessibility. Moreover, concerns related to privacy can further limit access to such data, making the process of obtaining it a complex and nuanced undertaking. To tackle this issue, our study introduces a new approach called Federated Instruction Tuning (FedIT), which leverages federated learning (FL) as the learning framework for the instruction tuning of LLMs. This marks the first exploration of FL-based instruction tuning for LLMs. This is especially important since text data is predominantly generated by end users. For example, collecting extensive amounts of everyday user conversations can be a useful approach to improving the generalizability of LLMs, allowing them to generate authentic and natural responses. Therefore, it is imperative to design and adapt FL approaches to effectively leverage these users’ diverse instructions stored on local devices while mitigating concerns related to the data sensitivity and the cost of data transmission. In this study, we leverage extensive qualitative analysis, including the prevalent GPT-4 auto-evaluation to illustrate how our FedIT framework enhances the performance of LLMs. Utilizing diverse instruction sets on the client side, FedIT outperforms centralized training with only limited local instructions.

Large language models (LLMs) have demonstrated powerful decision-making and planning capabilities in solving complicated real-world problems. LLM-based autonomous agents can interact with diverse tools (e.g., functional APIs) and generate solution plans that execute a series of API function calls in a step-by-step manner. The multitude of candidate API function calls significantly expands the action space, amplifying the critical need for efficient action space navigation. However, existing methods either struggle with unidirectional exploration in expansive action spaces, trapped into a locally optimal solution, or suffer from exhaustively traversing all potential actions, causing inefficient navigation. To address these issues, we propose ToolChain*, an efficient tree search-based planning algorithm for LLM-based agents. It formulates the entire action space as a decision tree, where each node represents a possible API function call involved in a solution plan. By incorporating the A$^*$ search algorithm with task-specific cost function design, it efficiently prunes high-cost branches that may involve incorrect actions, identifying the most low-cost valid path as the solution. Extensive experiments on multiple tool-use and reasoning tasks demonstrate that ToolChain* efficiently balances exploration and exploitation within an expansive action space. It outperforms state-of-the-art baselines on planning and reasoning tasks by 3.1% and 3.5% on average while requiring 7.35x and 2.31x less time, respectively.

Most real-world networks are noisy and incomplete samples from an unknown target distribution. Refining them by correcting corruptions or inferring unobserved regions typically improves downstream performance. Inspired by the impressive generative capabilities that have been used to correct corruptions in images, and the similarities between "in-painting" and filling in missing nodes and edges conditioned on the observed graph, we propose a novel graph generative framework, SGDM, which is based on subgraph diffusion. Our framework not only improves the scalability and fidelity of graph diffusion models, but also leverages the reverse process to perform novel, conditional generation tasks. In particular, through extensive empirical analysis and a set of novel metrics, we demonstrate that our proposed model effectively supports the following refinement tasks for partially observable networks: (T1) denoising extraneous subgraphs, (T2) expanding existing subgraphs and (T3) performing ``style" transfer by regenerating a particular subgraph to match the characteristics of a different node or subgraph.

The long-tail recommendation is a challenging task for traditional recommender systems, due to data sparsity and data imbalance issues. The recent development of large language models (LLMs) has shown their abilities in complex reasoning, which can help to deduce users' preferences based on very few previous interactions. However, since most LLM-based systems rely on items' semantic meaning as the sole evidence for reasoning, the collaborative information of user-item interactions is neglected, which can cause the LLM's reasoning to be misaligned with task-specific collaborative information of the dataset. To further align LLMs' reasoning to task-specific user-item interaction knowledge, we introduce collaborative retrieval-augmented LLMs, CoRAL, which directly incorporate collaborative evidence into the prompts. Based on the retrieved user-item interactions, the LLM can analyze shared and distinct preferences among users, and summarize the patterns indicating which types of users would be attracted by certain items. The retrieved collaborative evidence prompts the LLM to align its reasoning with the user-item interaction patterns in the dataset. However, since the capacity of the input prompt is limited, finding the minimally-sufficient collaborative information for recommendation tasks can be challenging. We propose to find the optimal interaction set through a sequential decision-making process and develop a retrieval policy learned through a reinforcement learning (RL) framework, CoRAL. Our experimental results show that CoRAL can significantly improve LLMs' reasoning abilities on specific recommendation tasks. Our analysis also reveals that CoRAL can more efficiently explore collaborative information through reinforcement learning.

Yu Xia, Tong Yu, Zhankui He, Handong Zhao, Julian McAuley, Shuai Li. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Yu Xia, Xu Liu, Tong Yu, Sungchul Kim, Ryan Rossi, Anup Rao, Tung Mai, Shuai Li. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

Large language models (LLMs) have demonstrated impressive capabilities in various reasoning tasks, aided by techniques like chain-of-thought prompting that elicits verbalized reasoning. However, LLMs often generate text with obvious mistakes and contradictions, raising doubts about their ability to robustly process and utilize generated rationales. In this work, we investigate reasoning in LLMs through the lens of internal representations, focusing on how these representations are influenced by generated rationales. Our preliminary analysis reveals that while generated rationales improve answer accuracy, inconsistencies emerge between the model’s internal representations in middle layers and those in final layers, potentially undermining the reliability of their reasoning processes. To address this, we propose internal consistency as a measure of the model’s confidence by examining the agreement of latent predictions decoded from intermediate layers. Extensive empirical studies across different models and datasets demonstrate that internal consistency effectively distinguishes between correct and incorrect reasoning paths. Motivated by this, we propose a new approach to calibrate reasoning by up-weighting reasoning paths with high internal consistency, resulting in a significant boost in reasoning performance. Further analysis uncovers distinct patterns in attention and feed-forward modules across layers, providing insights into the emergence of internal inconsistency. In summary, our results demonstrate the potential of using internal representations for self-evaluation of LLMs.

Large language models (LLMs) are known for their exceptional performance in natural language processing, making them highly effective in many human life-related tasks. The attention mechanism in the Transformer architecture is a critical component of LLMs, as it allows the model to selectively focus on specific input parts. The softmax unit, which is a key part of the attention mechanism, normalizes the attention scores. Hence, the performance of LLMs in various NLP tasks depends significantly on the crucial role played by the attention mechanism with the softmax unit. In-context learning is one of the celebrated abilities of recent LLMs. Without further parameter updates, Transformers can learn to predict based on few in-context examples. However, the reason why Transformers becomes in-context learners is not well understood. Recently, in-context learning has been studied from a mathematical perspective with simplified linear self-attention without softmax unit. Based on a linear regression formulation $\min_x\| Ax - b \|_2$, existing works show linear Transformers' capability of learning linear functions in context. The capability of Transformers with softmax unit approaching full Transformers, however, remains unexplored. In this work, we study the in-context learning based on a softmax regression formulation $\min_{x} \| \langle \exp(Ax), {\bf 1}_n \rangle^{-1} \exp(Ax) - b \|_2$. We show the upper bounds of the data transformations induced by a single self-attention layer with softmax unit and by gradient-descent on a $\ell_2$ regression loss for softmax prediction function. Our theoretical results imply that when training self-attention-only Transformers for fundamental regression tasks, the models learned by gradient-descent and Transformers show great similarity.

In recent years, the field of recommendation systems has witnessed significant advancements, with explainable recommendation systems gaining prominence as a crucial area of research. These systems aim to enhance user experience by providing transparent and compelling recommendations, accompanied by explanations. However, a persistent challenge lies in addressing biases that can influence the recommendations and explanations offered by these systems. Such biases often stem from a tendency to favor popular items and generate explanations that highlight their common attributes, thereby deviating from the objective of delivering personalized recommendations and explanations. While existing debiasing methods have been applied in explainable recommendation systems, they often overlook the model-generated explanations in tackling biases. Consequently, biases in model-generated explanations may persist, potentially compromising system performance and user satisfaction.

Web-based applications such as chatbots, search engines and news recommendations continue to grow in scale and complexity with the recent surge in the adoption of large language models (LLMs). Online model selection has thus garnered increasing attention due to the need to choose the best model among a diverse set while balancing task reward and exploration cost. Organizations faces decisions like whether to employ a costly API-based LLM or a locally finetuned small LLM, weighing cost against performance. Traditional selection methods often evaluate every candidate model before choosing one, which are becoming impractical given the rising costs of training and finetuning LLMs. Moreover, it is undesirable to allocate excessive resources towards exploring poor-performing models. While some recent works leverage online bandit algorithm to manage such exploration-exploitation trade-off in model selection, they tend to overlook the increasing-then-converging trend in model performances as the model is iteratively finetuned, leading to less accurate predictions and suboptimal model selections.

Abstract Rapid advancements of large language models (LLMs) have enabled the processing, understanding, and generation of human-like text, with increasing integration into systems that touch our social sphere. Despite this success, these models can learn, perpetuate, and amplify harmful social biases. In this article, we present a comprehensive survey of bias evaluation and mitigation techniques for LLMs. We first consolidate, formalize, and expand notions of social bias and fairness in natural language processing, defining distinct facets of harm and introducing several desiderata to operationalize fairness for LLMs. We then unify the literature by proposing three intuitive taxonomies, two for bias evaluation, namely, metrics and datasets, and one for mitigation. Our first taxonomy of metrics for bias evaluation disambiguates the relationship between metrics and evaluation datasets, and organizes metrics by the different levels at which they operate in a model: embeddings, probabilities, and generated text. Our second taxonomy of datasets for bias evaluation categorizes datasets by their structure as counterfactual inputs or prompts, and identifies the targeted harms and social groups; we also release a consolidation of publicly available datasets for improved access. Our third taxonomy of techniques for bias mitigation classifies methods by their intervention during pre-processing, in-training, intra-processing, and post-processing, with granular subcategories that elucidate research trends. Finally, we identify open problems and challenges for future work. Synthesizing a wide range of recent research, we aim to provide a clear guide of the existing literature that empowers researchers and practitioners to better understand and prevent the propagation of bias in LLMs.

We consider dynamic pricing strategies in a streamed longitudinal data set-up where the objective is to maximize, over time, the cumulative profit across a large number of customer segments. We consider a dynamic model with the consumers’ preferences as well as price sensitivity varying over time. Building on the well-known finding that consumers sharing similar characteristics act in similar ways, we consider a global shrinkage structure, which assumes that the consumers’ preferences across the different segments can be well approximated by a spatial autoregressive (SAR) model. In such a streamed longitudinal setup, we measure the performance of a dynamic pricing policy via regret, which is the expected revenue loss compared to a clairvoyant that knows the sequence of model parameters in advance. We propose a pricing policy based on penalized stochastic gradient descent (PSGD) and explicitly characterize its regret as functions of time, the temporal variability in the model parameters as well as the strength of the auto-correlation network structure spanning the varied customer segments. Our regret analysis results not only demonstrate asymptotic optimality of the proposed policy but also show that for policy planning it is essential to incorporate available structural information as policies based on unshrunken models are highly sub-optimal in the aforementioned set-up. We conduct simulation experiments across a wide range of regimes as well as real-world networks based studies and report encouraging performance for our proposed method.

Graph Neural Networks (GNNs) have become increasingly important due to their representational power and state-of-the-art predictive performance on many fundamental learning tasks. Despite this success, GNNs suffer from fairness issues that arise as a result of the underlying graph data and the fundamental aggregation mechanism that lies at the heart of the large class of GNN models. In this article, we examine and categorize fairness techniques for improving the fairness of GNNs. We categorize these techniques by whether they focus on improving fairness in the pre-processing, in-processing (during training), or post-processing phases. We discuss how such techniques can be used together whenever appropriate and highlight the advantages and intuition as well. We also introduce an intuitive taxonomy for fairness evaluation metrics, including graph-level fairness, neighborhood-level fairness, embedding-level fairness, and prediction-level fairness metrics. In addition, graph datasets that are useful for benchmarking the fairness of GNN models are summarized succinctly. Finally, we highlight key open problems and challenges that remain to be addressed.

Conversational recommendation systems (CRSs) are increasingly prevalent, but they are susceptible to the influence of corrupted user behaviors, such as deceptive click ratings. These behaviors can skew the recommendation process, resulting in suboptimal results. Traditional bandit algorithms, which are typically oriented to single users, do not capitalize on implicit social connections between users, which could otherwise enhance learning efficiency. Furthermore, they cannot identify corrupted users in a real-time, multi-user environment. In this paper, we propose a novel bandit problem, Online Learning and Detecting Corrupted Users (OLDCU), to learn and utilize unknown user relations from disrupted behaviors to speed up learning and detect corrupted users in an online setting. This problem is non-trivial due to the dynamic nature of user behaviors and the difficulty of online detection. To robustly learn and leverage the unknown relations among potentially corrupted users, we propose a novel bandit algorithm RCLUB-WCU, incorporating a conversational mechanism. This algorithm is designed to handle the complexities of disrupted behaviors and to make accurate user relation inferences. To detect corrupted users with bandit feedback, we further devise a novel online detection algorithm, OCCUD, which is based on RCLUB-WCU’s inferred user relations and designed to adapt over time. We prove a sub-linear regret bound for RCLUB-WCU, demonstrating its efficiency. We also analyze the detection accuracy of OCCUD, showing its effectiveness in identifying corrupted users. Through extensive experiments, we validate the performance of our methods. Our results show that RCLUB-WCU and OCCUD outperform previous bandit algorithms and achieve high corrupted user detection accuracy, providing robust and efficient solutions in the field of CRSs.

Exploratory data analytics (EDA) is a sequential decision making process where analysts choose subsequent queries that might lead to some interesting insights based on the previous queries and corresponding results. Data processing systems often execute the queries on samples to produce results with low latency. Different downsampling strategy preserves different statistics of the data and have different magnitude of latency reductions. The optimum choice of sampling strategy often depends on the particular context of the analysis flow and the hidden intent of the analyst. In this paper, we are the first to consider the impact of sampling in interactive data exploration settings as they introduce approximation errors. We propose a Deep Reinforcement Learning (DRL) based framework which can optimize the sample selection in order to keep the analysis and insight generation flow intact. Evaluations with real datasets show that our technique can preserve the original insight generation flow while improving the interaction latency, compared to baseline methods.

Modern software systems’ increasing complexity and scale makes it challenging to accurately detect system issues and outages, which have been tackled as an anomaly detection task. Conventionally, such anomalous events barely happen, and annotating them is time-consuming and impractical in big data streams. Even with automated anomaly detection, resolving issues promptly is a remaining challenge that can only be done by providing specific contexts such as root causes, target/affected services, and more. To address these fundamentally important problems, we present Grid Transformer (GT), a framework designed to detect and explain $\log$ anomalies in an unsupervised setting. We first train an Auto-Encoder model to generate pseudo labels. Then, we train the proposed grid transformer that not only predicts anomalies but also generates why a particular instance is an anomaly. Through extensive experiments, we demonstrate the effectiveness of our approach where it is shown to outperform the other $\log$ anomaly detection models by 20% while also able to generate time-wise and message-wise explanations of the anomalies.

Generative models have been widely studied in computer vision. Recently, diffusion models have drawn substantial attention due to the high quality of their generated images. A key desired property of image generative models is the ability to disentangle different attributes, which should enable modification towards a style without changing the semantic content, and the modification parameters should generalize to different images. Previous studies have found that generative adversarial networks (GANs) are inherently endowed with such disentanglement capability, so they can perform disentangled image editing without re-training or fine-tuning the network. In this work, we explore whether diffusion models are also inherently equipped with such a capability. Our finding is that for stable diffusion models, by partially changing the input text embedding from a neutral description (e.g., “a photo of person”) to one with style (e.g., “a photo of person with smile”) while fixing all the Gaussian random noises introduced during the denoising process, the generated images can be modified towards the target style without changing the semantic content. Based on this finding, we further propose a simple, light-weight image editing algorithm where the mixing weights of the two text embeddings are optimized for style matching and content preservation. This entire process only involves optimizing over around 50 parameters and does not fine-tune the diffusion model itself. Experiments show that the proposed method can modify a wide range of attributes, with the performance outperforming diffusion-model-based image-editing algorithms that require fine-tuning. The optimized weights generalize well to different images. Our code is publicly available at https://github.com/UCSB-NLP-Chang/DiffusionDisentanglement.

With the rapid increase of digital content like images or videos nowadays, compression technology contributes more to saving storage or transferring time with large-scale data. While some existing methods already achieved a great compression ratio, they are not applicable to certain live applications under low efficiency. In this work, we use massive parallelization to speed up the SOTA baseline FLIF, including bitwise-equivalent speedup and learning-based speedup. Our method achieves $38.7 \times$ throughputs for encoding and $2.45 \times$ throughputs for decoding, compared to the baseline FLIF.

Progressive image compression and syncing between devices is an important and challenging problem. When the users are collaboratively editing the same image online, they would expect the changes made by others to be instantly displayed on their side. Since such syncing can be very frequent and usually the image sizes are significantly larger than text data, image live co-editing cannot be easily achieved in the same way as those commonly seen in document co-editing tools. While previous compression techniques like PNG, JPEG and FLIF enable spatially progressive compression, they do not support content-aware compression. Thus, even though the image can be gradually displayed, users cannot prioritize the transmission and display of the most important bits of the image, and often times the resulting pixelation during syncing greatly hurts the user experience. Many existing works on saliency detection can be utilized to provide content awareness. However, many of those techniques are deep-learning-based and it would be computationally prohibitive to directly use them in a latency sensitive scenario like collaborative editing on client devices. In this work, we aim to find a middle ground between a good quality pixel prioritization strategy and extremely fast compression. We start with the pipeline proposed in FLIF and improve it with an entropy-based pixel prioritization strategy, which enables better progressive compression and syncing. Specifically, we modify the traditional Adam interlacing mode [1] to enable an arbitrary pixel transmission order avoiding spatial dependency issues. After constructing the MANIAC tree, we calculate entropy values for each leaf nodes and use them to determine the priority. In addition, we propose to use pixel masks of individual zoom levels to indicate the positions of the transmitted pixels. We further integrate the mask compression algorithm to reduce the communication cost. Through extensive experiments on over 2000 images, we show our proposed method outperforms the baseline methods.

Recent conversational recommender systems (CRSs) have achieved considerable success on addressing the cold-start problem. While they utilize conversational key-terms to efficiently elicit user preferences, most of them, however, neglect that key-terms can also introduce biases. Systems learning key-term-level user preferences may make a biased item recommendation based on an overrated key-term instead of the item itself. As key-term conversation is a crucial part of CRSs, it is important to properly handle such bias resulting from the item-key-term relationship. While many debiasing methods have been proposed for traditional recommender systems, most of them focus on items or item groups re-ranking or re-weighting strategies such as calibration and propensity score, which are not designed to model the relation between item and key-term user preference. There is also no effective way for traditional debiasing methods to measure potentially useful biases through conversational key-terms to enhance the recommendation performance.

Learning from noisy labels is an important and long-standing problem in machine learning for real applications. One of the main research lines focuses on learning a label corrector to purify potential noisy labels. However, these methods typically rely on strict assumptions and are limited to certain types of label noise. In this paper, we reformulate the label-noise problem from a generative-model perspective, *i.e.*, labels are generated by gradually refining an initial random guess. This new perspective immediately enables existing powerful diffusion models to seamlessly learn the stochastic generative process. Once the generative uncertainty is modeled, we can perform classification inference using maximum likelihood estimation of labels. To mitigate the impact of noisy labels, we propose the **L**abel-**R**etrieval-**A**ugmented (LRA) diffusion model, which leverages neighbor consistency to effectively construct pseudo-clean labels for diffusion training. Our model is flexible and general, allowing easy incorporation of different types of conditional information, *e.g.*, use of pre-trained models, to further boost model performance. Extensive experiments are conducted for evaluation. Our model achieves new state-of-the-art (SOTA) results on all the standard real-world benchmark datasets. Remarkably, by incorporating conditional information from the powerful CLIP model, our method can boost the current SOTA accuracy by 10-20 absolute points in many cases. Code is available: https://anonymous.4open.science/r/LRA-diffusion-5F2F

Discovering causal relations from observational time series without making the stationary assumption is a significant challenge. In practice, this challenge is common in many areas, such as retail sales, transportation systems, and medical science. Here, we consider this problem for a class of non-stationary time series. The structural causal model (SCM) of this type of time series, called the semi-stationary time series, exhibits that a finite number of different causal mechanisms occur sequentially and periodically across time. This model holds considerable practical utility because it can represent periodicity, including common occurrences such as seasonality and diurnal variation. We propose a constraint-based, non-parametric algorithm for discovering causal relations in this setting. The resulting algorithm, PCMCI$_{\Omega}$, can capture the alternating and recurring changes in the causal mechanisms and then identify the underlying causal graph with conditional independence (CI) tests. We show that this algorithm is sound in identifying causal relations on discrete time series. We validate the algorithm with extensive experiments on continuous and discrete simulated data. We also apply our algorithm to a real-world climate dataset.

In real-world online web systems, multiple users usually arrive sequentially into the system. For applications like click fraud and fake reviews, some users can maliciously perform corrupted (disrupted) behaviors to trick the system. Therefore, it is crucial to design efficient online learning algorithms to robustly learn from potentially corrupted user behaviors and accurately identify the corrupted users in an online manner. Existing works propose bandit algorithms robust to adversarial corruption. However, these algorithms are designed for a single user, and cannot leverage the implicit social relations among multiple users for more efficient learning. Moreover, none of them consider how to detect corrupted users online in the multiple-user scenario. In this paper, we present an important online learning problem named LOCUD to learn and utilize unknown user relations from disrupted behaviors to speed up learning, and identify the corrupted users in an online setting. To robustly learn and utilize the unknown relations among potentially corrupted users, we propose a novel bandit algorithm RCLUB-WCU. To detect the corrupted users, we devise a novel online detection algorithm OCCUD based on RCLUB-WCU's inferred user relations. We prove a regret upper bound for RCLUB-WCU, which asymptotically matches the lower bound with respect to $T$ up to logarithmic factors, and matches the state-of-the-art results in degenerate cases. We also give a theoretical guarantee for the detection accuracy of OCCUD. With extensive experiments, our methods achieve superior performance over previous bandit algorithms and high corrupted user detection accuracy.

Soft prompt tuning achieves superior performances across a wide range of few-shot tasks. However, the performances of prompt tuning can be highly sensitive to the initialization of the prompts. We have also empirically observed that conventional prompt tuning methods cannot encode and learn sufficient task-relevant information from prompt tokens. In this work, we develop an information-theoretic framework that formulates soft prompt tuning as maximizing the mutual information between prompts and other model parameters (or encoded representations). This novel view helps us to develop a more efficient, accurate and robust soft prompt tuning method, InfoPrompt. With this framework, we develop two novel mutual information based loss functions, to (i) explore proper prompt initialization for the downstream tasks and learn sufficient task-relevant information from prompt tokens and (ii) encourage the output representation from the pretrained language model to be more aware of the task-relevant information captured in the learnt prompts. Extensive experiments validate that InfoPrompt can significantly accelerate the convergence of the prompt tuning and outperform traditional prompt tuning methods. Finally, we provide a formal theoretical result to show that a gradient descent type algorithm can be used to train our mutual information loss.

Interactive recommendation with natural-language feedback can provide richer user feedback and has demonstrated advantages over traditional recommender systems. However, the classical online paradigm involves iteratively collecting experience via interaction with users, which is expensive and risky. We consider an offline interactive recommendation to exploit arbitrary experience collected by multiple unknown policies. A direct application of policy learning with such fixed experience suffers from the distribution shift. To tackle this issue, we develop a behavior-agnostic off-policy correction framework to make offline interactive recommendation possible. Specifically, we leverage the conservative Q-function to perform off-policy evaluation, which enables learning effective policies from fixed datasets without further interactions. Empirical results on the simulator derived from real-world datasets demonstrate the effectiveness of our proposed offline training framework.

Using natural-language feedback to guide image generation and manipulation can greatly lower the required efforts and skills. This topic has received increased attention in recent years through refinement of Generative Adversarial Networks (GANs); however, most existing works are limited to single-round interaction, which is not reflective of real world interactive image editing workflows. Furthermore, previous works dealing with multi-round scenarios are limited to predefined feedback sequences, which is also impractical. In this paper, we propose a novel framework for Text-based Interactive image generation and manipulation (TiGAN) that responds to users' natural-language feedback. TiGAN utilizes the powerful pre-trained CLIP model to understand users' natural-language feedback and exploits contrastive learning for a better text-to-image mapping. To maintain the image consistency during interactions, TiGAN generates intermediate feature vectors aligned with the feedback and selectively feeds these vectors to our proposed generative model. Empirical results on several datasets show that TiGAN improves both interaction efficiency and image quality while better avoids undesirable image manipulation during interactions.

Rui Wang, Tong Yu, Handong Zhao, Sungchul Kim, Subrata Mitra, Ruiyi Zhang, Ricardo Henao. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022.

The recent advances of conversational recommendations provide a promising way to efficiently elicit users' preferences via conversational interactions. To achieve this, the recommender system conducts conversations with users, asking their preferences for different items or item categories. Most existing conversational recommender systems for cold-start users utilize a multi-armed bandit framework to learn users' preference in an online manner. However, they rely on a pre-defined conversation frequency for asking about item categories instead of individual items, which may incur excessive conversational interactions that hurt user experience. To enable more flexible questioning about key-terms, we formulate a new conversational bandit problem that allows the recommender system to choose either a key-term or an item to recommend at each round and explicitly models the rewards of these actions. This motivates us to handle a new exploration-exploitation (EE) trade-off between key-term asking and item recommendation, which requires us to accurately model the relationship between key-term and item rewards. We conduct a survey and analyze a real-world dataset to find that, unlike assumptions made in prior works, key-term rewards are mainly affected by rewards of representative items. We propose two bandit algorithms, Hier-UCB and Hier-LinUCB, that leverage this observed relationship and the hierarchical structure between key-terms and items to efficiently learn which items to recommend. We theoretically prove that our algorithm can reduce the regret bound's dependency on the total number of items from previous work. We validate our proposed algorithms and regret bound on both synthetic and real-world data.

Cross language-image modality retrieval in E-commerce is a fundamental problem for product search, recommendation, and marketing services. Extensive efforts have been made to conquer the cross-modal retrieval problem in the general domain. When it comes to E-commerce, a com-mon practice is to adopt the pretrained model and finetune on E-commerce data. Despite its simplicity, the performance is sub-optimal due to overlooking the uniqueness of E-commerce multimodal data. A few recent efforts [10], [72] have shown significant improvements over generic methods with customized designs for handling product images. Unfortunately, to the best of our knowledge, no existing method has addressed the unique challenges in the e-commerce language. This work studies the outstanding one, where it has a large collection of special meaning entities, e.g., “Di s s e l (brand)”, “Top (category)”, “relaxed (fit)” in the fashion clothing business. By formulating such out-of-distribution finetuning process in the Causal Inference paradigm, we view the erroneous semantics of these special entities as confounders to cause the retrieval failure. To rectify these semantics for aligning with e-commerce do-main knowledge, we propose an intervention-based entity-aware contrastive learning framework with two modules, i.e., the Confounding Entity Selection Module and Entity-Aware Learning Module. Our method achieves competitive performance on the E-commerce benchmark Fashion-Gen. Particularly, in top-1 accuracy (R@l), we observe 10.3% and 10.5% relative improvements over the closest baseline in image-to-text and text-to-image retrievals, respectively.

One of the major challenges in training text-to-image generation models is the need of a large number of highquality image-text pairs. While image samples are often easily accessible, the associated text descriptions typically require careful human captioning, which is particularly time- and cost-consuming. In this paper, we propose the first work to train text-to-image generation models without any text data. Our method leverages the well-aligned multi-modal semantic space of the powerful pre-trained CLIP model: the requirement of text-conditioning is seamlessly alleviated via generating text features from image features. Extensive experiments are conducted to illustrate the effectiveness of the proposed method. We obtain state-of-the-art results in the standard text-to-image generation tasks. Importantly, the proposed language-free model outperforms most existing models trained with full image-text pairs. Furthermore, our method can be applied in fine-tuning pretrained models, which saves both training time and cost in training text-to-image generation models. Our pre-trained model obtains competitive results in zero-shot text-to-image generation on the MS-COCO dataset, yet with around only 1% of the model size and training data size relative to the recently proposed large DALL-E model.

Large pretrained multilingual language models (ML-LMs) have shown remarkable capabilities of zero-shot cross-lingual transfer, without direct cross-lingual supervision.While these results are promising, follow-up works found that, within the multilingual embedding spaces, there exists strong language identity information which hinders the expression of linguistic factors shared across languages.For semantic tasks like cross-lingual sentence retrieval, it is desired to remove such language identity signals to fully leverage semantic information.In this work, we provide a novel view of projecting away language-specific factors from a multilingual embedding space.Specifically, we discover that there exists a low-rank subspace that primarily encodes information irrelevant to semantics (e.g., syntactic information).To identify this subspace, we present a simple but effective unsupervised method based on singular value decomposition with multiple monolingual corpora as input.Once the subspace is found, we can directly project the original embeddings into the null space to boost language agnosticism without finetuning.We systematically evaluate our method on various tasks including the challenging language-agnostic QA retrieval task.Empirical results show that applying our method consistently leads to improvements over commonly used ML-LMs.

With a focus on both the fitness and cost of subset selection, we study stochastic local search (SLS) heuristics in this paper. In particular, we consider subset selection problems where the cost of fitness function evaluation needs to be accounted for. Here, cost can be fitness evaluation's computation time or energy cost. We propose and study an SLS method, SLS4CFF, tailored to such problems. SLS4CFF ("SLS for costly fitness functions") is an amalgamation of several existing heuristics. We develop a homogeneous Markov chain model that explicitly represents both fitness and cost of subset selection with SLS4CFF. This Markov chain, which can be lumped or compressed for certain fitness and cost functions, enables us to better understand and analyze hyperparameter optimization in a principled manner, via expected hitting times. Studies with synthetic and real-world problems improve the understanding of SLS and demonstrate the importance of cost-awareness.

The growth of the Internet of Things (IoT) sensing systems leads to a large number of multimodal datasets over different deployments. Labeling costs for these datasets, especially fine-grained labels, are often tremendous. On the other hand, different data distributions (domain variance) of these datasets prevent models built with labels of one dataset (source domain) from being directly used in another (target domain). This domain variance may be caused by one or more physical factors change in the deployments, such as buildings and/ or people. Existing model transfer studies mainly focus on adapting the model to the domain variance caused by only one physical factor change. When multiple factors change between the source and target domains, the model transfer often yields low accuracy due to significant domain variance. We present VMA, a model transfer framework for multimodal IoT sensing data that handles multi-factor domain variance. VMA first decouples the multi-factor domain variance between two datasets to multiple single-factor domain variance dataset pairs with other available datasets. Then, VMA leverages sensing modalities robust to each single-factor domain variance for accurate prediction by weighing them more in the fusion. We apply VMA to the fine-grained occupant activity recognition application with a multi-modal sensing system of structural vibration and wearable IMU. We collect real-world datasets to evaluate the proposed framework. VMA achieves a model transfer accuracy up to 76.1% on the target domain with multi-factor domain variance, demonstrating a 1.6x and 1.9x error reduction compared to direct prediction baselines with and without modality-aware learning design.

Tabular pre-training models have received increasing attention due to the wide-ranging applications for tabular data analysis. However, most of the existing solutions are directly built upon the tabular data with a mixture of non-semantic and semantic contents. According to the statistics, only 30% of tabular data in wikitables are semantic entities that are surrounded and isolated by enormous irregular characters such as numbers, strings, symbols, etc. Despite the small portion, such semantic entities are crucial for table understanding. This paper attempts to enhance the existing tabular pre-training model by injecting common-sense knowledge from external sources. Compared with the knowledge injection in the natural language pre-training models, the tabular model naturally requires overcoming the domain gaps between external knowledge and tabular data with significant differences in both structures and contents. To this end, we propose the dual-adapters inserted within the pre-trained tabular model for flexible and efficient knowledge injection. The two parallel adapters are trained by the knowledge graph triplets and semantically augmented tables respectively for infusion and alignment with the tabular data. In addition, a path-wise attention layer is attached below to fuse the cross-domain representation with the weighted contribution. Finally, to verify the effectiveness of our proposed knowledge injection framework, we extensively test it on 5 different application scenarios covering both zero-shot and finetuning-based tabular understanding tasks over the cell, column, and tables levels.

Existing interactive learning systems usually train models on simulators as surrogates for real users. Due to the limited amount of user data, trained simulators may lead to biased results as it fails to well represent real users. One solution is to model users as agents, and then simultaneously train the interactive system and user agents by multi-agent reinforcement learning (MARL) frameworks. However, developing efficient MARL frameworks for modern interactive multimodal systems is still challenging. First, given the existence of multimodal data, how to develop accurate multimodal fusion within and between agents in each interaction is challenging and unclear. Second, interactions between users and systems are complex and it is challenging to track and synchronize the interactions over time. The above multimodal fusion between agents and synchronization over time becomes even more challenging, when the amount of user data is limited. To jointly address these challenges and achieve more sample-efficient learning, we propose a novel spatial-temporal aligned sta multi-agent reinforcement learning framework to better align the multimodal data within and between agents over time. Based on our framework, we develop sample-efficient visual dialog systems. Through extensive experiments and analysis, we validate the effectiveness of our spatial-temporal aligned sta multi-agent reinforcement learning framework in visual dialog systems.

Bundle recommender systems recommend sets of items (e.g., pants, shirt, and shoes) to users, but they often suffer from two issues: significant interaction sparsity and a large output space. In this work, we extend multi-round conversational recommendation (MCR) to alleviate these issues. MCR—which uses a conversational paradigm to elicit user interests by asking user preferences on tags (e.g., categories or attributes) and handling user feedback across multiple rounds—is an emerging recommendation setting to acquire user feedback and narrow down the output space, but has not been explored in the context of bundle recommendation.

Occupant information inference with IoT sensor data enables many smart applications, such as patients'/older adults' in-home monitoring. The difficulty of collecting labeled real-world IoT sensor data often leads to reliability and scalability issues for those systems. Extensive prior works (e.g., domain adaptation) focus on the domain shift issues, i.e., the inconsistent data feature and label relationship, and dataset bias is often neglected. Dataset bias is commonly caused by limited and varied accessibility to labeled data for each class, and it is inevitable for real-world datasets. The model trained with a biased dataset fits into the bias, hence cannot further generalize to the testing data for accurate inference.

Interactive recommender systems (IRS) have received wide attention in recent years. To capture users' dynamic preferences and maximize their long-term engagement, IRS are usually formulated as reinforcement learning (RL) problems. Despite the promise to solve complex decision-making problems, RL-based methods generally require a large amount of online interaction, restricting their applications due to economic considerations. One possible direction to alleviate this issue is cross-domain recommendation that aims to leverage abundant logged interaction data from a source domain (e.g., adventure genre in movie recommendation) to improve the recommendation quality in the target domain (e.g., crime genre). Nevertheless, prior studies mostly focus on adapting the static representations of users/items. Few have explored how the temporally dynamic user-item interaction patterns transform across domains.

Contextual multi-armed bandit (MAB) is an important sequential decision-making problem in recommendation systems. A line of works, called the clustering of bandits (CLUB), utilize the collaborative effect over users and dramatically improve the recommendation quality. Owing to the increasing application scale and public concerns about privacy, there is a growing demand to keep user data decentralized and push bandit learning to the local server side. Existing CLUB algorithms, however, are designed under the centralized setting where data are available at a central server. We focus on studying the federated online clustering of bandit (FCLUB) problem, which aims to minimize the total regret while satisfying privacy and communication considerations. We design a new phase-based scheme for cluster detection and a novel asynchronous communication protocol for cooperative bandit learning for this problem. To protect users’ privacy, previous differential privacy (DP) definitions are not very suitable, and we propose a new DP notion that acts on the user cluster level. We provide rigorous proofs to show that our algorithm simultaneously achieves (clustered) DP, sublinear communication complexity and sublinear regret. Finally, experimental evaluations show our superior performance compared with benchmark algorithms.

Multilingual natural language understanding, which aims to comprehend multilingual documents, is an important task. Existing efforts have been focusing on the analysis of centrally stored text data, but in real practice, multilingual data is usually distributed. Federated learning is a promising paradigm to solve this problem, which trains local models with decentralized data on local clients and aggregates local models on the central server to achieve a good global model. However, existing federated learning methods assume that data are independent and identically distributed (IID), and cannot handle multilingual data, that are usually non-IID with severely skewed distributions: First, multilingual data is stored on local client devices such that there are only monolingual or bilingual data stored on each client. This makes it difficult for local models to know the information of documents in other languages. Second, the distribution over different languages could be skewed. High resource language data is much more abundant than low resource language data. The model trained on such skewed data may focus more on high resource languages but fail to consider the key information of low resource languages. To solve the aforementioned challenges of multilingual federated NLU, we propose a plug-and-play knowledge composition (KC) module, called FedKC, which exchanges knowledge among clients without sharing raw data. Specifically, we propose an effective way to calculate a consistency loss defined based on the shared knowledge across clients, which enables models trained on different clients achieve similar predictions on similar data. Leveraging this consistency loss, joint training is thus conducted on distributed data respecting the privacy constraints. We also analyze the potential risk of FedKC and provide theoretical bound to show that it is difficult to recover data from the corrupted data. We conduct extensive experiments on three public multilingual datasets for three typical NLU tasks, including paraphrase identification, question answering matching, and news classification. The experiment results show that the proposed FedKC can outperform state-of-the-art baselines on the three datasets significantly.

Conversational recommender systems (CRSs) have been proposed recently to mitigate the cold-start problem suffered by the traditional recommender systems. By introducing conversational key-terms, existing conversational recommenders can effectively reduce the need for extensive exploration and elicit the user preferences faster and more accurately. However, existing conversational recommenders leveraging key-terms heavily rely on the availability and quality of the key-terms, and their performances might degrade significantly when the key-terms are incomplete or not well labeled, which usually happens when there are new items being consistently incorporated into the systems and involving lots of human efforts to acquire well-labeled key-terms is costly. Besides, existing CRS methods leverage the feedback to different conversational key-terms separately, without considering the underlying relations between the key-terms. In this case, the learning of the conversational recommenders is sample inefficient, especially when there is a large number of candidate conversational key-terms.

Conversational recommender systems elicit user preference via interactive conversational interactions. By introducing conversational key-terms, existing conversational recommenders can effectively reduce the need for extensive exploration in a traditional interactive recommender. However, there are still limitations of existing conversational recommender approaches eliciting user preference via key-terms. First, the key-term data of the items needs to be carefully labeled, which requires a lot of human efforts. Second, the number of the human labeled key-terms is limited and the granularity of the key-terms is fixed, while the elicited user preference is usually from coarse-grained to fine-grained during the conversations. In this paper, we propose a clustering of conversational bandits algorithm. To avoid the human labeling efforts and automatically learn the key-terms with the proper granularity, we online cluster the items and generate meaningful key-terms for the items during the conversational interactions. Our algorithm is general and can also be used in the user clustering when the feedback from multiple users is available, which further leads to more accurate learning and generations of conversational key-terms. We analyze the regret bound of our learning algorithm. In the empirical evaluations, without using any human labeled key-terms, our algorithm effectively generates meaningful coarse-to-fine grained key-terms and performs as well as or better than the state-of-the-art baseline.

In vision-language retrieval systems, users provide natural language feedback to find target images. Vision-language explanations in the systems can better guide users to provide feedback and thus improve the retrieval. However, developing explainable vision-language retrieval systems can be challenging, due to limited labeled multimodal data. In the retrieval of complex scenes, the issue of limited labeled data can be more severe. With multiple objects in the complex scenes, each user query may not exhaustively describe all objects in the desired image and thus more labeled queries are needed. The issue of limited labeled data can cause data selection biases, and result in spurious correlations learned by the models. When learning spurious correlations, existing explainable models may not be able to accurately extract regions from images and keywords from user queries.

With the recent advances of conversational recommendations, the recommender system is able to actively and dynamically elicit user preference via conversational interactions. To achieve this, the system periodically queries users' preference on attributes and collects their feedback. However, most existing conversational recommender systems only enable the user to provide absolute feedback to the attributes. In practice, the absolute feedback is usually limited, as the users tend to provide biased feedback when expressing the preference. Instead, the user is often more inclined to express comparative preferences, since user preferences are inherently relative. To enable users to provide comparative preferences during conversational interactions, we propose a novel comparison-based conversational recommender system. The relative feedback, though more practical, is not easy to be incorporated since its feedback scale is always mismatched with users' absolute preferences. With effectively collecting and understanding the relative feedback from an interactive manner, we further propose a new bandit algorithm, which we call RelativeConUCB. The experiments on both synthetic and real-world datasets validate the advantage of our proposed method, compared to the existing bandit algorithms in the conversational recommender systems.

With the recent advances of multimodal interactive recommendations, the users are able to express their preference by natural language feedback to the item images, to find the desired items. However, the existing systems either retrieve only one item or require the user to specify (e.g., by click or touch) the commented items from a list of recommendations in each user interaction. As a result, the users are not hands-free and the recommendations may be impractical. We propose a hands-free visual dialog recommender system to interactively recommend a list of items. At each time, the system shows a list of items with visual appearance. The user can comment on the list in natural language, to describe the desired features they further want. With these multimodal data, the system chooses another list of items to recommend. To understand the user preference from these multimodal data, we develop neural network models which identify the described items among the list and further predict the desired attributes. To achieve efficient interactive recommendations, we leverage the inferred user preference and further develop a novel bandit algorithm. Specifically, to avoid the system exploring more than needed, the desired attributes are utilized to reduce the exploration space. More importantly, to achieve sample efficient learning in this hands-free setting, we derive additional samples from the user's relative preference expressed in natural language and design a pairwise logistic loss in bandit learning. Our bandit model is jointly updated by the pairwise logistic loss on the additional samples derived from natural language feedback and the traditional logistic loss. The empirical results show that the probability of finding the desired items by our system is about 3 times as high as that by the traditional interactive recommenders, after a few user interactions.

We propose a novel framework for structured bandits, which we call an influence diagram bandit. Our framework uses a graphical model to capture complex statistical dependencies between actions, latent variables, and observations; and thus unifies and extends many existing models, such as combinatorial semi-bandits, cascading bandits, and low-rank bandits. We develop novel online learning algorithms that learn to act efficiently in our models. The key idea is to track a structured posterior distribution of model parameters, either exactly or approximately. To act, we sample model parameters from their posterior and then use the structure of the influence diagram to find the most optimistic action under the sampled parameters. We empirically evaluate our algorithms in three structured bandit problems, and show that they perform as well as or better than problem-specific state-of-the-art baselines.

Thompson sampling (TS) is a class of algorithms for sequential decision-making, which requires maintaining a posterior distribution over a reward model. However, calculating exact posterior distributions is intractable for all but the simplest models. Consequently, how to computationally-efficiently approximate a posterior distribution is a crucial problem for scalable TS with complex models, such as neural networks. In this paper, we use distribution optimization techniques to approximate the posterior distribution, solved via Wasserstein gradient flows. Based on the framework, a principled particle-optimization algorithm is developed for TS to approximate the posterior efficiently. Our approach is scalable and does not make explicit distribution assumptions on posterior approximations. Extensive experiments on both synthetic data and large-scale real data demonstrate the superior performance of the proposed methods.

Traditional recommender systems rely on user feedback such as ratings or clicks to the items, to analyze the user interest and provide personalized recommendations. However, rating or click feedback are limited in that they do not exactly tell why users like or dislike an item. If a user does not like the recommendations and can not effectively express the reasons via rating and clicking, the feedback from the user may be very sparse. These limitations lead to inefficient model learning of the recommender system. To address these limitations, more effective user feedback to the recommendations should be designed, so that the system can effectively understand a user's preference and improve the recommendations over time. In this paper, we propose a novel dialog-based recommender system to interactively recommend a list of items with visual appearance. At each time, the user receives a list of recommended items with visual appearance. The user can point to some items and describe their feedback, such as the desired features in the items they want in natural language. With this natural language based feedback, the recommender system updates and provides another list of items. To model the user behaviors of viewing, commenting and clicking on a list of items, we propose a visual dialog augmented cascade model. To efficiently understand the user preference and learn the model, exploration should be encouraged to provide more diverse recommendations to quickly collect user feedback on more attributes of the items. We propose a variant of the cascading bandits, where the neural representations of the item images and user feedback in natural language are utilized. In a task of recommending a list of footwear, we show that our visual dialog augmented interactive recommender needs around 41.03% rounds of recommendations, compared to the traditional interactive recommender only relying on the user click behavior.

Interactive recommenders have demonstrated the advantage over traditional recommenders with dynamic change of items. However, the traditional user feedback in the format of clicks or ratings, provides limited user preference information and limited history tracking capabilities. As a result, it takes a user many interactions to find a desired item. Data of other modalities, such as item visual appearance and user comments in natural language, may enable richer user feedback. However, there are several critical challenges to be addressed when utilizing these multimodal data: multimodal matching, user preference tracking, and adaptation to dynamic unseen items. Without properly handling these challenges, the recommendations can easily violate the users' preference from their past natural language feedback. In this paper, we introduce a novel approach, called vision-language recommendation, that enables users to provide natural language feedback on visual products to have more natural and effective interactions. To model more explicit and accurate multimodal matching, we propose a novel visual attribute augmented reinforcement learning approach that enhances the grounding of natural language to visual items. Furthermore, to effectively track the users' preference and overcome the performance deficiency on dynamic unseen items after deployment, we propose a novel history multimodal matching reward to continuously adapt the model on-the-fly. Empirical results show that, our system augmented by visual attribute and history multimodal matching can significantly increase the success rate, reduce the number of recommendations that violate the user's previous feedback, and need less number of user interactions to find the desired items.

Text-based interactive recommendation provides richer user preferences and has demonstrated advantages over traditional interactive recommender systems. However, recommendations can easily violate preferences of users from their past natural-language feedback, since the recommender needs to explore new items for further improvement. To alleviate this issue, we propose a novel constraint-augmented reinforcement learning (RL) framework to efficiently incorporate user preferences over time. Specifically, we leverage a discriminator to detect recommendations violating user historical preference, which is incorporated into the standard RL objective of maximizing expected cumulative future rewards. Our proposed framework is general and is further extended to the task of constrained text generation. Empirical results show that the proposed method yields consistent improvement relative to standard RL methods.

Dialog response selection is an important step towards natural response generation in conversational agents. Existing work on neural conversational models mainly focuses on offline supervised learning using a large set of context-response pairs. In this paper, we focus on online learning of response selection in retrieval-based dialog systems. We propose a contextual multi-armed bandit model with a nonlinear reward function that uses distributed representation of text for online response selection. A bidirectional LSTM is used to produce the distributed representations of dialog context and responses, which serve as the input to a contextual bandit. In learning the bandit, we propose a customized Thompson sampling method that is applied to a polynomial feature space in approximating the reward. Experimental results on the Ubuntu Dialogue Corpus demonstrate significant performance gains of the proposed method over conventional linear contextual bandits. Moreover, we report encouraging response selection performance of the proposed neural bandit model using the Recall@k metric for a small set of online training samples.

Deep neural networks, including recurrent networks, have been successfully applied to human activity recognition. Unfortunately, the final representation learned by recurrent networks might encode some noise (irrelevant signal components, unimportant sensor modalities, etc.). Besides, it is difficult to interpret the recurrent networks to gain insight into the models' behavior. To address these issues, we propose two attention models for human activity recognition: temporal attention and sensor attention. These two mechanisms adaptively focus on important signals and sensor modalities. To further improve the understandability and mean Fl score, we add continuity constraints, considering that continuous sensor signals are more robust than discrete ones. We evaluate the approaches on three datasets and obtain state-of-the-art results. Furthermore, qualitative analysis shows that the attention learned by the models agree well with human intuition.

Labeled data used for training activity recognition classifiers are usually limited in terms of size and diversity. Thus, the learned model may not generalize well when used in real-world use cases. Semi-supervised learning augments labeled examples with unlabeled examples, often resulting in improved performance. However, the semi-supervised methods studied in the activity recognition literatures assume that feature engineering is already done. In this paper, we lift this assumption and present two semi-supervised methods based on convolutional neural networks (CNNs) to learn discriminative hidden features. Our semi-supervised CNNs learn from both labeled and unlabeled data while also performing feature learning on raw sensor data. In experiments on three real world datasets, we show that our CNNs outperform supervised methods and traditional semi-supervised learning methods by up to 18% in mean F1-score (Fm).

Deep neural networks (DNNs) are popular in diverse fields such as computer vision and natural language processing. DNN inference tasks are emerging as a service provided by cloud computing environments. However, cloud-hosted DNN inference faces new challenges in workload scheduling for the best Quality of Service (QoS), due to dependence on batch size, model complexity and resource allocation. This paper represents the QoS metric as a utility function of response delay and inference accuracy. We first propose a simple and effective heuristic approach that keeps low response delay and satisfies the requirement on processing throughput. Then we describe an advanced deep reinforcement learning (RL) approach that learns to schedule from experience. The RL scheduler is trained to maximize QoS, using a set of system statuses as the input to the RL policy model. Our approach performs scheduling actions only when there are free GPUs, thus reduces scheduling overhead over common RL schedulers that run at every continuous time step. We evaluate the schedulers on a simulation platform and demonstrate the advantages of RL over heuristics.

This work considers that multiple mobile clients offload various continuous sensing applications with end-to-end delay constraints, to a cluster of machines as the server. Contention for shared computing resources on a server can result in delay degradation and application malfunction. We present ATOMS (Accurate Timing prediction and Offloading for Mobile Systems), a framework to mitigate multi-tenant resource contention and to improve delay using a two-phase Plan-Schedule approach. The planning phase includes methods to predict future workloads from all clients, to estimate contention, and to devise offloading schedule to reduce contention. The scheduling phase dispatches arriving offloaded workload to the server machine that minimizes contention, based on the running workloads on each machine.

Mobile devices, such as smartphones and tablets, have become prevalent given their ample functionality brought by a variety of applications. Unfortunately, these devices face security and privacy threats due to unauthorized access. Ordinary protection mechanisms such as passcode and fingerprint verification are widely employed to mitigate the threats. To achieve strong security without sacrificing usability, extensive research efforts have been devoted to continuous authentication through passive sensing and behavior modeling. Nowadays, more and more users own multiple devices. This trend presents opportunities for further optimization of authentication across devices. In this paper, we conduct an empirical study on how a behavioral model created on one device can be transferred to other devices to bootstrap continuous authentication. To pursue this goal, we collect 160 sets of usage data on multiple mobile devices and perform a proof-of-concept experiment. The results demonstrate that we can leverage the similarity between user behaviors on different devices to enable cross-device authentication and anomaly detection.

Recommender systems play a key role in personalizing service experiences by recommending relevant items to users. One popular technique for producing such personalization at scale is collaborative filtering via Matrix Factorization (MF). The essence of MF is to train a model by factorizing a sparse rating matrix consisting of users' ratings of item. Unfortunately, existing MF methods require model Learning from Scratch when new data (for users, items, or user ratings) arrive. Learning large models from scratch incurs significant computation cost and typically also results in stale recommendations. With increasing amounts of data and a need for real-time recommendations, incremental learning is desirable. In this paper, we develop a novel but simple method for incremental learning of MF models, called One-sided Least Squares, and demonstrate its parallel implementation via Apache Spark. We also describe how to integrate it with batch learning via Alternating Least Squares (ALS). Unlike previous incremental learning methods, we study our method's approximation of the results of ALS, while significantly reducing compute and storage costs. Our theoretical analysis and experimental results on three real-world datasets suggest that One-sided Least Squares achieves prediction accuracy close to Learning from Scratch with ALS at substantially faster learning speeds. This fast and accurate method for incremental learning enables improved Web-scale recommender systems.

Sensors on mobile phones and wearables, and in general sensors on IoT (Internet of Things), bring forth a couple of new challenges to big data research. First, the power consumption for analyzing sensor data must be low, since most wearables and portable devices are power-strapped. Second, the velocity of analyzing big data on these devices must be high, otherwise the limited local storage may overflow. This paper presents our hardware-software co-design of a classifier for wearables to detect a person's transportation mode (i.e., still, walking, running, biking, and on a vehicle). We particularly focus on addressing the big-data small-footprint requirement by designing a classifier that is low in both computational complexity and memory requirement. Together with a sensor-hub configuration, we are able to drastically reduce power consumption by 99%, while maintaining competitive mode-detection accuracy. The data used in the paper is made publicly available for conducting research.