Systematic reviews (SRs) are vital for evidencebased practice in high stakes disciplines, such as healthcare, but are often impeded by laborintensive and lengthy processes that can span months.Due to the high demand for domain expertise, existing automatic summarization methods fail to accurately identify relevant studies and generate high-quality summaries.To that end, we introduce InsightAgent, a human-centered interactive AI agent powered by large language models that revolutionizes the systematic review workflow.In-sightAgent partitions a large literature corpus based on semantics and employs a multi-agent design for more focused processing of literature, leading to significant improvement in the quality of generated SRs.InsightAgent also provides intuitive visualizations of the corpus and agent trajectories, allowing users to effortlessly monitor the actions of the agent and provide real-time feedback based on their expertise.Our user studies with 9 medical professionals demonstrate that the visualization and interaction mechanisms can effectively improve the quality of synthesized SRs by 27.2%, reaching 79.7% of human-written quality.At the same time, user satisfaction is improved by 34.4%.With InsightAgent, it only takes a clinician about 1.5 hours, rather than months, to complete a high-quality systematic review.InsightAgent demonstrates great potential in facilitating more timely and informed decisionmaking in high stake application scenarios 1 .

Xiang Yue, Tianyu Zheng, Yuansheng Ni, Yubo Wang, Kai Zhang, Shengbang Tong, Yuxuan Sun, Botao Yu, Ge Zhang, Huan Sun, Yu Su, Wenhu Chen, Graham Neubig. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

We present a simple approach to make pre-trained Vision Transformers (ViTs) interpretable for fine-grained analysis, aiming to identify and localize the traits that distinguish visually similar categories, such as bird species. Pretrained ViTs, such as DINO, have demonstrated remarkable capabilities in extracting localized, discriminative features. However, saliency maps like Grad-CAM often fail to identify these traits, producing blurred, coarse heatmaps that highlight entire objects instead. We propose a novel approach, Prompt Class Attention Map (Prompt-CAM), to address this limitation. Prompt-CAM learns class-specific prompts for a pre-trained ViT and uses the corresponding outputs for classification. To correctly classify an image, the true-class prompt must attend to unique image patches not present in other classes’ images (i.e., traits). As a result, the true class’s multi-head attention maps reveal traits and their locations. Implementation-wise, Prompt-CAM is almost a "free lunch," requiring only a modification to the prediction head of Visual Prompt Tuning (VPT). This makes Prompt-CAM easy to train and apply, in stark contrast to other interpretable methods that require designing specific models and training processes. Extensive empirical studies on a dozen datasets from various domains (e.g., birds, fishes, insects, fungi, flowers, food, and cars) validate the superior interpretation capability of Prompt-CAM. The source code and demo are available at https://github.com/Imageomics/Prompt_CAM.

Spatial understanding is a crucial capability that enables robots to perceive their surroundings, reason about their environment, and interact with it meaningfully. In modern robotics, these capabilities are increasingly provided by vision-language models. However, these models face significant challenges in spatial reasoning tasks, as their training data are based on general-purpose image datasets that often lack sophisticated spatial understanding. For example, datasets frequently do not capture reference frame comprehension, yet effective spatial reasoning requires understanding whether to reason from ego-, world-, or object-centric perspectives. To address this issue, we introduce RoboSpatial, a large-scale dataset for spatial understanding in robotics. It consists of real indoor and tabletop scenes, captured as 3D scans and egocentric images, and annotated with rich spatial information relevant to robotics. The dataset includes 1M images, 5k 3D scans, and 3M annotated spatial relationships, and the pairing of 2D egocentric images with 3D scans makes it both 2D- and 3D- ready. Our experiments show that models trained with RoboSpatial outperform baselines on downstream tasks such as spatial affordance prediction, spatial relationship prediction, and robotics manipulation.

Class activation map (CAM) has been widely used to highlight image regions that contribute to class predictions. Despite its simplicity and computational efficiency, CAM often struggles to identify discriminative regions that distinguish visually similar fine-grained classes. Prior efforts address this limitation by introducing more sophisticated explanation processes, but at the cost of extra complexity. In this paper, we propose Finer-CAM, a method that retains CAM’s efficiency while achieving precise localization of discriminative regions. Our key insight is that the deficiency of CAM lies not in "how" it explains, but in "what" it explains. Specifically, previous methods attempt to identify all cues contributing to the target class’s logit value, which inadvertently also activates regions predictive of visually similar classes. By explicitly comparing the target class with similar classes and spotting their differences, Finer-CAM suppresses features shared with other classes and emphasizes the unique, discriminative details of the target class. Finer-CAM is easy to implement, compatible with various CAM methods, and can be extended to multi-modal models for accurate localization of specific concepts. Additionally, Finer-CAM allows adjustable comparison strength, enabling users to selectively highlight coarse object contours or fine discriminative details. Quantitatively, we show that masking out the top 5% of activated pixels by Finer-CAM results in a larger relative confidence drop compared to baselines. The source code and demo are available at https://github.com/Imageomics/Finer-CAM.

Kunlun Zhu, Jiaxun Zhang, Ziheng Qi, Nuoxing Shang, Zijia Liu, Peixuan Han, Yue Su, Haofei Yu, Jiaxuan You. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

The advancements of language language models (LLMs) have piqued growing interest in developing LLM-based language agents to automate scientific discovery end-to-end, which has sparked both excitement and skepticism about the true capabilities of such agents. In this work, we argue that for an agent to fully automate scientific discovery, it must be able to complete all essential tasks in the workflow. Thus, we call for rigorous assessment of agents on individual tasks in a scientific workflow before making bold claims on end-to-end automation. To this end, we present ScienceAgentBench, a new benchmark for evaluating language agents for data-driven scientific discovery. To ensure the scientific authenticity and real-world relevance of our benchmark, we extract 102 tasks from 44 peer-reviewed publications in four disciplines and engage nine subject matter experts to validate them. We unify the target output for every task to a self-contained Python program file and employ an array of evaluation metrics to examine the generated programs, execution results, and costs. Each task goes through multiple rounds of manual validation by annotators and subject matter experts to ensure its annotation quality and scientific plausibility. We also propose two effective strategies to mitigate data contamination concerns. Using our benchmark, we evaluate five open-weight and proprietary LLMs, each with three frameworks: direct prompting, OpenHands, and self-debug. Given three attempts for each task, the best-performing agent can only solve 32.4% of the tasks independently and 34.3% with expert-provided knowledge. These results underscore the limited capacities of current language agents in generating code for data-driven discovery, let alone end-to-end automation for scientific research.

Information retrieval (IR) systems have played a vital role in modern digital life and have cemented their continued usefulness in this new era of generative AI via retrieval-augmented generation. With strong language processing capabilities and remarkable versatility, large language models (LLMs) have become popular choices for zero-shot re-ranking in IR systems. So far, LLM-based re-ranking methods rely on strong generative capabilities, which restricts their use to either specialized or powerful proprietary models. Given these restrictions, we ask: is autoregressive generation necessary and optimal for LLMs to perform re-ranking? We hypothesize that there are abundant signals relevant to re-ranking within LLMs that might not be used to their full potential via generation. To more directly leverage such signals, we propose in-context re-ranking (ICR), a novel method that leverages the change in attention pattern caused by the search query for accurate and efficient re-ranking. We assume that more relevant documents should receive more attention weights when an LLM is processing the query tokens, and leverage such signals for re-ranking. To mitigate the intrinsic biases in LLMs, we propose a calibration method using a content-free query. Due to the absence of generation, ICR only requires two ($O(1)$) forward passes to re-rank $N$ documents, making it substantially more efficient than generative re-ranking methods that require at least $O(N)$ forward passes. Our novel design also enables ICR to be applied to any LLM without specialized training while guaranteeing a well-formed ranking. Extensive experiments with two popular open-weight LLMs on standard single-hop and multi-hop information retrieval benchmarks show that ICR outperforms RankGPT while cutting the latency by more than 60% in practice. Through detailed analyses, we show that ICR's performance is specially strong on tasks that require more complex re-ranking signals, such as handling contextualization and contradiction between the query and passages, as well as information integration across multiple passages. Our findings call for further exploration on novel ways of utilizing open-weight LLMs beyond text generation.

Multimodal large language models (MLLMs) are transforming the capabilities of graphical user interface (GUI) agents, facilitating their transition from controlled simulations to complex, real-world applications across various platforms. However, the effectiveness of these agents hinges on the robustness of their grounding capability. Current GUI agents predominantly utilize text-based representations such as HTML or accessibility trees, which, despite their utility, often introduce noise, incompleteness, and increased computational overhead. In this paper, we advocate a human-like embodiment for GUI agents that perceive the environment entirely visually and directly perform pixel-level operations on the GUI. The key is visual grounding models that can accurately map diverse referring expressions of GUI elements to their coordinates on the GUI across different platforms. We show that a simple recipe, which includes web-based synthetic data and slight adaptation of the LLaVA architecture, is surprisingly effective for training such visual grounding models. We collect the largest dataset for GUI visual grounding so far, containing 10M GUI elements and their referring expressions over 1.3M screenshots, and use it to train UGround, a strong universal visual grounding model for GUI agents. Empirical results on six benchmarks spanning three categories (grounding, offline agent, and online agent) show that 1) UGround substantially outperforms existing visual grounding models for GUI agents, by up to 20\% absolute, and 2) agents with UGround outperform state-of-the-art agents, despite the fact that existing agents use additional text-based input while ours only uses visual perception. These results provide strong support for the feasibility and promises of GUI agents that navigate the digital world as humans do.

Large Multimodal Models (LMMs) have ushered in a new era in artificial intelligence, merging capabilities in both language and vision to form highly capable \textbf{Visual Foundation Agents} that are postulated to excel across a myriad of tasks. However, existing benchmarks fail to sufficiently challenge or showcase the full potential of LMMs as visual foundation agents in complex, real-world environments. To address this gap, we introduce VisualAgentBench (VAB), a comprehensive and unified benchmark specifically designed to train and evaluate LMMs as visual foundation agents across diverse scenarios in one standard setting, including Embodied, Graphical User Interface, and Visual Design, with tasks formulated to probe the depth of LMMs' understanding and interaction capabilities. Through rigorous testing across 9 proprietary LMM APIs and 9 open models (18 in total), we demonstrate the considerable yet still developing visual agent capabilities of these models. Additionally, VAB explores the synthesizing of visual agent trajectory data through hybrid methods including Program-based Solvers, LMM Agent Bootstrapping, and Human Demonstrations, offering insights into obstacles, solutions, and trade-offs one may meet in developing open LMM agents. Our work not only aims to benchmark existing models but also provides an instrumental playground for future development into visual foundation agents. Code, train, and test data are available at \url{https://github.com/THUDM/VisualAgentBench}.

Our ability to continuously acquire, organize, and leverage knowledge is a key feature of human intelligence that AI systems must approximate to unlock their full potential. Given the challenges in continual learning with large language models (LLMs), retrieval-augmented generation (RAG) has become the dominant way to introduce new information. However, its reliance on vector retrieval hinders its ability to mimic the dynamic and interconnected nature of human long-term memory. Recent RAG approaches augment vector embeddings with various structures like knowledge graphs to address some of these gaps, namely sense-making and associativity. However, their performance on more basic factual memory tasks drops considerably below standard RAG. We address this unintended deterioration and propose HippoRAG 2, a framework that outperforms standard RAG comprehensively on factual, sense-making, and associative memory tasks. HippoRAG 2 builds upon the Personalized PageRank algorithm used in HippoRAG and enhances it with deeper passage integration and more effective online use of an LLM. This combination pushes this RAG system closer to the effectiveness of human long-term memory, achieving a 7% improvement in associative memory tasks over the state-of-the-art embedding model while also exhibiting superior factual knowledge and sense-making memory capabilities. This work paves the way for non-parametric continual learning for LLMs. Code and data are available at https://github.com/OSU-NLP-Group/HippoRAG.

Nishant Subramani, Jason Eisner, Justin Svegliato, Benjamin Van Durme, Yu Su, Sam Thomson. 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.

Agentic search such as Deep Research systems-where agents autonomously browse the web, synthesize information, and return comprehensive citation-backed answers-represents a major shift in how users interact with web-scale information. While promising greater efficiency and cognitive offloading, the growing complexity and open-endedness of agentic search have outpaced existing evaluation benchmarks and methodologies, which largely assume short search horizons and static answers. In this paper, we introduce Mind2Web 2, a benchmark of 130 realistic, high-quality, and long-horizon tasks that require real-time web browsing and extensive information synthesis, constructed with over 1000 hours of human labor. To address the challenge of evaluating time-varying and complex answers, we propose a novel Agent-as-a-Judge framework. Our method constructs task-specific judge agents based on a tree-structured rubric design to automatically assess both answer correctness and source attribution. We conduct a comprehensive evaluation of ten frontier agentic search systems and human performance, along with a detailed error analysis to draw insights for future development. The best-performing system, OpenAI Deep Research, can already achieve 50-70% of human performance while spending half the time, highlighting its great potential. Altogether, Mind2Web 2 provides a rigorous foundation for developing and benchmarking the next generation of agentic search systems.

Foundation models trained at scale exhibit remarkable emergent behaviors, learning new capabilities beyond their initial training objectives. We find such emergent behaviors in biological vision models via large-scale contrastive vision-language training. To achieve this, we first curate TreeOfLife-200M, comprising 214 million images of living organisms, the largest and most diverse biological organism image dataset to date. We then train BioCLIP 2 on TreeOfLife-200M to distinguish different species. Despite the narrow training objective, BioCLIP 2 yields extraordinary accuracy when applied to various biological visual tasks such as habitat classification and trait prediction. We identify emergent properties in the learned embedding space of BioCLIP 2. At the inter-species level, the embedding distribution of different species aligns closely with functional and ecological meanings (e.g., beak sizes and habitats). At the intra-species level, instead of being diminished, the intra-species variations (e.g., life stages and sexes) are preserved and better separated in subspaces orthogonal to inter-species distinctions. We provide formal proof and analyses to explain why hierarchical supervision and contrastive objectives encourage these emergent properties. Crucially, our results reveal that these properties become increasingly significant with larger-scale training data, leading to a biologically meaningful embedding space.

While large reasoning models demonstrate strong performance on complex tasks, they lack the ability to adjust reasoning token usage based on task difficulty. This often leads to the "overthinking" problem—excessive and unnecessary reasoning—which, although potentially mitigated by human intervention to control the token budget, still fundamentally contradicts the goal of achieving fully autonomous AI. In this work, we propose Adaptive Reasoning Model (ARM), a reasoning model capable of adaptively selecting appropriate reasoning formats based on the task at hand. These formats include three efficient ones—Direct Answer, Short CoT, and Code—as well as a more elaborate format, Long CoT. To train ARM, we introduce Ada-GRPO, an adaptation of Group Relative Policy Optimization (GRPO), which addresses the format collapse issue in traditional GRPO. Ada-GRPO enables ARM to achieve high token efficiency, reducing tokens by an average of $\sim$30%, and up to $\sim$70%, while maintaining performance comparable to the model that relies solely on Long CoT. Furthermore, not only does it improve inference efficiency through reduced token generation, but it also brings a $\sim$2$\times$ speedup in training. In addition to the default Adaptive Mode, ARM supports two additional reasoning modes: 1) Instruction-Guided Mode, which allows users to explicitly specify the reasoning format via special tokens—ideal when the appropriate format is known for a batch of tasks. 2) Consensus-Guided Mode, which aggregates the outputs of the three efficient formats and resorts to Long CoT in case of disagreement, prioritizing performance with higher token usage. All the resources will be released.

In this paper, we examine how large language models (LLMs) solve multi-step problems under a language agent framework with three components: a generator, a discriminator, and a planning method.We investigate the practical utility of two advanced planning methods, iterative correction and tree search.We present a comprehensive analysis of how discrimination accuracy affects the overall performance of agents when using these two methods or a simpler method, re-ranking.Experiments on two tasks, text-to-SQL parsing and mathematical reasoning, show that: (1) advanced planning methods demand discriminators with at least 90% accuracy to achieve significant improvements over re-ranking; (2) current LLMs' discrimination abilities have not met the needs of advanced planning methods to achieve such improvements; (3) with LLM-based discriminators, advanced planning methods may not adequately balance accuracy and efficiency.For example, compared to the other two methods, tree search is at least 10-20 times slower but leads to negligible performance gains, which hinders its real-world applications.1

Tools are essential for large language models (LLMs) to acquire up-to-date information and take consequential actions in external environments.Existing work on tool-augmented LLMs primarily focuses on the broad coverage of tools and the flexibility of adding new tools.However, a critical aspect that has surprisingly been understudied is simply how accurately an LLM uses tools for which it has been trained.We find that existing LLMs, including GPT-4 and open-source LLMs specifically fine-tuned for tool use, only reach a correctness rate in the range of 30% to 60%, far from reliable use in practice.We propose a biologically inspired method for tool-augmented LLMs, simulated trial and error (STE), that orchestrates three key mechanisms for successful tool use behaviors in the biological system: trial and error, imagination, and memory.Specifically, STE leverages an LLM's 'imagination' to simulate plausible scenarios for using a tool, after which the LLM interacts with the tool to learn from its execution feedback.Both short-term and long-term memory are employed to improve the depth and breadth of the exploration, respectively.Comprehensive experiments on Tool-Bench show that STE substantially improves tool learning for LLMs under both in-context learning and fine-tuning settings, bringing a boost of 46.7% to Mistral-Instruct-7B and enabling it to outperform GPT-4.We also show effective continual learning of tools via a simple experience replay strategy.1 * Work done as an intern at Microsoft Semantic Machines. 1 Code and data available at https://github.com/ microsoft/simulated-trial-and-error.

Automatic web navigation aims to build a web agent that can follow language instructions to execute complex and diverse tasks on real-world websites. Existing work primarily takes HTML documents as input, which define the contents and action spaces (i.e., actionable elements and operations) of webpages. Nevertheless, HTML documents may not provide a clear task-related context for each element, making it hard to select the right (sequence of) actions. In this paper, we propose to contextualize HTML elements through their “dual views” in webpage screenshots: each HTML element has its corresponding bounding box and visual content in the screenshot. We build upon the insight-web developers tend to arrange task-related elements nearby on webpages to enhance user experiences-and propose to contextualize each element with its neighbor elements, using both tex-tual and visual features. The resulting representations of HTML elements are more informative for the agent to take action. We validate our method on the recently released Mind2Web dataset, which features diverse navigation domains and tasks on real-world websites. Our method consistently outperforms the baseline in all the scenarios, in-cluding cross-task, cross-website, and cross-domain ones.

Images of the natural world, collected by a variety of cameras, from drones to individual phones, are increasingly abundant sources of biological information. There is an ex-plosion of computational methods and tools, particularly computer vision, for extracting biologically relevant information from images for science and conservation. Yet most of these are bespoke approaches designed for a specific task and are not easily adaptable or extendable to new questions, contexts, and datasets. A vision model for general or-ganismal biology questions on images is of timely need. To approach this, we curate and release Tree Of Life-10m, the largest and most diverse ML-ready dataset of biology images. We then develop Bioclip, a foundation model for the tree of life, leveraging the unique properties of bi-ology captured by Treeoflife-10m, namely the abun-dance and variety of images of plants, animals, and fungi, together with the availability of rich structured biological knowledge. We rigorously benchmark our approach on di-verse fine-grained biology classification tasks and find that BloCLIP consistently and substantially outperforms existing baselines (by 16% to 17% absolute). Intrinsic evaluation reveals that BloCLIP has learned a hierarchical representation conforming to the tree of life, shedding light on its strong generalizability.11imageomics.github.io/bioclip has models, data and code.

We introduce MMMU: a new benchmark designed to evaluate multimodal models on massive multi-discipline tasks demanding college-level subject knowledge and deliberate reasoning. MMMU includes 11.5K meticulously collected multimodal questions from college exams, quizzes, and text-books, covering six core disciplines: Art & Design, Busi-ness, Science, Health & Medicine, Humanities & Social Science, and Tech & Engineering. These questions span 30 subjects and 183 subfields, comprising 30 highly het-erogeneous image types, such as charts, diagrams, maps, tables, music sheets, and chemical structures. Unlike existing benchmarks, MMMU focuses on advanced perception and reasoning with domain-specific knowledge, challenging models to perform tasks akin to those faced by experts. The evaluation of 28 open-source LMMs as well as the propri-etary GPT-4V(ision) and Gemini highlights the substantial challenges posed by MMMU. Even the advanced GPT-4V and Gemini Ultra only achieve accuracies of 56% and 59% respectively, indicating significant room for improvement. We believe MMMU will stimulate the community to build next-generation multimodal foundation models towards expert artificial general intelligence.

The applications of large language models (LLMs) have expanded well beyond the confines of text processing, signaling a new era where LLMs are envisioned as generalist agents capable of operating within complex environments.These environments are often highly expansive, making it impossible for the LLM to process them within its short-term memory.Motivated by recent research on extending the capabilities of LLMs with tools, we seek to investigate the intriguing potential of tools to augment LLMs in handling such complexity by introducing a novel class of tools, termed middleware, to aid in the proactive exploration within these massive environments.Such specialized tools can serve as a middleware layer shielding the LLM from environmental complexity.In two representative complex environmentsknowledge bases (KBs) and databases-we demonstrate the significant potential of augmenting language agents with tools in complex environments.Notably, equipped with the middleware, GPT-4 achieves 2.8× the performance of the best baseline in tasks requiring access to database content and 2.2× in KB tasks.Our findings illuminate the path for advancing language agents in real-world applications.1

The potential of Large Language Model (LLM) as agents has been widely acknowledged recently. Thus, there is an urgent need to quantitatively evaluate LLMs as agents on challenging tasks in interactive environments. We present AgentBench, a multi-dimensional benchmark that consists of 8 distinct environments to assess LLM-as-Agent's reasoning and decision-making abilities. Our extensive test over 29 API-based and open-sourced (OSS) LLMs shows that, while top commercial LLMs present a strong ability of acting as agents in complex environments, there is a significant disparity in performance between them and many OSS competitors that are no larger than 70B. We identify the typical reasons of failures in environments and LLMs, showing that poor long-term reasoning, decision-making, and instruction following abilities are the main obstacles for developing usable LLM agents. Improving instruction following and training on high quality multi-round alignment data could improve agent performance. And different from existing assumptions, training on code present ambivalent impacts on different agent tasks. Datasets, environments, and an integrated evaluation package for AgentBench are released at https://github.com/THUDM/AgentBench.

In the realm of large language models (LLMs), enhancing instruction-following capability often involves curating expansive training data. This is achieved through two primary schemes: i) Scaling-Inputs: Amplifying (input, output) pairs per task instruction, aiming for better instruction adherence. ii) Scaling Input-Free Tasks: Enlarging tasks, each composed of an (instruction, output) pair (without requiring a separate input anymore). However, LLMs under Scaling-Inputs tend to be overly sensitive to inputs, leading to misinterpretation or non-compliance with instructions. Conversely, Scaling Input-Free Tasks demands a substantial number of tasks but is less effective in instruction following when dealing with instances in Scaling-Inputs. This work introduces MUFFIN, a new scheme of instruction-following dataset curation. Specifically, we automatically Scale Tasks per Input by diversifying these tasks with various input facets. Experimental results across four zero-shot benchmarks, spanning both Scaling-Inputs and Scaling Input-Free Tasks schemes, reveal that LLMs, at various scales, trained on MUFFIN generally demonstrate superior instruction-following capabilities compared to those trained on the two aforementioned schemes.

We present a novel usage of Transformers to make image classification interpretable. Unlike mainstream classifiers that wait until the last fully connected layer to incorporate class information to make predictions, we investigate a proactive approach, asking each class to search for itself in an image. We realize this idea via a Transformer encoder-decoder inspired by DEtection TRansformer (DETR). We learn ''class-specific'' queries (one for each class) as input to the decoder, enabling each class to localize its patterns in an image via cross-attention. We name our approach INterpretable TRansformer (INTR), which is fairly easy to implement and exhibits several compelling properties. We show that INTR intrinsically encourages each class to attend distinctively; the cross-attention weights thus provide a faithful interpretation of the prediction. Interestingly, via ''multi-head'' cross-attention, INTR could identify different ''attributes'' of a class, making it particularly suitable for fine-grained classification and analysis, which we demonstrate on eight datasets. Our code and pre-trained models are publicly accessible at the Imageomics Institute GitHub site: https://github.com/Imageomics/INTR.

By providing external information to large language models (LLMs), tool augmentation (including retrieval augmentation) has emerged as a promising solution for addressing the limitations of LLMs' static parametric memory. However, how receptive are LLMs to such external evidence, especially when the evidence conflicts with their parametric memory? We present the first comprehensive and controlled investigation into the behavior of LLMs when encountering knowledge conflicts. We propose a systematic framework to elicit high-quality parametric memory from LLMs and construct the corresponding counter-memory, which enables us to conduct a series of controlled experiments. Our investigation reveals seemingly contradicting behaviors of LLMs. On the one hand, different from prior wisdom, we find that LLMs can be highly receptive to external evidence even when that conflicts with their parametric memory, given that the external evidence is coherent and convincing. On the other hand, LLMs also demonstrate a strong confirmation bias when the external evidence contains some information that is consistent with their parametric memory, despite being presented with conflicting evidence at the same time. These results pose important implications that are worth careful consideration for the further development and deployment of tool- and retrieval-augmented LLMs. Resources are available at https://github.com/OSU-NLP-Group/LLM-Knowledge-Conflict.

We introduce MAmmoTH, a series of open-source large language models (LLMs) specifically tailored for general math problem-solving. The MAmmoTH models are trained on MathInstruct, our meticulously curated instruction tuning dataset. MathInstruct is compiled from 13 math datasets with intermediate rationales, six of which have rationales newly curated by us. It presents a unique hybrid of chain-of-thought (CoT) and program-of-thought (PoT) rationales, and also ensures extensive coverage of diverse fields in math. The hybrid of CoT and PoT not only unleashes the potential of tool use but also allows different thought processes for different math problems. As a result, the MAmmoTH series substantially outperform existing open-source models on nine mathematical reasoning datasets across all scales with an average accuracy gain between 16% and 32%. Remarkably, our MAmmoTH-7B model reaches 33% on MATH (a competition-level dataset), which exceeds the best open-source 7B model (WizardMath) by 23%, and the MAmmoTH-34B model achieves 44% accuracy on MATH, even surpassing GPT-4’s CoT result. Our work underscores the importance of diverse problem coverage and the use of hybrid rationales in developing superior math generalist models.

Image retrieval, i.e., finding desired images given a reference image, inherently encompasses rich, multi-faceted search intents that are difficult to capture solely using image-based measures. Recent works leverage text instructions to allow users to more freely express their search intents. However, they primarily focus on image pairs that are visually similar and/or can be characterized by a small set of pre-defined relations. The core thesis of this paper is that text instructions can enable retrieving images with richer relations beyond visual similarity. To show this, we introduce MagicLens, a series of self-supervised image retrieval models that support open-ended instructions. MagicLens is built on a key novel insight: image pairs that naturally occur on the same web pages contain a wide range of implicit relations (e.g., inside view of), and we can bring those implicit relations explicit by synthesizing instructions via foundation models. Trained on 36.7M (query image, instruction, target image) triplets with rich semantic relations mined from the web, MagicLens achieves results comparable with or better than prior best on eight benchmarks of various image retrieval tasks, while maintaining high parameter efficiency with a significantly smaller model size. Additional human analyses on a 1.4M-image unseen corpus further demonstrate the diversity of search intents supported by MagicLens. Code and models are publicly available at the https://open-vision-language.github.io/MagicLens/.

Planning has been part of the core pursuit for artificial intelligence since its conception, but earlier AI agents mostly focused on constrained settings because many of the cognitive substrates necessary for human-level planning have been lacking. Recently, language agents powered by large language models (LLMs) have shown interesting capabilities such as tool use and reasoning. Are these language agents capable of planning in more complex settings that are out of the reach of prior AI agents? To advance this investigation, we propose TravelPlanner, a new planning benchmark that focuses on travel planning, a common real-world planning scenario. It provides a rich sandbox environment, various tools for accessing nearly four million data records, and 1,225 meticulously curated planning intents and reference plans. Comprehensive evaluations show that the current language agents are not yet capable of handling such complex planning tasks—even GPT-4 only achieves a success rate of 0.6%. Language agents struggle to stay on task, use the right tools to collect information, or keep track of multiple constraints. However, we note that the mere possibility for language agents to tackle such a complex problem is in itself non-trivial progress. TravelPlanner provides a challenging yet meaningful testbed for future language agents.

The recent development on large multimodal models (LMMs), especially GPT-4V(ision) and Gemini, has been quickly expanding the capability boundaries of multimodal models beyond traditional tasks like image captioning and visual question answering. In this work, we explore the potential of LMMs like GPT-4V as a generalist web agent that can follow natural language instructions to complete tasks on any given website. We propose SEEACT, a generalist web agent that harnesses the power of LMMs for integrated visual understanding and acting on the web. We evaluate on the recent MIND2WEB benchmark. In addition to standard offline evaluation on cached websites, we enable a new online evaluation setting by developing a tool that allows running web agents on live websites. We show that GPT-4V presents a great potential for web agents—it can successfully complete 51.1% of the tasks on live websites if we manually ground its textual plans into actions on the websites. This substantially outperforms text-only LLMs like GPT-4 or smaller models (FLAN-T5 and BLIP-2) specifically fine-tuned for web agents. However, grounding still remains a major challenge. Existing LMM grounding strategies like set-of-mark prompting turns out to be not effective for web agents, and the best grounding strategy we develop in this paper leverages both the HTML structure and visuals. Yet, there is still a substantial gap with oracle grounding, leaving ample room for further improvement. All code, data, and evaluation tools are available at https://github.com/OSU-NLP-Group/SeeAct.

In order to thrive in hostile and ever-changing natural environments, mammalian brains evolved to store large amounts of knowledge about the world and continually integrate new information while avoiding catastrophic forgetting. Despite the impressive accomplishments, large language models (LLMs), even with retrieval-augmented generation (RAG), still struggle to efficiently and effectively integrate a large amount of new experiences after pre-training. In this work, we introduce HippoRAG, a novel retrieval framework inspired by the hippocampal indexing theory of human long-term memory to enable deeper and more efficient knowledge integration over new experiences. HippoRAG synergistically orchestrates LLMs, knowledge graphs, and the Personalized PageRank algorithm to mimic the different roles of neocortex and hippocampus in human memory. We compare HippoRAG with existing RAG methods on multi-hop question answering (QA) and show that our method outperforms the state-of-the-art methods remarkably, by up to 20%. Single-step retrieval with HippoRAG achieves comparable or better performance than iterative retrieval like IRCoT while being 10-20 times cheaper and 6-13 times faster, and integrating HippoRAG into IRCoT brings further substantial gains. Finally, we show that our method can tackle new types of scenarios that are out of reach of existing methods.

Fine-tuning is arguably the most straightforward way to tailor a pre-trained model (e.g., a foundation model) to downstream applications, but it also comes with the risk of losing valuable knowledge the model had learned in pre-training. For example, fine-tuning a pre-trained classifier capable of recognizing a large number of classes to master a subset of classes at hand is shown to drastically degrade the model's accuracy in the other classes it had previously learned. As such, it is hard to further use the fine-tuned model when it encounters classes beyond the fine-tuning data. In this paper, we systematically dissect the issue, aiming to answer the fundamental question, "What has been damaged in the fine-tuned model?" To our surprise, we find that the fine-tuned model neither forgets the relationship among the other classes nor degrades the features to recognize these classes. Instead, the fine-tuned model often produces more discriminative features for these other classes, even if they were missing during fine-tuning! What really hurts the accuracy is the discrepant logit scales between the fine-tuning classes and the other classes, implying that a simple post-processing calibration would bring back the pre-trained model's capability and at the same time unveil the feature improvement over all classes. We conduct an extensive empirical study to demonstrate the robustness of our findings and provide preliminary explanations underlying them, suggesting new directions for future theoretical analysis.

Images are increasingly becoming the currency for documenting biodiversity on the planet, providing novel opportunities for accelerating scientific discoveries in the field of organismal biology, especially with the advent of large vision-language models (VLMs). We ask if pre-trained VLMs can aid scientists in answering a range of biologically relevant questions without any additional fine-tuning. In this paper, we evaluate the effectiveness of $12$ state-of-the-art (SOTA) VLMs in the field of organismal biology using a novel dataset, VLM4Bio, consisting of $469K$ question-answer pairs involving $30K$ images from three groups of organisms: fishes, birds, and butterflies, covering five biologically relevant tasks. We also explore the effects of applying prompting techniques and tests for reasoning hallucination on the performance of VLMs, shedding new light on the capabilities of current SOTA VLMs in answering biologically relevant questions using images.

We study whether transformers can learn to *implicitly* reason over parametric knowledge, a skill that even the most capable language models struggle with. Focusing on two representative reasoning types, composition and comparison, we consistently find that transformers *can* learn implicit reasoning, but only through *grokking*, i.e., extended training far beyond overfitting. The levels of generalization also vary across reasoning types: when faced with out-of-distribution examples, transformers fail to systematically generalize for composition but succeed for comparison. We delve into the model's internals throughout training, conducting analytical experiments that reveal: 1) the mechanism behind grokking, such as the formation of the generalizing circuit and its relation to the relative efficiency of generalizing and memorizing circuits, and 2) the connection between systematicity and the configuration of the generalizing circuit. Our findings guide data and training setup to better induce implicit reasoning and suggest potential improvements to the transformer architecture, such as encouraging cross-layer knowledge sharing. Furthermore, we demonstrate that for a challenging reasoning task with a large search space, GPT-4-Turbo and Gemini-1.5-Pro based on non-parametric memory fail badly regardless of prompting styles or retrieval augmentation, while a fully grokked transformer can achieve near-perfect accuracy, showcasing the power of parametric memory for complex reasoning.

OBJECTIVE: Although the mechanisms behind pharmacokinetic (PK) drug-drug interactions (DDIs) are well-documented, bridging the gap between this knowledge and clinical evidence of DDIs, especially for serious adverse drug reactions (SADRs), remains challenging. While leveraging the FDA Adverse Event Reporting System (FAERS) database along with disproportionality analysis tends to detect a vast number of DDI signals, this abundance complicates further investigation, such as validation through clinical trials. Our study proposed a framework to efficiently prioritize these signals and assessed their reliability using multi-source Electronic Health Records (EHR) to identify top candidates for further investigation. METHODS: We analyzed FAERS data spanning from January 2004 to March 2023, employing four established disproportionality methods: Proportional Reporting Ratio (PRR), Reporting Odds Ratio (ROR), Multi-item Gamma Poisson Shrinker (MGPS), and Bayesian Confidence Propagating Neural Network (BCPNN). Building upon these models, we developed four ranking models to prioritize DDI-SADR signals and cross-referenced signals with DrugBank. To validate the top-ranked signals, we employed longitudinal EHRs from Vanderbilt University Medical Center and the All of Us research program. The performance of each model was assessed by counting how many of the top-ranked signals were confirmed by EHRs and calculating the average ranking of these confirmed signals. RESULTS: Out of 189 DDI-SADR signals identified by all four disproportionality methods, only two were documented in the DrugBank database. By prioritizing the top 20 signals as determined by each of the four disproportionality methods and our four ranking models, 58 unique DDI-SADR signals were selected for EHR validations. Of these, five signals were confirmed. The ranking model, which integrated the MGPS and BCPNN, demonstrated superior performance by assigning the highest priority to those five EHR-confirmed signals. CONCLUSION: The fusion of disproportionality analysis with ranking models, validated through multi-source EHRs, presents a groundbreaking approach to pharmacovigilance. Our study's confirmation of five significant DDI-SADRs, previously unrecorded in the DrugBank database, highlights the essential role of advanced data analysis techniques in identifying ADRs.

A key missing capacity of current language models (LMs) is grounding to real-world environments.Most existing work for grounded language understanding uses LMs to directly generate plans that can be executed in the environment to achieve the desired effects.It thereby casts the burden of ensuring grammaticality, faithfulness, and controllability all on the LMs.We propose Pangu, a generic framework for grounded language understanding that capitalizes on the discriminative ability of LMs instead of their generative ability.Pangu consists of a symbolic agent and a neural LM working in a concerted fashion: The agent explores the environment to incrementally construct valid plans, and the LM evaluates the plausibility of the candidate plans to guide the search process.A case study on the challenging problem of knowledge base question answering (KBQA), which features a massive environment, demonstrates the remarkable effectiveness and flexibility of Pangu: A BERT-base LM is sufficient for setting a new record on standard KBQA datasets, and larger LMs further bring substantial gains.Pangu also enables, for the first time, effective few-shot in-context learning for KBQA with large LMs such as Codex. 1

Question answering over knowledge bases is considered a difficult problem due to the challenge of generalizing to a wide variety of possible natural language questions.Additionally, the heterogeneity of knowledge base schema items between different knowledge bases often necessitates specialized training for different knowledge base question-answering (KBQA) datasets.To handle questions over diverse KBQA datasets with a unified trainingfree framework, we propose KB-BINDER, which for the first time enables few-shot incontext learning over KBQA tasks.Firstly, KB-BINDER leverages large language models like Codex to generate logical forms as the draft for a specific question by imitating a few demonstrations.Secondly, KB-BINDER grounds on the knowledge base to bind the generated draft to an executable one with BM25 score matching.The experimental results on four public heterogeneous KBQA datasets show that KB-BINDER can achieve a strong performance with only a few in-context demonstrations.Especially on GraphQA and 3-hop MetaQA, KB-BINDER can even outperform the state-of-the-art trained models.On GrailQA and WebQSP, our model is also on par with other fully-trained models.We believe KB-BINDER can serve as an important baseline for future research.Our code is available at

Fatemehsadat Mireshghallah, Yu Su, Tatsunori Hashimoto, Jason Eisner, Richard Shin. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023.

This paper studies a new task of federated learning (FL) for semantic parsing, where multiple clients collaboratively train one global model without sharing their semantic parsing data.By leveraging data from multiple clients, the FL paradigm can be especially beneficial for clients that have little training data to develop a data-hungry neural semantic parser on their own.We propose an evaluation setup to study this task, where we re-purpose widely-used single-domain text-to-SQL datasets as clients to form a realistic heterogeneous FL setting and collaboratively train a global model.As standard FL algorithms suffer from the high client heterogeneity in our realistic setup, we further propose a novel LOss Reduction Adjusted Reweighting (Lorar) mechanism to mitigate the performance degradation, which adjusts each client's contribution to the global model update based on its training loss reduction during each round.Our intuition is that the larger the loss reduction, the further away the current global model is from the client's local optimum, and the larger weight the client should get.By applying Lorar to three widely adopted FL algorithms (FedAvg, FedOPT and FedProx), we observe that their performance can be improved substantially on average (4%-20% absolute gain under MacroAvg) and that clients with smaller datasets enjoy larger performance gains.In addition, the global model converges faster for almost all the clients. 1

Knowledge graph (KG) link prediction aims to infer new facts based on existing facts in the KG. Recent studies have shown that using the graph neighborhood of a node via graph neural networks (GNNs) provides more useful information compared to just using the query information. Conventional GNNs for KG link prediction follow the standard message-passing paradigm on the entire KG, which leads to superfluous computation, over-smoothing of node representations, and also limits their expressive power. On a large scale, it becomes computationally expensive to aggregate useful information from the entire KG for inference. To address the limitations of existing KG link prediction frameworks, we propose a novel retrieve-and-read framework, which first retrieves a relevant subgraph context for the query and then jointly reasons over the context and the query with a high-capacity reader. As part of our exemplar instantiation for the new framework, we propose a novel Transformer-based GNN as the reader, which incorporates graph-based attention structure and cross-attention between query and context for deep fusion. This simple yet effective design enables the model to focus on salient context information relevant to the query. Empirical results on two standard KG link prediction datasets demonstrate the competitive performance of the proposed method. Furthermore, our analysis yields valuable insights for designing improved retrievers within the framework.

This study focuses on using large language models (LLMs) as a planner for embodied agents that can follow natural language instructions to complete complex tasks in a visually-perceived environment. The high data cost and poor sample efficiency of existing methods hinders the development of versatile agents that are capable of many tasks and can learn new tasks quickly. In this work, we propose a novel method, LLM-Planner, that harnesses the power of large language models to do few-shot planning for embodied agents. We further propose a simple but effective way to enhance LLMs with physical grounding to generate and update plans that are grounded in the current environment. Experiments on the ALFRED dataset show that our method can achieve very competitive few-shot performance: Despite using less than 0.5% of paired training data, LLM-Planner achieves competitive performance with recent baselines that are trained using the full training data. Existing methods can barely complete any task successfully under the same few-shot setting. Our work opens the door for developing versatile and sample-efficient embodied agents that can quickly learn many tasks.1

We propose a learning problem involving adapting a pre-trained source model to the target domain for classifying all classes that appeared in the source data, using target data that covers only a partial label space. This problem is practical, as it is unrealistic for the target end-users to collect data for all classes prior to adaptation. However, it has received limited attention in the literature. To shed light on this issue, we construct benchmark datasets and conduct extensive experiments to uncover the inherent challenges. We found a dilemma --- on the one hand, adapting to the new target domain is important to claim better performance; on the other hand, we observe that preserving the classification accuracy of classes missing in the target adaptation data is highly challenging, let alone improving them. To tackle this, we identify two key directions: 1) disentangling domain gradients from classification gradients, and 2) preserving class relationships. We present several effective solutions that maintain the accuracy of the missing classes and enhance the overall performance, establishing solid baselines for holistic transfer of pre-trained models with partial target data.

We introduce Mind2Web, the first dataset for developing and evaluating generalist agents for the web that can follow language instructions to complete complex tasks on any website. Existing datasets for web agents either use simulated websites or only cover a limited set of websites and tasks, thus not suitable for generalist web agents. With over 2,000 open-ended tasks collected from 137 websites spanning 31 domains and crowdsourced action sequences for the tasks, Mind2Web provides three necessary ingredients for building generalist web agents: 1) diverse domains, websites, and tasks, 2) use of real-world websites instead of simulated and simplified ones, and 3) a broad spectrum of user interaction patterns. Based on Mind2Web, we conduct an initial exploration of using large language models (LLMs) for building generalist web agents. While the raw HTML of real-world websites are often too large to be fed to LLMs, we show that first filtering it with a small LM significantly improves the effectiveness and efficiency of LLMs. Our solution demonstrates a decent level of performance, even on websites or entire domains the model has never seen before, but there is still a substantial room to improve towards truly generalizable agents. We open-source our dataset, model implementation, and trained models (https://osu-nlp-group.github.io/Mind2Web) to facilitate further research on building a generalist agent for the web.

Text-guided image editing is widely needed in daily life, ranging from personal use to professional applications such as Photoshop.However, existing methods are either zero-shot or trained on an automatically synthesized dataset, which contains a high volume of noise.Thus, they still require lots of manual tuning to produce desirable outcomes in practice.To address this issue, we introduce MagicBrush, the first large-scale, manually annotated dataset for instruction-guided real image editing that covers diverse scenarios: single-turn, multi-turn, mask-provided, and mask-free editing.MagicBrush comprises over 10K manually annotated triplets (source image, instruction, target image), which supports trainining large-scale text-guided image editing models.We fine-tune InstructPix2Pix on MagicBrush and show that the new model can produce much better images according to human evaluation.We further conduct extensive experiments to evaluate current image editing baselines from multiple dimensions including quantitative, qualitative, and human evaluations.The results reveal the challenging nature of our dataset and the gap between current baselines and real-world editing needs.

Lingbo Mo, Shijie Chen, Ziru Chen, Xiang Deng, Ashley Lewis, Sunit Singh, Samuel Stevens, Chang-You Tai, Zhen Wang, Xiang Yue, Tianshu Zhang, Yu Su, Huan Sun. Proceedings of the 24th Annual Meeting of the Special Interest Group on Discourse and Dialogue. 2023.

Text-to-SQL parsers map natural language questions to programs that are executable over tables to generate answers, and are typically evaluated on large-scale datasets like SPIDER (Yu et al., 2018).We argue that existing benchmarks fail to capture a certain out-of-domain generalization problem that is of significant practical importance: matching domain specific phrases to composite operations over columns.To study this problem, we propose a synthetic dataset and a re-purposed train/test split of the SQUALL dataset (Shi et al., 2020) as new benchmarks to quantify domain generalization over column operations.Our results indicate that existing state-of-the-art parsers struggle in these benchmarks.We propose to address this problem by incorporating prior domain knowledge by preprocessing table schemas, and design a method that consists of two components: schema expansion and schema pruning.This method can be easily applied to multiple existing base parsers, and we show that it significantly outperforms baseline parsers on this domain generalization problem, boosting the underlying parsers' overall performance by up to 13.8% relative accuracy gain (5.1% absolute) on the new SQUALL data split.

Question answering on knowledge bases (KBQA) poses a unique challenge for semantic parsing research due to two intertwined challenges: large search space and ambiguities in schema linking. Conventional ranking-based KBQA models, which rely on a candidate enumeration step to reduce the search space, struggle with flexibility in predicting complicated queries and have impractical running time. In this paper, we present ArcaneQA, a novel generation-based model that addresses both the large search space and the schema linking challenges in a unified framework with two mutually boosting ingredients: dynamic program induction for tackling the large search space and dynamic contextualized encoding for schema linking. Experimental results on multiple popular KBQA datasets demonstrate the highly competitive performance of ArcaneQA in both effectiveness and efficiency.

We study the problem of developing autonomous agents that can follow human instructions to infer and perform a sequence of actions to complete the underlying task. Significant progress has been made in recent years, especially for tasks with short horizons. However, when it comes to long-horizon tasks with extended sequences of actions, an agent can easily ignore some instructions or get stuck in the middle of the long instructions and eventually fail the task. To address this challenge, we propose a modelagnostic milestone-based task tracker (M-TRACK) to guide the agent and monitor its progress. Specifically, we propose a milestone builder that tags the instructions with navigation and interaction milestones which the agent needs to complete step by step, and a milestone checker that systemically checks the agent's progress in its current milestone and determines when to proceed to the next. On the challenging ALFRED dataset, our M-Track leads to a notable 33% and 52% relative improvement in unseen success rate over two competitive base models.

Elias Stengel-Eskin, Emmanouil Antonios Platanios, Adam Pauls, Sam Thomson, Hao Fang, Benjamin Van Durme, Jason Eisner, Yu Su. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022.

Vardaan Pahuja, Yu Gu, Wenhu Chen, Mehdi Bahrami, Lei Liu, Wei-Peng Chen, Yu Su. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). 2021.

We present ReasonBERT, a pre-training method that augments language models with the ability to reason over long-range relations and multiple, possibly hybrid, contexts.Unlike existing pre-training methods that only harvest learning signals from local contexts of naturally occurring texts, we propose a generalized notion of distant supervision to automatically connect multiple pieces of text and tables to create pre-training examples that require long-range reasoning.Different types of reasoning are simulated, including intersecting multiple pieces of evidence, bridging from one piece of evidence to another, and detecting unanswerable cases.We conduct a comprehensive evaluation on a variety of extractive question answering datasets ranging from single-hop to multi-hop and from text-only to table-only to hybrid that require various reasoning capabilities and show that ReasonBERT achieves remarkable improvement over an array of strong baselines.Fewshot experiments further demonstrate that our pre-training method substantially improves sample efficiency. 1

Pengcheng Yin, Hao Fang, Graham Neubig, Adam Pauls, Emmanouil Antonios Platanios, Yu Su, Sam Thomson, Jacob Andreas. Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2021.

Existing studies on question answering on knowledge bases (KBQA) mainly operate with the standard i.i.d. assumption, i.e., training distribution over questions is the same as the test distribution. However, i.i.d. may be neither achievable nor desirable on large-scale KBs because 1) true user distribution is hard to capture and 2) randomly sampling training examples from the enormous space would be data-inefficient. Instead, we suggest that KBQA models should have three levels of built-in generalization: i.i.d., compositional, and zero-shot. To facilitate the development of KBQA models with stronger generalization, we construct and release a new large-scale, high-quality dataset with 64,331 questions, GrailQA, and provide evaluation settings for all three levels of generalization. In addition, we propose a novel BERT-based KBQA model. The combination of our dataset and model enables us to thoroughly examine and demonstrate, for the first time, the key role of pre-trained contextual embeddings like BERT in the generalization of KBQA.1

Neural natural language generation (NLG) models have recently shown remarkable progress in fluency and coherence.However, existing studies on neural NLG are primarily focused on surface-level realizations with limited emphasis on logical inference, an important aspect of human thinking and language.In this paper, we suggest a new NLG task where a model is tasked with generating natural language statements that can be logically entailed by the facts in an open-domain semi-structured table.To facilitate the study of the proposed logical NLG problem, we use the existing Tab-Fact dataset (Chen et al., 2019) featured with a wide range of logical/symbolic inferences as our testbed, and propose new automatic metrics to evaluate the fidelity of generation models w.r.t.logical inference.The new task poses challenges to the existing monotonic generation frameworks due to the mismatch between sequence order and logical order.In our experiments, we comprehensively survey different generation architectures (LSTM, Transformer, Pre-Trained LM) trained with different algorithms (RL, Adversarial Training, Coarse-to-Fine) on the dataset and made following observations: 1) Pre-Trained LM can significantly boost both the fluency and logical fidelity metrics, 2) RL and Adversarial Training are trading fluency for fidelity, 3) Coarse-to-Fine generation can help partially alleviate the fidelity issue while maintaining high language fluency.

Data-to-text generation has recently attracted substantial interests due to its wide applications. Existing methods have shown impressive performance on an array of tasks. However, they rely on a significant amount of labeled data for each task, which is costly to acquire and thus limits their application to new tasks and domains. In this paper, we propose to leverage pre-training and transfer learning to address this issue. We propose a knowledge-grounded pre-training (KGPT), which consists of two parts, 1) a general knowledge-grounded generation model to generate knowledge-enriched text. 2) a pre-training paradigm on a massive knowledge-grounded text corpus crawled from the web. The pre-trained model can be fine-tuned on various data-to-text generation tasks to generate task-specific text. We adopt three settings, namely fully-supervised, zero-shot, few-shot to evaluate its effectiveness. Under the fully-supervised setting, our model can achieve remarkable gains over the known baselines. Under zero-shot setting, our model without seeing any examples achieves over 30 ROUGE-L on WebNLG while all other baselines fail. Under the few-shot setting, our model only needs about one-fifteenth as many labeled examples to achieve the same level of performance as baseline models. These experiments consistently prove the strong generalization ability of our proposed framework.

Despite the widely successful applications, building a semantic parser is still a tedious process in practice with challenges from costly data annotation and privacy risks.We suggest an alternative, human-in-the-loop methodology for learning semantic parsers directly from users.A semantic parser should be introspective of its uncertainties and prompt for user demonstrations when uncertain.In doing so it also gets to imitate the user behavior and continue improving itself autonomously with the hope that eventually it may become as good as the user in interpreting their questions.To combat the sparsity of demonstrations, we propose a novel annotation-efficient imitation learning algorithm, which iteratively collects new datasets by mixing demonstrated states and confident predictions and retrains the semantic parser in a Dataset Aggregation fashion (Ross et al., 2011).We provide a theoretical analysis of its cost bound and also empirically demonstrate its promising performance on the text-to-SQL problem. 1

We describe an approach to task-oriented dialogue in which dialogue state is represented as a dataflow graph. A dialogue agent maps each user utterance to a program that extends this graph. Programs include metacomputation operators for reference and revision that reuse dataflow fragments from previous turns. Our graph-based state enables the expression and manipulation of complex user intents, and explicit metacomputation makes these intents easier for learned models to predict. We introduce a new dataset, SMCalFlow, featuring complex dialogues about events, weather, places, and people. Experiments show that dataflow graphs and metacomputation substantially improve representability and predictability in these natural dialogues. Additional experiments on the MultiWOZ dataset show that our dataflow representation enables an otherwise off-the-shelf sequence-to-sequence model to match the best existing task-specific state tracking model. The SMCalFlow dataset, code for replicating experiments, and a public leaderboard are available at https://www.microsoft.com/en-us/research/project/dataflow-based-dialogue-semantic-machines .

Although promising results have been achieved in video captioning, existing models are limited to the fixed inventory of activities in the training corpus, and do not generalize to open vocabulary scenarios. Here we introduce a novel task, zeroshot video captioning, that aims at describing out-of-domain videos of unseen activities. Videos of different activities usually require different captioning strategies in many aspects, i.e. word selection, semantic construction, and style expression etc, which poses a great challenge to depict novel activities without paired training data. But meanwhile, similar activities share some of those aspects in common. Therefore, we propose a principled Topic-Aware Mixture of Experts (TAMoE) model for zero-shot video captioning, which learns to compose different experts based on different topic embeddings, implicitly transferring the knowledge learned from seen activities to unseen ones. Besides, we leverage external topic-related text corpus to construct the topic embedding for each activity, which embodies the most relevant semantic vectors within the topic. Empirical results not only validate the effectiveness of our method in utilizing semantic knowledge for video captioning, but also show its strong generalization ability when describing novel activities.

Pre-trained embeddings such as word embeddings and sentence embeddings are fundamental tools facilitating a wide range of downstream NLP tasks.In this work, we investigate how to learn a general-purpose embedding of textual relations, defined as the shortest dependency path between entities.Textual relation embedding provides a level of knowledge between word/phrase level and sentence level, and we show that it can facilitate downstream tasks requiring relational understanding of the text.To learn such an embedding, we create the largest distant supervision dataset by linking the entire English ClueWeb09 corpus to Freebase.We use global co-occurrence statistics between textual and knowledge base relations as the supervision signal to train the embedding.Evaluation on two relational understanding tasks demonstrates the usefulness of the learned textual relation embedding.

We consider a discrete-time linear quadratic Gaussian networked control setting where the (full information) observer and controller are separated by a fixed-rate noiseless channel. We study the event-triggered control setup in which the encoder may choose to either transmit a packet or remain silent. We recast this problem into that of fixed-rate quantization with an extra symbol that corresponds to the silence event. This way, controlling the average transmission rate is possible by constraining the minimal probability of the silence symbol. We supplement our theoretical framework with numerical simulations.

Ziyu Yao, Yu Su, Huan Sun, Wen-tau Yih. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019.

In this work we study the hierarchical organization of technical documents, where given a set of documents and a hierarchy of categories, the goal is to assign documents to their corresponding categories. Unlike prior work on supervised hierarchical document categorization that relies on large amount of labeled training data, which is expensive to obtain in closed technical domain and tends to stale as new knowledge emerges, we study this problem in a weak supervision setting, by leveraging semantic information from concepts. The core idea is to project both documents and categories into a common concept embedding space, where their fine-grained similarity can be easily and effectively computed. Experiments over real-world datasets from the subject of computer science, physics & mathematics, and medicine demonstrated the superior performance of our approach over a wide range of state of the art baseline approaches.

The recent advance in deep learning and semantic parsing has significantly improved the translation accuracy of natural language questions to structured queries.However, further improvement of the existing approaches turns out to be quite challenging.Rather than solely relying on algorithmic innovations, in this work, we introduce DialSQL, a dialoguebased structured query generation framework that leverages human intelligence to boost the performance of existing algorithms via user interaction.DialSQL is capable of identifying potential errors in a generated SQL query and asking users for validation via simple multi-choice questions.User feedback is then leveraged to revise the query.We design a generic simulator to bootstrap synthetic training dialogues and evaluate the performance of DialSQL on the WikiSQL dataset.Using SQLNet as a black box query generation tool, DialSQL improves its performance from 61.3% to 69.0% using only 2.4 validation questions per dialogue.

Task-oriented dialog systems are becoming pervasive, and many companies heavily rely on them to complement human agents for customer service in call centers.With globalization, the need for providing cross-lingual customer support becomes more urgent than ever.However, cross-lingual support poses great challenges-it requires a large amount of additional annotated data from native speakers.In order to bypass the expensive human annotation and achieve the first step towards the ultimate goal of building a universal dialog system, we set out to build a cross-lingual state tracking framework.Specifically, we assume that there exists a source language with dialog belief tracking annotations while the target languages have no annotated dialog data of any form.Then, we pre-train a state tracker for the source language as a teacher, which is able to exploit easy-to-access parallel data.We then distill and transfer its own knowledge to the student state tracker in target languages.We specifically discuss two types of common parallel resources: bilingual corpus and bilingual dictionary, and design different transfer learning strategies accordingly.Experimentally, we successfully use English state tracker as the teacher to transfer its knowledge to both Italian and German trackers and achieve promising results.

WikiSQL is a newly released dataset for studying the natural language sequence to SQL translation problem.The SQL queries in Wik-iSQL are simple: Each involves one relation and does not have any join operation.Despite of its simplicity, none of the publicly reported structured query generation models can achieve an accuracy beyond 62%, which is still far from enough for practical use.In this paper, we ask two questions, "Why is the accuracy still low for such simple queries?" and "What does it take to achieve 100% accuracy on WikiSQL?"To limit the scope of our study, we focus on the WHERE clause in SQL.The answers will help us gain insights about the directions we should explore in order to further improve the translation accuracy.We will then investigate alternative solutions to realize the potential ceiling performance on WikiSQL.Our proposed solution can reach up to 88.6% condition accuracy on the WikiSQL dataset.

In this work, we study the problem of concept mining, which serves as the first step in transforming unstructured text into structured information, and supports downstream analytical tasks such as information extraction, organization, recommendation and search. Previous work mainly relies on statistical signals, existing knowledge bases, or predefined linguistic patterns. In this work, we propose a novel approach that mines concepts based on their occurrence contexts, by learning embedding vector representations that summarize the context information for each possible candidates, and use these embeddings to evaluate the concept's global quality and their fitness to each local context. Experiments over several real-world corpora demonstrate the superior performance of our method. A publicly available implementation is provided at https://github.com/kleeeeea/ECON.

Yu Su, Honglei Liu, Semih Yavuz, Izzeddin Gür, Huan Sun, Xifeng Yan. Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers). 2018.

Most conventional document categorization methods require a large number of documents with labeled categories for training. These methods are hard to be applied in scenarios, such as scientific publications, where training data is expensive to obtain and categories could change over years and across domains. In this work, we propose UNEC, an unsupervised representation learning model that directly categories documents without the need of labeled training data. Specifically, we develop a novel cascade embedding approach. We first embed concepts, i.e., significant phrases mined from scientific publications, into continuous vectors, which capture concept semantics. Based on the concept similarity graph built from the concept embedding, we further embed concepts into a hidden category space, where the category information of concepts becomes explicit. Finally we categorize documents by jointly considering the category attribution of their concepts. Our experimental results show that UNEC significantly outperforms several strong baselines on a number of real scientific corpora, under both automatic and manual evaluation.

The rapidly increasing ubiquity of computing puts a great demand on next-generation human-machine interfaces. Natural language interfaces, exemplified by virtual assistants like Apple Siri and Microsoft Cortana, are widely believed to be a promising direction. However, current natural language interfaces provide users with little help in case of incorrect interpretation of user commands. We hypothesize that the support of fine-grained user interaction can greatly improve the usability of natural language interfaces. In the specific setting of natural language interface to web APIs, we conduct a systematic study to verify our hypothesis. To facilitate this study, we propose a novel modular sequence-to-sequence model to create interactive natural language interfaces. By decomposing the complex prediction process of a typical sequence-to-sequence model into small, highly-specialized prediction units called modules, it becomes straightforward to explain the model prediction to the user, and solicit user feedback to correct possible prediction errors at a fine-grained level. We test our hypothesis by comparing an interactive natural language interface with its non-interactive version through both simulation and human subject experiments with real-world APIs. We show that with the interactive natural language interface, users can achieve a higher success rate and a lower task completion time, which lead to greatly improved user satisfaction.

No abstract available.

As the Web evolves towards a service-oriented architecture, application program interfaces (APIs) are becoming an increasingly important way to provide access to data, services, and devices. We study the problem of natural language interface to APIs (NL2APIs), with a focus on web APIs for web services. Such NL2APIs have many potential benefits, for example, facilitating the integration of web services into virtual assistants.

Existing studies on semantic parsing mainly focus on the in-domain setting.We formulate cross-domain semantic parsing as a domain adaptation problem: train a semantic parser on some source domains and then adapt it to the target domain.Due to the diversity of logical forms in different domains, this problem presents unique and intriguing challenges.By converting logical forms into canonical utterances in natural language, we reduce semantic parsing to paraphrasing, and develop an attentive sequence-to-sequence paraphrase model that is general and flexible to adapt to different domains.We discover two problems, small micro variance and large macro variance, of pretrained word embeddings that hinder their direct use in neural networks, and propose standardization techniques as a remedy.On the popular OVERNIGHT dataset, which contains eight domains, we show that both cross-domain training and standardized pre-trained word embedding can bring significant improvement.

The existing factoid QA systems often lack a post-inspection component that can help models recover from their own mistakes.In this work, we propose to crosscheck the corresponding KB relations behind the predicted answers and identify potential inconsistencies.Instead of developing a new model that accepts evidences collected from these relations, we choose to plug them back to the original questions directly and check if the revised question makes sense or not.A bidirectional LSTM is applied to encode revised questions.We develop a scoring mechanism over the revised question encodings to refine the predictions of a base QA system.This approach can improve the F 1 score of STAGG (Yih et al., 2015), one of the leading QA systems, from 52.5% to 53.9% on WE-BQUESTIONS data.

Given a question and a set of answer candidates, answer triggering determines whether the candidate set contains any correct answers.If yes, it then outputs a correct one.In contrast to existing pipeline methods which first consider individual candidate answers separately and then make a prediction based on a threshold, we propose an end-to-end deep neural network framework, which is trained by a novel group-level objective function that directly optimizes the answer triggering performance.Our objective function penalizes three potential types of error and allows training the framework in an end-to-end manner.Experimental results on the WIKIQA benchmark show that our framework outperforms the state of the arts by a 6.6% absolute gain under F 1 measure 1 .

We present a semi-automated framework for constructing factoid question answering (QA) datasets, where an array of question characteristics are formalized, including structure complexity, function, commonness, answer cardinality, and paraphrasing.Instead of collecting questions and manually characterizing them, we employ a reverse procedure, first generating a kind of graph-structured logical forms from a knowledge base, and then converting them into questions.Our work is the first to generate questions with explicitly specified characteristics for QA evaluation.We construct a new QA dataset with over 5,000 logical form-question pairs, associated with answers from the knowledge base, and show that datasets constructed in this way enable finegrained analyses of QA systems.The dataset can be found in https://github.com/ysu1989/GraphQuestions.

In this work, we show the possibility of inferring the answer type before solving a factoid question and leveraging the type information to improve semantic parsing.By replacing the topic entity in a question with its type, we are able to generate an abstract form of the question, whose answer corresponds to the answer type of the original question.A bidirectional LSTM model is built to train over the abstract form of questions and infer their answer types.It is also observed that if we convert a question into a statement form, our LSTM model achieves better accuracy.Using the predicted type information to rerank the logical forms returned by AgendaIL, one of the leading semantic parsers, we are able to improve the F1-score from 49.7% to 52.6% on the WE-BQUESTIONS data.

As a fundamental technique for graph analysis, graph kernels have been successfully applied to a wide range of problems. Unfortunately, the high computational complexity of existing graph kernels is limiting their further applications to larger-scale graph datasets. In this paper, we propose a fast graph kernel, the descriptor matching (DM) kernel, for graphs with both categorical and numerical attributes. The computation time of the DM kernel is linear with respect to graph size. On graphs with n nodes and m edges, the kernel computation for two graphs can be done in O(n+m) time. Although there are other linear-time graph kernels, most of them are restricted to graphs with only categorical attributes; their efficiency mainly comes from the sparseness of the feature space resulted from the mutually orthogonal categorical attributes. Extensive experiments on both synthetic and real-world graph datasets show promising performance of DM in both accuracy and efficiency: On graphs with both categorical and numerical attributes, DM is orders of magnitude faster than several state-of-the-art graph kernels, while being much more accurate than the only graph kernel that is more efficient.

Tables are pervasive on the Web. Informative web tables range across a large variety of topics, which can naturally serve as a significant resource to satisfy user information needs. Driven by such observations, in this paper, we investigate an important yet largely under-addressed problem: Given millions of tables, how to precisely retrieve table cells to answer a user question. This work proposes a novel table cell search framework to attack this problem. We first formulate the concept of a relational chain which connects two cells in a table and represents the semantic relation between them. With the help of search engine snippets, our framework generates a set of relational chains pointing to potentially correct answer cells. We further employ deep neural networks to conduct more fine-grained inference on which relational chains best match the input question and finally extract the corresponding answer cells. Based on millions of tables crawled from the Web, we evaluate our framework in the open-domain question answering (QA) setting, using both the well-known WebQuestions dataset and user queries mined from Bing search engine logs. On WebQuestions, our framework is comparable to state-of-the-art QA systems based on knowledge bases (KBs), while on Bing queries, it outperforms other systems with a 56.7% relative gain. Moreover, when combined with results from our framework, KB-based QA performance can obtain a relative improvement of 28.1% to 66.7%, demonstrating that web tables supply rich knowledge that might not exist or is difficult to be identified in existing KBs.

The big data era is witnessing a prevalent shift of data from homogeneous to heterogeneous, from isolated to linked. Exemplar outcomes of this shift are a wide range of graph data such as information, social, and knowledge graphs. The unique characteristics of graph data are challenging traditional search techniques like SQL and keyword search. Graph query is emerging as a promising complementary search form. In this paper, we study how to improve graph query by relevance feedback. Specifically, we focus on knowledge graph query, and formulate the graph relevance feedback (GRF) problem. We propose a general GRF framework that is able to (1) tune the original ranking function based on user feedback and (2) further enrich the query itself by mining new features from user feedback. As a consequence, a query-specific ranking function is generated, which is better aligned with the user search intent. Given a newly learned ranking function based on user feedback, we further investigate whether we shall re-rank the existing answers, or choose to search from scratch. We propose a strategy to train a binary classifier to predict which action will be more beneficial for a given query. The GRF framework is applied to searching DBpedia with graph queries derived from YAGO and Wikipedia. Experiment results show that GRF can improve the mean average precision by 80% to 100%.

Mobile networking researchers have long searched for large-scale, fine-grained traces of human movement, which have remained elusive for both privacy and logistical reasons. Recently, researchers have begun to focus on geosocial mobility traces, e.g. Foursquare checkin traces, because of their availability and scale. But are we conceding correctness in our zeal for data? In this paper, we take initial steps towards quantifying the value of geosocial datasets using a large ground truth dataset gathered from a user study. By comparing GPS traces against Foursquare checkins, we find that a large portion of visited locations is missing from checkins, and most checkin events are either forged or superfluous events. We characterize extraneous checkins, describe possible techniques for their detection, and show that both extraneous and missing checkins introduce significant errors into applications driven by these traces.