arXiv subjects

Jie Fu

Publications and source records attributed to Jie Fu.

At least 73 records · Page 4Linked to original sources

arXiv2025

Thompson Sampling in Online RLHF with General Function Approximation

Reinforcement learning from human feedback (RLHF) has achieved great empirical success in aligning large language models (LLMs) with human preference, and it is of great importance to study the statistical efficiency of RLHF algorithms from a theoretical perspective. In this work, we consider the online RLHF setting where the preference data is revealed during the learning process and study action value function approximation. We design a model-free posterior sampling algorithm for online RLHF inspired by Thompson sampling and provide its theoretical guarantee. Specifically, we adopt Bellman eluder (BE) dimension as the complexity measure of the function class and establish $O(\sqrt{T})$ regret bound for the proposed algorithm with other multiplicative factor depending on the horizon, BE dimension and the $log$-bracketing number of the function class. Further, in the analysis, we first establish the concentration-type inequality of the squared Bellman error bound based on the maximum likelihood estimator (MLE) generalization bound, which plays the crucial rules in obtaining the eluder-type regret bound and may be of independent interest.

cs.LG↗

arXiv2025

Automata Learning of Preferences over Temporal Logic Formulas from Pairwise Comparisons

Many preference elicitation algorithms consider preference over propositional logic formulas or items with different attributes. In sequential decision making, a user's preference can be a preorder over possible outcomes, each of which is a temporal sequence of events. This paper considers a class of preference inference problems where the user's unknown preference is represented by a preorder over regular languages (sets of temporal sequences), referred to as temporal goals. Given a finite set of pairwise comparisons between finite words, the objective is to learn both the set of temporal goals and the preorder over these goals. We first show that a preference relation over temporal goals can be modeled by a Preference Deterministic Finite Automaton (PDFA), which is a deterministic finite automaton augmented with a preorder over acceptance conditions. The problem of preference inference reduces to learning the PDFA. This problem is shown to be computationally challenging, with the problem of determining whether there exists a PDFA of size smaller than a given integer $k$, consistent with the sample, being NP-Complete. We formalize the properties of characteristic samples and develop an algorithm that guarantees to learn, given a characteristic sample, the minimal PDFA equivalent to the true PDFA from which the sample is drawn. We present the method through a running example and provide detailed analysis using a robotic motion planning problem.

cs.AI↗

arXiv2025

Generating Symbolic World Models via Test-time Scaling of Large Language Models

Solving complex planning problems requires Large Language Models (LLMs) to explicitly model the state transition to avoid rule violations, comply with constraints, and ensure optimality-a task hindered by the inherent ambiguity of natural language. To overcome such ambiguity, Planning Domain Definition Language (PDDL) is leveraged as a planning abstraction that enables precise and formal state descriptions. With PDDL, we can generate a symbolic world model where classic searching algorithms, such as A*, can be seamlessly applied to find optimal plans. However, directly generating PDDL domains with current LLMs remains an open challenge due to the lack of PDDL training data. To address this challenge, we propose to scale up the test-time computation of LLMs to enhance their PDDL reasoning capabilities, thereby enabling the generation of high-quality PDDL domains. Specifically, we introduce a simple yet effective algorithm, which first employs a Best-of-N sampling approach to improve the quality of the initial solution and then refines the solution in a fine-grained manner with verbalized machine learning. Our method outperforms o1-mini by a considerable margin in the generation of PDDL domains, achieving over 50\% success rate on two tasks (i.e., generating PDDL domains from natural language description or PDDL problems). This is done without requiring additional training. By taking advantage of PDDL as state abstraction, our method is able to outperform current state-of-the-art methods on almost all competition-level planning tasks.

cs.AI↗

arXiv2025

MAP: Low-compute Model Merging with Amortized Pareto Fronts via Quadratic Approximation

Model merging has emerged as an effective approach to combine multiple single-task models into a multitask model. This process typically involves computing a weighted average of the model parameters without any additional training. Existing model-merging methods focus on enhancing average task accuracy. However, interference and conflicts between the objectives of different tasks can lead to trade-offs during the merging process. In real-world applications, a set of solutions with various trade-offs can be more informative, helping practitioners make decisions based on diverse preferences. In this paper, we introduce a novel and low-compute algorithm, Model Merging with Amortized Pareto Front (MAP). MAP efficiently identifies a Pareto set of scaling coefficients for merging multiple models, reflecting the trade-offs involved. It amortizes the substantial computational cost of evaluations needed to estimate the Pareto front by using quadratic approximation surrogate models derived from a pre-selected set of scaling coefficients. Experimental results on vision and natural language processing tasks demonstrate that MAP can accurately identify the Pareto front, providing practitioners with flexible solutions to balance competing task objectives. We also introduce Bayesian MAP for scenarios with a relatively low number of tasks and Nested MAP for situations with a high number of tasks, further reducing the computational cost of evaluation.

cs.LG↗

arXiv2025

NeuralGrok: Accelerate Grokking by Neural Gradient Transformation

Grokking is proposed and widely studied as an intricate phenomenon in which generalization is achieved after a long-lasting period of overfitting. In this work, we propose NeuralGrok, a novel gradient-based approach that learns an optimal gradient transformation to accelerate the generalization of transformers in arithmetic tasks. Specifically, NeuralGrok trains an auxiliary module (e.g., an MLP block) in conjunction with the base model. This module dynamically modulates the influence of individual gradient components based on their contribution to generalization, guided by a bilevel optimization algorithm. Our extensive experiments demonstrate that NeuralGrok significantly accelerates generalization, particularly in challenging arithmetic tasks. We also show that NeuralGrok promotes a more stable training paradigm, constantly reducing the model's complexity, while traditional regularization methods, such as weight decay, can introduce substantial instability and impede generalization. We further investigate the intrinsic model complexity leveraging a novel Absolute Gradient Entropy (AGE) metric, which explains that NeuralGrok effectively facilitates generalization by reducing the model complexity. We offer valuable insights on the grokking phenomenon of Transformer models, which encourages a deeper understanding of the fundamental principles governing generalization ability.

cs.LG↗

arXiv2025

VCR: A Task for Pixel-Level Complex Reasoning in Vision Language Models via Restoring Occluded Text

We introduce Visual Caption Restoration (VCR), a novel vision-language task that challenges models to accurately restore partially obscured texts using pixel-level hints within images. This task stems from the observation that text embedded in images is intrinsically different from common visual elements and natural language due to the need to align the modalities of vision, text, and text embedded in images. While numerous works have integrated text embedded in images into visual question-answering tasks, approaches to these tasks generally rely on optical character recognition or masked language modeling, thus reducing the task to mainly text-based processing. However, text-based processing becomes ineffective in VCR as accurate text restoration depends on the combined information from provided images, context, and subtle cues from the tiny exposed areas of masked texts. We develop a pipeline to generate synthetic images for the VCR task using image-caption pairs, with adjustable caption visibility to control the task difficulty. With this pipeline, we construct a dataset for VCR called VCR-Wiki using images with captions from Wikipedia, comprising 2.11M English and 346K Chinese entities in both easy and hard split variants. Our results reveal that current vision language models significantly lag behind human performance in the VCR task, and merely fine-tuning the models on our dataset does not lead to notable improvements. We release VCR-Wiki and the data construction code to facilitate future research.

cs.CV↗

arXiv2025

Integrated Control and Active Perception in POMDPs for Temporal Logic Tasks and Information Acquisition

This paper studies the synthesis of a joint control and active perception policy for a stochastic system modeled as a partially observable Markov decision process (POMDP), subject to temporal logic specifications. The POMDP actions influence both system dynamics (control) and the emission function (perception). Beyond task completion, the planner seeks to maximize information gain about certain temporal events (the secret) through coordinated perception and control. To enable active information acquisition, we introduce minimizing the Shannon conditional entropy of the secret as a planning objective, alongside maximizing the probability of satisfying the temporal logic formula within a finite horizon. Using a variant of observable operators in hidden Markov models (HMMs) and POMDPs, we establish key properties of the conditional entropy gradient with respect to policy parameters. These properties facilitate efficient policy gradient computation. We validate our approach through graph-based examples, inspired by common security applications with UAV surveillance.

eess.SY↗

arXiv2025

CodeEditorBench: Evaluating Code Editing Capability of Large Language Models

Large Language Models (LLMs) for code are rapidly evolving, with code editing emerging as a critical capability. We introduce CodeEditorBench, an evaluation framework designed to rigorously assess the performance of LLMs in code editing tasks, including debugging, translating, polishing, and requirement switching. Unlike existing benchmarks focusing solely on code generation, CodeEditorBench emphasizes real-world scenarios and practical aspects of software development. We curate diverse coding challenges and scenarios from five sources, covering various programming languages, complexity levels, and editing tasks. Evaluation of 19 LLMs reveals that closed-source models (particularly Gemini-Ultra and GPT-4), outperform open-source models in CodeEditorBench, highlighting differences in model performance based on problem types and prompt sensitivities. CodeEditorBench aims to catalyze advancements in LLMs by providing a robust platform for assessing code editing capabilities. We will release all prompts and datasets to enable the community to expand the dataset and benchmark emerging LLMs. By introducing CodeEditorBench, we contribute to the advancement of LLMs in code editing and provide a valuable resource for researchers and practitioners.

cs.SE↗

arXiv2025

AI Alignment: A Comprehensive Survey

AI alignment aims to make AI systems behave in line with human intentions and values. As AI systems grow more capable, so do risks from misalignment. To provide a comprehensive and up-to-date overview of the alignment field, in this survey, we delve into the core concepts, methodology, and practice of alignment. First, we identify four principles as the key objectives of AI alignment: Robustness, Interpretability, Controllability, and Ethicality (RICE). Guided by these four principles, we outline the landscape of current alignment research and decompose them into two key components: forward alignment and backward alignment. The former aims to make AI systems aligned via alignment training, while the latter aims to gain evidence about the systems' alignment and govern them appropriately to avoid exacerbating misalignment risks. On forward alignment, we discuss techniques for learning from feedback and learning under distribution shift. On backward alignment, we discuss assurance techniques and governance practices. We also release and continually update the website (www.alignmentsurvey.com) which features tutorials, collections of papers, blog posts, and other resources.

cs.AI↗

arXiv2025

OpenCoder: The Open Cookbook for Top-Tier Code Large Language Models

Large language models (LLMs) for code have become indispensable in various domains, including code generation, reasoning tasks and agent systems. While open-access code LLMs are increasingly approaching the performance levels of proprietary models, high-quality code LLMs suitable for rigorous scientific investigation, particularly those with reproducible data processing pipelines and transparent training protocols, remain limited. The scarcity is due to various challenges, including resource constraints, ethical considerations, and the competitive advantages of keeping models advanced. To address the gap, we introduce OpenCoder, a top-tier code LLM that not only achieves performance comparable to leading models but also serves as an "open cookbook" for the research community. Unlike most prior efforts, we release not only model weights and inference code, but also the reproducible training data, complete data processing pipeline, rigorous experimental ablation results, and detailed training protocols for open scientific research. Through this comprehensive release, we identify the key ingredients for building a top-tier code LLM: (1) code optimized heuristic rules for data cleaning and methods for data deduplication, (2) recall of text corpus related to code and (3) high-quality synthetic data in both annealing and supervised fine-tuning stages. By offering this level of openness, we aim to broaden access to all aspects of a top-tier code LLM, with OpenCoder serving as both a powerful model and an open foundation to accelerate research, and enable reproducible advancements in code AI.

cs.CL↗

arXiv2025

Layerwise Recurrent Router for Mixture-of-Experts

The scaling of large language models (LLMs) has revolutionized their capabilities in various tasks, yet this growth must be matched with efficient computational strategies. The Mixture-of-Experts (MoE) architecture stands out for its ability to scale model size without significantly increasing training costs. Despite their advantages, current MoE models often display parameter inefficiency. For instance, a pre-trained MoE-based LLM with 52 billion parameters might perform comparably to a standard model with 6.7 billion parameters. Being a crucial part of MoE, current routers in different layers independently assign tokens without leveraging historical routing information, potentially leading to suboptimal token-expert combinations and the parameter inefficiency problem. To alleviate this issue, we introduce the Layerwise Recurrent Router for Mixture-of-Experts (RMoE). RMoE leverages a Gated Recurrent Unit (GRU) to establish dependencies between routing decisions across consecutive layers. Such layerwise recurrence can be efficiently parallelly computed for input tokens and introduces negotiable costs. Our extensive empirical evaluations demonstrate that RMoE-based language models consistently outperform a spectrum of baseline models. Furthermore, RMoE integrates a novel computation stage orthogonal to existing methods, allowing seamless compatibility with other MoE architectures. Our analyses attribute RMoE's gains to its effective cross-layer information sharing, which also improves expert selection and diversity. Our code is at https://github.com/qiuzh20/RMoE .

cs.CL↗

arXiv2025

Learning from Failures in Multi-Attempt Reinforcement Learning

Recent advancements in reinforcement learning (RL) for large language models (LLMs), exemplified by DeepSeek R1, have shown that even a simple question-answering task can substantially improve an LLM's reasoning capabilities. In this work, we extend this approach by modifying the task into a multi-attempt setting. Instead of generating a single response per question, the model is given multiple attempts, with feedback provided after incorrect responses. The multi-attempt task encourages the model to refine its previous attempts and improve search efficiency. Experimental results show that even a small LLM trained on a multi-attempt task achieves significantly higher accuracy when evaluated with more attempts, improving from 45.6% with 1 attempt to 52.5% with 2 attempts on the math benchmark. In contrast, the same LLM trained on a standard single-turn task exhibits only a marginal improvement, increasing from 42.3% to 43.2% when given more attempts during evaluation. The results indicate that, compared to the standard single-turn task, an LLM trained on a multi-attempt task achieves slightly better performance on math benchmarks while also learning to refine its responses more effectively based on user feedback. Full code is available at https://github.com/DualityRL/multi-attempt

cs.CL↗

arXiv2025

PVU: Design and Implementation of a Posit Vector Arithmetic Unit (PVU) for Enhanced Floating-Point Computing in Edge and AI Applications

With the rapid development of edge computing, artificial intelligence and other fields, the accuracy and efficiency of floating-point computing have become increasingly crucial. However, the traditional IEEE 754 floating-point system faces bottlenecks in energy consumption and computing accuracy, which have become major constraints. To address this issue, the Posit digital system characterized by adaptive accuracy, broader dynamic range and low hardware consumption has been put forward. Despite its widespread adoption, the existing research mainly concentrates on scalar computation, which is insufficient to meet the requirements of large-scale parallel data processing. This paper proposes, for the first time, a Posit Vector Arithmetic Unit (PVU) designed using the Chisel language. It supports vector operations such as addition, subtraction, multiplication, division, and dot product, thereby overcoming the limitations of traditional scalar designs and integrating the RISC-V instruction extension. The contributions of this paper include the efficient implementation of the vector arithmetic unit, the parametric and modular hardware design as well as the verification of the practical application of the positive digital system. This paper extracts the quantized data of the first convolutional layer for verification. Experiments indicate that the accuracy rate of the division operation is 95.84\%, and the accuracy rate of the remaining operations is 100\%. Moreover, the PVU is implemented with only 65,407 LUTs. Therefore, PVU has great potential as a new-generation floating-point computing platform in various fields.

cs.DC↗

arXiv2025

T3D: Advancing 3D Medical Vision-Language Pre-training by Learning Multi-View Visual Consistency

While 3D visual self-supervised learning (vSSL) shows promising results in capturing visual representations, it overlooks the clinical knowledge from radiology reports. Meanwhile, 3D medical vision-language pre-training (MedVLP) remains underexplored due to the lack of a large-scale, publicly available 3D medical image-report dataset. To bridge this gap, we introduce **CT-3DVLP**, the first and largest **public** 3D volume-report dataset, establishing a comprehensive benchmark for 3D MedVLP research. Meanwhile, we propose the **T3D** framework, which enhances 3D MedVLP beyond naive CLIP-style alignment that directly pairs volumes with reports but neglects local visual representations. Instead, we introduce **Text-informed Multi-view Alignment (TMA)**, a novel approach that clusters volumetric data while enforcing consistency across different views of the same volume-report pair. TMA integrates textual features into fine-grained visual representations, ensuring contextual coherence across views. We evaluate T3D across multiple downstream tasks in both unimodal and cross-modal settings, including zero-shot and fine-tuned classification, cross-modal retrieval, report generation, and semantic segmentation. Our results show that T3D consistently outperforms existing vSSL and multimodal methods, demonstrating superior zero-shot and fine-tuning capabilities and setting a new benchmark for 3D medical image understanding.

cs.CV↗

arXiv2025

Synthesis of Dynamic Masks for Information-Theoretic Opacity in Stochastic Systems

In this work, we investigate the synthesis of dynamic information releasing mechanisms, referred to as ''masks'', to minimize information leakage from a stochastic system to an external observer. Specifically, for a stochastic system, an observer aims to infer whether the final state of the system trajectory belongs to a set of secret states. The dynamic mask seeks to regulate sensor information in order to maximize the observer's uncertainty about the final state, a property known as final-state opacity. While existing supervisory control literature on dynamic masks primarily addresses qualitative opacity, we propose quantifying opacity in stochastic systems by conditional entropy, which is a measure of information leakage in information security. We then formulate a constrained optimization problem to synthesize a dynamic mask that maximizes final-state opacity under a total cost constraint on masking. To solve this constrained optimal dynamic mask synthesis problem, we develop a novel primal-dual policy gradient method. Additionally, we present a technique for computing the gradient of conditional entropy with respect to the masking policy parameters, leveraging observable operators in hidden Markov models. To demonstrate the effectiveness of our approach, we apply our method to an illustrative example and a stochastic grid world scenario, showing how our algorithm optimally enforces final-state opacity under cost constraints.

eess.SY↗

arXiv2025

Active Inference through Incentive Design in Markov Decision Processes

We present a method for active inference with partial observations in stochastic systems through incentive design, also known as the leader-follower game. Consider a leader agent who aims to infer a follower agent's type given a finite set of possible types. Different types of followers differ in either the dynamical model, the reward function, or both. We assume the leader can partially observe a follower's behavior in the stochastic system modeled as a Markov decision process, in which the follower takes an optimal policy to maximize a total reward. To improve inference accuracy and efficiency, the leader can offer side payments (incentives) to the followers such that different types of them, under the incentive design, can exhibit diverging behaviors that facilitate the leader's inference task. We show the problem of active inference through incentive design can be formulated as a special class of leader-follower games, where the leader's objective is to balance the information gain and cost of incentive design. The information gain is measured by the entropy of the estimated follower's type given partial observations. Furthermore, we demonstrate that this problem can be solved by reducing a single-level optimization through softmax temporal consistency between followers' policies and value functions. This reduction allows us to develop an efficient gradient-based algorithm. We utilize observable operators in the hidden Markov model (HMM) to compute the necessary gradients and demonstrate the effectiveness of our approach through experiments in stochastic grid world environments.

eess.SY↗

arXiv2025

Sequential Decision Making in Stochastic Games with Incomplete Preferences over Temporal Objectives

Ensuring that AI systems make strategic decisions aligned with the specified preferences in adversarial sequential interactions is a critical challenge for developing trustworthy AI systems, especially when the environment is stochastic and players' incomplete preferences leave some outcomes unranked. We study the problem of synthesizing preference-satisfying strategies in two-player stochastic games on graphs where players have opposite (possibly incomplete) preferences over a set of temporal goals. We represent these goals using linear temporal logic over finite traces (LTLf), which enables modeling the nuances of human preferences where temporal goals need not be mutually exclusive and comparison between some goals may be unspecified. We introduce a solution concept of non-dominated almost-sure winning, which guarantees to achieve a most preferred outcome aligned with specified preferences while maintaining robustness against the adversarial behaviors of the opponent. Our results show that strategy profiles based on this concept are Nash equilibria in the game where players are risk-averse, thus providing a practical framework for evaluating and ensuring stable, preference-aligned outcomes in the game. Using a drone delivery example, we demonstrate that our contributions offer valuable insights not only for synthesizing rational behavior under incomplete preferences but also for designing games that motivate the desired behavior from the players in adversarial conditions.

cs.GT↗

arXiv2025

OpenCodeInterpreter: Integrating Code Generation with Execution and Refinement

The introduction of large language models has significantly advanced code generation. However, open-source models often lack the execution capabilities and iterative refinement of advanced systems like the GPT-4 Code Interpreter. To address this, we introduce OpenCodeInterpreter, a family of open-source code systems designed for generating, executing, and iteratively refining code. Supported by Code-Feedback, a dataset featuring 68K multi-turn interactions, OpenCodeInterpreter integrates execution and human feedback for dynamic code refinement. Our comprehensive evaluation of OpenCodeInterpreter across key benchmarks such as HumanEval, MBPP, and their enhanced versions from EvalPlus reveals its exceptional performance. Notably, OpenCodeInterpreter-33B achieves an accuracy of 83.2 (76.4) on the average (and plus versions) of HumanEval and MBPP, closely rivaling GPT-4's 84.2 (76.2) and further elevates to 91.6 (84.6) with synthesized human feedback from GPT-4. OpenCodeInterpreter brings the gap between open-source code generation models and proprietary systems like GPT-4 Code Interpreter.

cs.SE↗