arXiv subjects

Wei Ye

Publications and source records attributed to Wei Ye.

At least 55 records · Page 3Linked to original sources

arXiv2026

Instruction Data Selection via Answer Divergence

Instruction tuning relies on large instruction-response corpora whose quality and composition strongly affect downstream performance. We propose Answer Divergence-Guided Selection (ADG), which selects instruction data based on the geometric structure of multi-sample outputs. ADG draws several high-temperature generations per instruction, maps responses into an embedding space, and computes an output divergence score that jointly encodes dispersion magnitude and shape anisotropy. High scores correspond to instructions whose answers are both far apart and multi-modal, rather than clustered paraphrases along a single direction. Across two backbones and three public instruction pools, fine-tuning on only 10K ADG-selected examples consistently outperforms strong selectors on six benchmarks spanning reasoning, knowledge, and coding. Analyses further show that both dispersion magnitude and shape anisotropy are necessary, supporting answer divergence as a practical signal for instruction data selection. Code and appendix are included in the supplementary materials.

cs.CL↗

arXiv2026

Retrieval as Generation: A Unified Framework with Self-Triggered Information Planning

We revisit retrieval-augmented generation (RAG) by embedding retrieval control directly into generation. Instead of treating retrieval as an external intervention, we express retrieval decisions within token-level decoding, enabling end-to-end coordination without additional controllers or classifiers. Under the paradigm of Retrieval as Generation, we propose \textbf{GRIP} (\textbf{G}eneration-guided \textbf{R}etrieval with \textbf{I}nformation \textbf{P}lanning), a unified framework in which the model regulates retrieval behavior through control-token emission. Central to GRIP is \textit{Self-Triggered Information Planning}, which allows the model to decide when to retrieve, how to reformulate queries, and when to terminate, all within a single autoregressive trajectory. This design tightly couples retrieval and reasoning and supports dynamic multi-step inference with on-the-fly evidence integration. To supervise these behaviors, we construct a structured training set covering answerable, partially answerable, and multi-hop queries, each aligned with specific token patterns. Experiments on five QA benchmarks show that GRIP surpasses strong RAG baselines and is competitive with GPT-4o while using substantially fewer parameters.

cs.CL↗

arXiv2026

A Survey of Reinforcement Learning for Large Language Models under Data Scarcity: Challenges and Solutions

Reinforcement learning (RL) has emerged as a powerful post-training paradigm for enhancing the reasoning capabilities of large language models (LLMs). However, reinforcement learning for LLMs faces substantial data scarcity challenges, including the limited availability of high-quality external supervision and the constrained volume of model-generated experience. These limitations make data-efficient reinforcement learning a critical research direction. In this survey, we present the first systematic review of reinforcement learning for LLMs under data scarcity. We propose a bottom-up hierarchical framework built around three complementary perspectives: the data-centric perspective, the training-centric perspective, and the framework-centric perspective. We develop a taxonomy of existing methods, summarize representative approaches in each category, and analyze their strengths and limitations. Our taxonomy aims to provide a clear conceptual foundation for understanding the design space of data-efficient RL for LLMs and to guide researchers working in this emerging area. We hope this survey offers a comprehensive roadmap for future research and inspires new directions toward more efficient and scalable reinforcement learning post-training for LLMs.

cs.LG↗

arXiv2026

Learning from Contrasts: Synthesizing Reasoning Paths from Diverse Search Trajectories

Monte Carlo Tree Search (MCTS) has been widely used for automated reasoning data exploration, but current supervision extraction methods remain inefficient. Standard approaches retain only the single highest-reward trajectory, discarding the comparative signals present in the many explored paths. Here we introduce \textbf{Contrastive Reasoning Path Synthesis (CRPS)}, a framework that transforms supervision extraction from a filtering process into a synthesis procedure. CRPS uses a structured reflective process to analyze the differences between high- and low-quality search trajectories, extracting explicit information about strategic pivots and local failure modes. These insights guide the synthesis of reasoning chains that incorporate success patterns while avoiding identified pitfalls. We show empirically that models fine-tuned on just 60K CRPS-synthesized examples match or exceed the performance of baselines trained on 590K examples derived from standard rejection sampling, a 20$\times$ reduction in dataset size. Furthermore, CRPS improves generalization on out-of-domain benchmarks, demonstrating that learning from the contrast between success and failure produces more transferable reasoning capabilities than learning from success alone.

cs.AI↗

arXiv2026

Mitigating Visual Context Degradation in Large Multimodal Models: A Training-Free Decoupled Agentic Framework

With the continuous expansion of Large Language Models (LLMs) and advances in reinforcement learning, LLMs have demonstrated exceptional reasoning capabilities, enabling them to address a wide range of complex problems. Inspired by these achievements, researchers have extended related techniques to Large Multimodal Models (LMMs). However, a critical limitation has emerged, reflected in the progressive loss of visual grounding. As the reasoning chain grows longer, LMMs tend to rely increasingly on the textual information generated in earlier steps, while the initially extracted visual information is rarely revisited or incorporated. This phenomenon often causes the reasoning process to drift away from the actual image content, resulting in visually implausible or even erroneous conclusions. To overcome this fundamental limitation, we propose a novel, training-free agentic paradigm that Decouples cognitive Reasoning from visual Perception (DRP). In this framework, a powerful LLM serves as a strategic Reasoner, orchestrating the inference process by explicitly querying an LMM-acting as a dedicated Observer-to retrieve fine-grained visual details on demand. This approach is lightweight, model-agnostic, and plug-and-play, necessitating no additional training or architectural modifications. Extensive experiments demonstrate our framework DRP's efficacy in regulating the visual reasoning trajectory, significantly mitigating reasoning drift, and enforcing robust visual grounding. Notably, on the MathVision benchmark, the integration of Qwen2.5-VL-7B and Qwen3-32B achieves an accuracy of 47.2\%, outperforming GPT-4o's 40.6\%. These findings underscore the potential of our approach to enhance multimodal reasoning reliability without the need for costly retraining. Our code is publicly available at https://github.com/hongruijia/DRP.

cs.CV↗

arXiv2026

An Empirical Study on Influence-Based Pretraining Data Selection for Code Large Language Models

Recent advancements in code large language models (Code-LLMs) have demonstrated remarkable capabilities in resolving programming related tasks. Meanwhile, researchers have recognized that the quality of pre-training data is crucial for improving LLM performance. However, most of the existing research on pre-training data filtering has focused on general datasets, and little attention for programming datasets. In this paper, we aim to address this gap by exploring the effectiveness of a widely used general data filtering technique, i.e., data-influence-score filtering, within the context of programming-related datasets. To this end, we first introduce a method for calculating data-influence-score for generative programming tasks which involves transforming a variety of downstream coding tasks into validation sets and using the models loss on these sets as a performance metric. Next, we pre-train a Code-LLMs with 1 billion parameters from scratch on a dataset of 100 billion code tokens. Based on it, we conduct an extensive empirical study to evaluate the effectiveness of data-influence-score filtering methods. Specifically, we examine how well this technique improves model performance, investigate how the characteristics of beneficial training data vary across different training stages and programming tasks, and assess the feasibility of prediction-based data-influence-score filtering method. Our findings show that data-influence-score filtering based on validation-set-loss can enhance models programming performance. Moreover, we observe that the criteria of beneficial training data differ significantly across various downstream programming tasks.

cs.SE↗

arXiv2026

GALA: Multimodal Graph Alignment for Bug Localization in Automated Program Repair

Large Language Model (LLM)-based Automated Program Repair (APR) has shown strong potential on textual benchmarks, yet struggles in multimodal scenarios where bugs are reported with GUI screenshots. Existing methods typically convert images into plain text, which discards critical spatial relationships and causes a severe disconnect between visual observations and code components, leading localization to degrade into imprecise keyword matching. To bridge this gap, we propose GALA (Graph Alignment for Localization in APR), a framework that shifts multimodal APR from implicit semantic guessing to explicit structural reasoning. GALA operates in four stages: it first constructs an Image UI Graph to capture visual elements and their structural relationships; then performs file-level alignment by cross-referencing this UI graph with repository-level structures (e.g., file references) to locate candidate files; next conducts function-level alignment by reasoning over fine-grained code dependencies (e.g., call graphs) to precisely ground visual elements to corresponding code components; and finally performs patch generation within the grounded code context based on the aligned files and functions. By systematically enforcing both semantic and relational consistency across modalities, GALA establishes a highly accurate visual-to-code mapping. Evaluations on the SWE-bench Multimodal benchmark demonstrate that GALA achieves state-of-the-art performance, highlighting the effectiveness of hierarchical structural alignment.

cs.SE↗

arXiv2026

QKVQA: Question-Focused Filtering for Knowledge-based VQA

Visual Question Answering (VQA) is the task of answering questions based on image content. Building upon this, Knowledge-Based VQA (KB-VQA) requires models to answer questions that depend on external knowledge beyond the visual content of an image. In such settings, effective knowledge filtering is essential for achieving high question answering accuracy. Typical filtering methods suffer from two issues: they fail to focus on parts relevant to the question during candidate section encoding, and they use similarity metrics to locate a section from a single article, resulting in information limitation. To address these issues, this paper proposes a question-focused, cross-article filtering method. Specifically, we design a trainable Question-Focused Filter (QFF) and a Chunk-based Dynamic Cross-Article Selection module (CDA). This approach maintains inference time comparable to the optimal method with the shorter context length, efficiently obtaining high-quality filtered knowledge. The accuracy outperforms current state-of-the-art methods by 3.2 and 2.2 percentage points on Encyclopedic-VQA and InfoSeek, respectively. The code is publicly available at: https://github.com/leaffeall/QKVQA.

cs.IR↗

arXiv2026

Squeezing-Enhanced Two-Phase Estimation with N-Particle W-type States

We investigate the simultaneous estimation of two optical phases in a three-mode interferometer assisted by optical parametric amplification (OPA). By employing the normally ordered characteristic-function formalism, we analytically obtain all photon-number moments of the output quantum state, enabling an explicit evaluation of the quantum Fisher information matrix for multiparameter phase estimation. In the lossless scenario, we show that uniformly applied OPA significantly enhances the attainable precision beyond that of an unamplified interferometer. By analyzing the second-order correlation functions, we demonstrate that this enhancement originates from the amplification of intra-mode photon-number correlations, rather than from inter-mode correlations. We further extend our analysis to realistic interferometers with photon loss using a purification-based variational approach. Although loss degrades the achievable precision, the OPA-assisted scheme retains a clear advantage for moderate loss, indicating a degree of robustness against dissipation. Our results clarify the physical mechanism underlying OPA-enhanced multiparameter quantum metrology and provide guidelines for optimizing phase estimation protocols in realistic noisy environments.

quant-ph↗

arXiv2026

SteerRM: Debiasing Reward Models via Sparse Autoencoders

Reward models (RMs) are critical components of alignment pipelines, yet they exhibit biases toward superficial stylistic cues, preferring better-presented responses over semantically superior ones. Existing debiasing methods typically require retraining or architectural modifications, while direct activation suppression degrades performance due to representation entanglement. We propose SteerRM, the first training-free method for debiasing reward models using Sparse Autoencoder (SAE)-based interventions. SteerRM isolates stylistic effects using contrastive paired responses, identifies bias-related SAE features with a strength-stability criterion, and suppresses them at inference time. Across six reward models on RM-Bench, SteerRM improves Hard-split accuracy by 7.3 points on average while preserving overall performance. Results on a Gemma-based reward model and a controlled non-format bias further suggest generalization across RM architectures and bias types. We further find that format-related features are concentrated in shallow layers and transfer across models, revealing shared architecture-level bias encoding patterns. These results show that SAE-based interventions can mitigate reward-model biases without retraining, providing a practical and interpretable solution for alignment pipelines.

cs.CL↗

arXiv2026

Bitcoin Price Prediction using Machine Learning and Combinatorial Fusion Analysis

In this work, we propose to apply a new model fusion and learning paradigm, known as Combinatorial Fusion Analysis (CFA), to the field of Bitcoin price prediction. Price prediction of financial product has always been a big topic in finance, as the successful prediction of the price can yield significant profit. Every machine learning model has its own strength and weakness, which hinders progress toward robustness. CFA has been used to enhance models by leveraging rank-score characteristic (RSC) function and cognitive diversity in the combination of a moderate set of diverse and relatively well-performed models. Our method utilizes both score and rank combinations as well as other weighted combination techniques. Key metrics such as RMSE and MAPE are used to evaluate our methodology performance. Our proposal presents a notable MAPE performance of 0.19\%. The proposed method greatly improves upon individual model performance, as well as outperforms other Bitcoin price prediction models.

q-fin.ST↗

arXiv2026

Revealing Combinatorial Reasoning of GNNs via Graph Concept Bottleneck Layer

Despite their success in various domains, the growing dependence on GNNs raises a critical concern about the nature of the combinatorial reasoning underlying their predictions, which is often hidden within their black-box architectures. Addressing this challenge requires understanding how GNNs translate topological patterns into logical rules. However, current works only uncover the hard logical rules over graph concepts, which cannot quantify the contribution of each concept to prediction. Moreover, they are post-hoc interpretable methods that generate explanations after model training and may not accurately reflect the true combinatorial reasoning of GNNs, since they approximate it with a surrogate. In this work, we develop a graph concept bottleneck layer that can be integrated into any GNN architectures to guide them to predict the selected discriminative global graph concepts. The predicted concept scores are further projected to class labels by a sparse linear layer. It enforces the combinatorial reasoning of GNNs' predictions to fit the soft logical rule over graph concepts and thus can quantify the contribution of each concept. To further improve the quality of the concept bottleneck, we treat concepts as "graph words" and graphs as "graph sentences", and leverage language models to learn graph concept embeddings. Extensive experiments on multiple datasets show that our method GCBMs achieve state-of-the-art performance both in classification and interpretability.

cs.LG↗

arXiv2026

Two-Stream Interactive Joint Learning of Scene Parsing and Geometric Vision Tasks

Inspired by the human visual system, which operates on two parallel yet interactive streams for contextual and spatial understanding, this article presents Two Interactive Streams (TwInS), a novel bio-inspired joint learning framework capable of simultaneously performing scene parsing and geometric vision tasks. TwInS adopts a unified, general-purpose architecture in which multi-level contextual features from the scene parsing stream are infused into the geometric vision stream to guide its iterative refinement. In the reverse direction, decoded geometric features are projected into the contextual feature space for selective heterogeneous feature fusion via a novel cross-task adapter, which leverages rich cross-view geometric cues to enhance scene parsing. To eliminate the dependence on costly human-annotated correspondence ground truth, TwInS is further equipped with a tailored semi-supervised training strategy, which unleashes the potential of large-scale multi-view data and enables continuous self-evolution without requiring ground-truth correspondences. Extensive experiments conducted on three public datasets validate the effectiveness of TwInS's core components and demonstrate its superior performance over existing state-of-the-art approaches. The source code will be made publicly available upon publication.

cs.CV↗

arXiv2026

OPE: Overcoming Information Saturation in Parallel Thinking via Outline-Guided Path Exploration

Parallel thinking has emerged as a new paradigm for large reasoning models (LRMs) in tackling complex problems. Recent methods leverage Reinforcement Learning (RL) to enhance parallel thinking, aiming to address the limitations in computational resources and effectiveness encountered with supervised fine-tuning. However, most existing studies primarily focus on optimizing the aggregation phase, with limited attention to the path exploration stage. In this paper, we theoretically analyze the optimization of parallel thinking under the Reinforcement Learning with Verifiable Rewards (RLVR) setting, and identify that the mutual information bottleneck among exploration paths fundamentally restricts overall performance. To address this, we propose Outline-Guided Path Exploration (OPE), which explicitly partitions the solution space by generating diverse reasoning outlines prior to parallel path reasoning, thereby reducing information redundancy and improving the diversity of information captured across exploration paths. We implement OPE with an iterative RL strategy that optimizes outline planning and outline-guided reasoning independently. Extensive experiments across multiple challenging mathematical benchmarks demonstrate that OPE effectively improves reasoning performance in different aggregation strategies, enabling LRMs to more reliably discover correct solutions.

cs.AI↗

arXiv2026

Rebenchmarking Unsupervised Monocular 3D Occupancy Prediction

Inferring the 3D structure from a single image, particularly in occluded regions, remains a fundamental yet unsolved challenge in vision-centric autonomous driving. Existing unsupervised approaches typically train a neural radiance field and treat the network outputs as occupancy probabilities during evaluation, overlooking the inconsistency between training and evaluation protocols. Moreover, the prevalent use of 2D ground truth fails to reveal the inherent ambiguity in occluded regions caused by insufficient geometric constraints. To address these issues, this paper presents a reformulated benchmark for unsupervised monocular 3D occupancy prediction. We first interpret the variables involved in the volume rendering process and identify the most physically consistent representation of the occupancy probability. Building on these analyses, we improve existing evaluation protocols by aligning the newly identified representation with voxel-wise 3D occupancy ground truth, thereby enabling unsupervised methods to be evaluated in a manner consistent with that of supervised approaches. Additionally, to impose explicit constraints in occluded regions, we introduce an occlusion-aware polarization mechanism that incorporates multi-view visual cues to enhance discrimination between occupied and free spaces in these regions. Extensive experiments demonstrate that our approach not only significantly outperforms existing unsupervised approaches but also matches the performance of supervised ones. Our source code and evaluation protocol will be made available upon publication.

cs.CV↗

arXiv2026

MPF-Net: Exposing High-Fidelity AI-Generated Video Forgeries via Hierarchical Manifold Deviation and Micro-Temporal Fluctuations

With the rapid advancement of video generation models such as Veo and Wan, the visual quality of synthetic content has reached a level where macro-level semantic errors and temporal inconsistencies are no longer prominent. However, this does not imply that the distinction between real and cutting-edge high-fidelity fake is untraceable. We argue that AI-generated videos are essentially products of a manifold-fitting process rather than a physical recording. Consequently, the pixel composition logic of consecutive adjacent frames residual in AI videos exhibits a structured and homogenous characteristic. We term this phenomenon `Manifold Projection Fluctuations' (MPF). Driven by this insight, we propose a hierarchical dual-path framework that operates as a sequential filtering process. The first, the Static Manifold Deviation Branch, leverages the refined perceptual boundaries of Large-Scale Vision Foundation Models (VFMs) to capture residual spatial anomalies or physical violations that deviate from the natural real-world manifold (off-manifold). For the remaining high-fidelity videos that successfully reside on-manifold and evade spatial detection, we introduce the Micro-Temporal Fluctuation Branch as a secondary, fine-grained filter. By analyzing the structured MPF that persists even in visually perfect sequences, our framework ensures that forgeries are exposed regardless of whether they manifest as global real-world manifold deviations or subtle computational fingerprints.

cs.CV↗

arXiv2026

What Do Agents Learn from Trajectory-SFT: Semantics or Interfaces?

Large language models are increasingly evaluated as interactive agents, yet standard agent benchmarks conflate two qualitatively distinct sources of success: semantic tool-use and interface-specific interaction pattern memorization. Because both mechanisms can yield identical task success on the original interface, benchmark scores alone are not identifiable evidence of environment-invariant capability. We propose PIPE, a protocol-level evaluation augmentation for diagnosing interface reliance by minimally rewriting environment interfaces while preserving task semantics and execution behavior. Across 16 environments from AgentBench and AgentGym and a range of open-source and API-based agents, PIPE reveals that trajectory-SFT substantially amplifies interface shortcutting: trained agents degrade sharply under minimal interface rewrites, while non-trajectory-trained models remain largely stable. We further introduce Interface Reliance (IR), a counterbalanced alias-based metric that quantifies preference for training-time interfaces, and show that interface shortcutting exhibits environment-dependent, non-monotonic training dynamics that remain invisible under standard evaluation. Our code is available at https://anonymous.4open.science/r/What-Do-Agents-Learn-from-Trajectory-SFT-Semantics-or-Interfaces--0831/.

cs.LG↗

arXiv2026

Rethinking the Sampling Criteria in Reinforcement Learning for LLM Reasoning: A Competence-Difficulty Alignment Perspective

Reinforcement learning exhibits potential in enhancing the reasoning abilities of large language models, yet it is hard to scale for the low sample efficiency during the rollout phase. Existing methods attempt to improve efficiency by scheduling problems based on problem difficulties. However, these approaches suffer from unstable and biased estimations of problem difficulty and fail to capture the alignment between model competence and problem difficulty in RL training, leading to suboptimal results. To tackle these limitations, this paper introduces $\textbf{C}$ompetence-$\textbf{D}$ifficulty $\textbf{A}$lignment $\textbf{S}$ampling ($\textbf{CDAS}$), which enables accurate and stable estimation of problem difficulties by aggregating historical performance discrepancies of problems. Then the model competence is quantified to adaptively select problems whose difficulty is in alignment with the model's current competence using a fixed-point system. Experimental results across a range of challenging mathematical benchmarks show that CDAS achieves great improvements in both accuracy and efficiency. CDAS attains the highest average accuracy against baselines and exhibits significant speed advantages compared to Dynamic Sampling, a competitive strategy in DAPO, which is 2.33 times slower than CDAS.

cs.LG↗