Recent advances in reasoning with large language models (LLMs) have popularized Long Chain-of-Thought (LCoT), a strategy that encourages deliberate and step-by-step reasoning before producing a final answer.While LCoTs have enabled expert-level performance in complex tasks, how the internal structures of their reasoning chains drive, or even predict, the correctness of final answers remains a critical yet underexplored question.In this work, we present LCoT2Tree, an automated framework that converts sequential LCoTs into hierarchical tree structures and thus enables deeper structural analysis of LLM reasoning.Using graph neural networks (GNNs), we reveal that structural patterns extracted by LCoT2Tree, including exploration, backtracking, and verification, serve as stronger predictors of final performance across a wide range of tasks and models.Leveraging an explainability technique, we further identify critical thought patterns such as over-branching that account for failures.Beyond diagnostic insights, the structural patterns by LCoT2Tree support practical applications, including improving Best-of-N decoding effectiveness.Overall, our results underscore the critical role of internal structures of reasoning chains, positioning LCoT2Tree as a powerful tool for diagnosing, interpreting, and improving reasoning in LLMs.The code is available at our GitHub repository.

Catastrophic forgetting (CF) poses a significant challenge in machine learning, where a model forgets previously learned information upon learning new tasks. Despite the advanced capabilities of Large Language Models (LLMs), they continue to face challenges with CF during continual learning. The majority of existing research focuses on analyzing forgetting patterns through a singular training sequence, thereby overlooking the intricate effects that diverse tasks have on model behavior. Our study explores CF across various settings, discovering that model forgetting is influenced by both the specific training tasks and the models themselves. To this end, we interpret forgetting by examining the function vector (FV), a compact representation of functions in LLMs, offering a model-dependent indicator for the occurrence of CF. Through theoretical and empirical analyses, we demonstrated that CF in LLMs primarily stems from biases in function activation rather than the overwriting of task processing functions. Leveraging these insights, we propose a novel function vector guided training methodology, incorporating a regularization technique to stabilize the FV and mitigate forgetting. Empirical tests on four benchmarks confirm the effectiveness of our proposed training method, substantiating our theoretical framework concerning CF and model function dynamics.

Abstract The advent of large language models marks a revolutionary breakthrough in artificial intelligence. With the unprecedented scale of training and model parameters, the capability of large language models has been dramatically improved, leading to human-like performances in understanding, language synthesizing, common-sense reasoning, etc. Such a major leap forward in general AI capacity will fundamentally change the pattern of how personalization is conducted. For one thing, it will reform the way of interaction between humans and personalization systems. Instead of being a passive medium of information filtering, like conventional recommender systems and search engines, large language models present the foundation for active user engagement. On top of such a new foundation, users’ requests can be proactively explored, and users’ required information can be delivered in a natural, interactable, and explainable way. For another thing, it will also considerably expand the scope of personalization, making it grow from the sole function of collecting personalized information to the compound function of providing personalized services. By leveraging large language models as a general-purpose interface, the personalization systems may compile user’s requests into plans, calls the functions of external tools (e.g., search engines, calculators, service APIs, etc.) to execute the plans, and integrate the tools’ outputs to complete the end-to-end personalization tasks. Today, large language models are still being rapidly developed, whereas the application in personalization is largely unexplored. Therefore, we consider it to be right the time to review the challenges in personalization and the opportunities to address them with large language models. In particular, we dedicate this perspective paper to the discussion of the following aspects: the development and challenges for the existing personalization system, the newly emerged capabilities of large language models, and the potential ways of making use of large language models for personalization.

Predictive Autoscaling is used to forecast the workloads of servers and prepare the resources in advance to ensure service level objectives (SLOs) in dynamic cloud environments. However, in practice, its prediction task often suffers from performance degradation under abnormal traffics caused by external events (such as sales promotional activities and applications' re-configurations), for which a common solution is to re-train the model with data of a long historical period, but at the expense of high computational and storage costs. To better address this problem, we propose a replay-based continual learning method, i.e., Density-based Memory Selection and Hint-based Network Learning Model (DMSHM), using only a small part of the historical log to achieve accurate predictions. First, we discover the phenomenon of sample overlap when applying replay-based continual learning in prediction tasks. In order to surmount this challenge and effectively integrate new sample distribution, we propose a density-based sample selection strategy that utilizes kernel density estimation to calculate sample density as a reference to compute sample weight, and employs weight sampling to construct a new memory set. Then we implement hint-based network learning based on hint representation to optimize the parameters. Finally, we conduct experiments on public and industrial datasets to demonstrate that our proposed method outperforms state-of-the-art continual learning methods in terms of memory capacity and prediction accuracy. Furthermore, we demonstrate remarkable practicability of DMSHM in real industrial applications.

Neural-based multi-task learning (MTL) has gained significant improvement, and it has been successfully applied to recommendation system (RS). Recent deep MTL methods for RS (e.g. MMoE, PLE) focus on designing soft gating-based parameter-sharing networks that implicitly learn a generalized representation for each task. However, MTL methods may suffer from performance degeneration when dealing with conflicting tasks, as negative transfer effects can occur on the task-shared bottom representation. This can result in a reduced capacity for MTL methods to capture task-specific characteristics, ultimately impeding their effectiveness and hindering the ability to generalize well on all tasks. In this paper, we focus on the bottom representation learning of MTL in RS and propose the Deep Task-specific Bottom Representation Network (DTRN) to alleviate the negative transfer problem. DTRN obtains task-specific bottom representation explicitly by making each task have its own representation learning network in the bottom representation modeling stage. Specifically, it extracts the user's interests from multiple types of behavior sequences for each task through the parameter-efficient hypernetwork. To further obtain the dedicated representation for each task, DTRN refines the representation of each feature by employing a SENet-like network for each task. The two proposed modules can achieve the purpose of getting task-specific bottom representation to relieve tasks' mutual interference. Moreover, the proposed DTRN is flexible to combine with existing MTL methods. Experiments on one public dataset and one industrial dataset demonstrate the effectiveness of the proposed DTRN.

Neural Temporal Point Processes (TPPs) are the prevalent paradigm for modeling continuous-time event sequences, such as user activities on the web and financial transactions. In real world applications, the event data typically comes in a streaming manner, where the distribution of the patterns may shift over time. Under the privacy and memory constraints commonly seen in real scenarios, how to continuously monitor a TPP to learn the streaming event sequence is an important yet under-investigated problem. In this work, we approach this problem by adopting Continual Learning (CL), which aims to enable a model to continuously learn a sequence of tasks without catastrophic forgetting. While CL for event sequence is less well studied, we present a simple yet effective framework, PromptTPP, by integrating the base TPP with a continuous-time retrieval prompt pool. In our proposed framework, prompts are small learnable parameters, maintained in a memory space and jointly optimized with the base TPP so that the model is properly instructed to learn event streams arriving sequentially without buffering past examples or task-specific attributes. We formalize a novel and realistic experimental setup for modeling event streams, where PromptTPP consistently sets state-of-the-art performance across two real user behavior datasets.

Recent efforts on text erasing have shown promising results. However, existing methods require rich yet costly label annotations to obtain robust models, which limits their use for practical applications. To this end, we study an unsupervised scenario by proposing a novel Self-supervised Text Erasing (STE) framework that jointly learns to synthesize training images with erasure ground-truth and accurately erase texts in the real world. We first design a style-aware image synthesis function to generate synthetic images with diverse styled texts based on two synthetic mechanisms. To bridge the text style gap between the synthetic and real-world data, a policy network is constructed to control the synthetic mechanisms by picking style parameters with the guidance of two specifically designed rewards. The synthetic training images with ground-truth are then fed to train a coarse-to-fine erasing network. To produce better erasing outputs, a triplet erasure loss is designed to enforce the refinement stage to recover background textures. Moreover, we provide a new dataset (called PosterErase), which contains 60K high-resolution posters and is more challenging for the erasing task. The proposed method has been extensively evaluated with both PosterErase and the widely-used SCUT-Enstext dataset. Notably, on PosterErase, our method achieves 5.07 in terms of FID, with a relative improvement of 20.9% over existing supervised baselines.

Ad creatives are one of the prominent mediums for online e-commerce advertisements. Ad creatives with enjoyable visual appearance may increase the click-through rate (CTR) of products. Ad creatives are typically handcrafted by advertisers and then delivered to the advertising platforms for advertisement. In recent years, advertising platforms are capable of instantly compositing ad creatives with arbitrarily designated elements of each ingredient, so advertisers are only required to provide basic materials. While facilitating the advertisers, a great number of potential ad creatives can be composited, making it difficult to accurately estimate CTR for them given limited real-time feedback. To this end, we propose an Adaptive and Efficient ad creative Selection (AES) framework based on a tree structure. The tree structure on compositing ingredients enables dynamic programming for efficient ad creative selection on the basis of CTR. Due to limited feedback, the CTR estimator is usually of high variance. Exploration techniques based on Thompson sampling are widely used for reducing variances of the CTR estimator, alleviating feedback sparsity. Based on the tree structure, Thompson sampling is adapted with dynamic programming, leading to efficient exploration for potential ad creatives with the largest CTR. We finally evaluate the proposed algorithm on the synthetic dataset and the real-world dataset. The results show that our approach can outperform competing baselines in terms of convergence rate and overall CTR.

The factorization-based models have achieved great success in online advertisements and recommender systems due to the capability of efficiently modeling combinational features. These models encode feature interactions by the vector product between feature embedding. Despite the improvement of generalization, the memory consumption of these models grows significantly, because they usually take hundreds to thousands of large categorical features as input. Several existing works try to reduce the memory footprint by hashing, randomized embedding composition, and dimensionality search, but they suffer from either substantial performance degradation or limited memory compression. To this end, in this paper, we propose an extremely memory-efficient Factorization Machine (xLightFM), where each category embedding is composited with latent vectors selected from codebooks. Based on the characteristics of each categorical feature, we further propose to adapt the codebook size with the neural architecture search techniques for compositing the embedding of each categorical feature. This further pushes the limits of memory compression while incurring negligible degradation or even some improvements in prediction performance. We extensively evaluate the proposed algorithm with two real-world datasets. The results demonstrate that xLightFM can outperform the state-of-the-art lightweight factorization-based methods in terms of both prediction quality and memory footprint, and achieve more than 18x and 27x memory compression compared to the vanilla FM on these two datasets, respectively.

Advertising creatives are ubiquitous in E-commerce advertisements and aesthetic creatives may improve the click-through rate (CTR) of the products. Nowadays smart advertisement platforms provide the function of compositing creatives based on source materials provided by advertisers. Since a great number of creatives can be generated, it is difficult to accurately predict their CTR given a limited amount of feedback. Factorization machine (FM), which models inner product interaction between features, can be applied for the CTR prediction of creatives. However, interactions between creative elements may be more complex than the inner product, and the FM-estimated CTR may be of high variance due to limited feedback. To address these two issues, we propose an Automated Creative Optimization (AutoCO) framework to model complex interaction between creative elements and to balance between exploration and exploitation. Specifically, motivated by AutoML, we propose one-shot search algorithms for searching effective interaction functions between elements. We then develop stochastic variational inference to estimate the posterior distribution of parameters based on the reparameterization trick, and apply Thompson Sampling for efficiently exploring potentially better creatives. We evaluate the proposed method with both a synthetic dataset and two public datasets. The experimental results show our method can outperform competing baselines with respect to cumulative regret. The online A/B test shows our method leads to a 7% increase in CTR compared to the baseline.