arXiv subjects

Wei Shi

Publications and source records attributed to Wei Shi.

At least 19 recordsLinked to original sources

arXiv2026

UGR: Unified Generative Ranking for Heterogeneous E-commerce Recommendation

List-wise ranking in e-commerce recommendation must balance user utility and commercial value across organic content, ads, and candidates from other businesses, while modeling candidate interactions and sequential dependencies. Existing generative ranking methods typically learn from logged exposure sequences and operate on truncated candidate sets from a single business. Exposure-sequence learning may not fully exploit post-exposure feedback signals. Meanwhile, generative ranking over heterogeneous, multi-business candidates remains less explored, particularly when recommendation scores and advertising value must be jointly modeled under generalized second-price (GSP) auctions. We propose Unified Generative Ranking (UGR), which performs generative ranking over the full request-level multi-business heterogeneous candidate set, typically comprising around a thousand candidates in production, and directly generates multi-business ranking lists through candidate-aware autoregressive decoding. To go beyond exposure-sequence imitation, UGR introduces complementary feedback supervision through weighted supervised fine-tuning and permutation preference optimization, which respectively leverage observed post-exposure feedback and feedback-derived preferences among permutations. UGR uses a training-time advertising debiasing tower to model historical ad exposure effects and incorporates current ad value during serving while preserving the existing allocation and GSP pricing rules. Multi-value sequence retrieval produces candidate ranking lists, and the existing downstream evaluator selects the final list for exposure. Offline experiments demonstrate consistent gains in ranking quality, while online A/B tests on Taobao's Guess You Like recommendation service yield improvements of 1.76% in Core IPV, 3.29% in Transaction Volume, and 2.10% in GMV over the production baseline.

cs.IR↗

arXiv2026

ACWM-RM: A Reference-Free Reward Model from Complementary Mismatch Preferences

Reinforcement learning has gained attention for improving action-conditioned world models (ACWMs), as it enables reward-driven optimization of visual quality (VQ) and action-following fidelity (AF) beyond flow-matching optimization. Realizing this potential requires reward signals that reliably rank videos by both VQ and AF. However, existing rewards capture only one dimension: metric-based rewards assess VQ but require ground-truth videos and do not directly assess AF, while inverse dynamics model (IDM) rewards assess AF without references but do not guarantee VQ. Moreover, IDMs, typically trained on real videos, may provide unreliable rewards for generated videos containing visual artifacts and physical inconsistencies. In this work, we introduce the Action and Video Mismatch Benchmark (AVM-Eval), comprising 67.2k pairs that evaluate how reliably IDMs assess AF on real and generated videos. We then propose ACWM-RM, a reward model that jointly assesses VQ and AF without ground-truth videos during reward computation. To train it, we first construct the Action and Video Mismatch Preference Dataset (AVM-Pref) with 506.3k pairs providing complementary supervision: action-mismatch pairs supervise AF, while video-mismatch pairs supervise VQ and expose the model to generated videos. We then pretrain an action-conditioned predictor on real videos to learn forward dynamics before jointly training it with a patch-weighted score head on AVM-Pref to transfer these dynamics to AF assessment on generated videos. Evaluation on AVM-Eval reveals the limited reliability of IDMs on generated videos, while downstream RL experiments show that ACWM-RM outperforms reward baselines in both criteria.

cs.CV↗

arXiv2026

SAPD: Learning from Agent Experience without Environment Interaction

Scaling the training of Large Language Model (LLM) agents typically requires a large collection of executable environments, whose construction, maintenance, and operation constrain the expansion of training experience. Existing offline trajectories offer an opportunity to reuse interaction experience, yet effectively refining the current policy without revisiting these environments remains challenging: supervised fine-tuning (SFT) can lead to catastrophic forgetting, while online reinforcement learning requires real-time interaction with the environment. To address this challenge, we propose Step-Advantage Policy Distillation (SAPD), a method for learning from offline trajectories using relative step advantages. SAPD decomposes multi-turn trajectories into individual decision contexts for on-policy distillation, allowing the student to generate responses conditioned on recorded histories, and introduces relative step advantage signals to stabilize training. The entire process requires neither new environment observations nor additional multi-turn interactions. Experiments on ALFWorld, WebShop, and Search-based QA show that SAPD improves performance across different teacher configurations, mitigates catastrophic forgetting compared with SFT, and runs at least $4.7\times$ faster per training step than online OPD. We will release our code soon.

cs.AI↗

arXiv2026

To Call or Not to Call: Diagnosing Intrinsic Over-Calling Bias in LLM Agents

LLM agents exhibit a consistent tendency to over-call, invoking tools even in situations where none is needed. On the When2Call benchmark, six models from three families show high call accuracy but much lower no-call accuracy, leaving overall accuracy in the 55%-70% range. We trace this to an Intrinsic Bias Hypothesis (IBH): the call/no-call decision mapping carries an activation-independent call offset, so the model favors call even at activation parity. Using Sparse Autoencoders (SAEs), we recover behavior-aligned feature bases for the call/no_call decision, reduce them to a signed activation margin, and estimate the offset directly. Across all six models, the model is decision-neutral only when no_call activation outweighs call activation, consistent with IBH. We then causally test IBH with Adaptive Margin-Calibrated Steering (AMCS), a closed-form counter-bias shift along SAE decoder directions. Cancelling the diagnosed offset mitigates over-calling and improves overall accuracy with a negligible drop in call accuracy. Our work recasts over-calling from an empirical phenomenon into a mechanistic object amenable to causal correction. Code is available at https://github.com/SKURA502/agent-sae/.

cs.LG↗

arXiv2026

OPD Before RL: Warm-Starting Rubric-Based RL with On-Policy Distillation

Many useful language-model tasks cannot be evaluated by exact outcome verification. Rubric-based reinforcement learning (RL) addresses this issue by scoring open-ended responses against explicit criteria. However, because the reward is assigned after the complete response, the training signal does not directly identify which individual decisions contributed to the final score. We propose a two-stage training framework that uses rubrics first as privileged teacher context for dense token-level supervision, then as rewards for further RL. In the first stage, rubric-privileged on-policy distillation (RP-OPD), a student without access to the rubric matches a rubric-aware teacher's next-token distributions at student-generated prefixes. In the second stage, RL directly optimizes the rubric reward and improves beyond the observed distillation plateau. We evaluate the framework on health and science tasks using open-weight models. Across HealthBench, ResearchQA, and RubricHub Science, we compare post-training methods and vary the amount of SFT or RP-OPD training before RL, finding that our two-stage framework achieves the highest scores among the methods evaluated. RP-OPD + RL shows limited signs of reward hacking on RubricHub Science, whereas the SFT + RL baseline increasingly receives high rewards for claims of rubric compliance without providing the required content. These findings support using rubrics to guide on-policy distillation before applying rubric-based RL.

cs.LG↗

arXiv2026

Faithful Activation Verbalization: Reducing Hallucinations in LLM Representation Interpretation

Activation verbalization methods such as Activation Oracle and Natural Language Autoencoders decode hidden representations of large language models into human-readable natural language. However, existing methods can produce incomplete or hallucinated descriptions, making their activation verbalizations difficult to trust and use reliably in practice. To this end, we introduce AVPO, a two-stage framework that first reconstructs source text from a hidden activation and then evaluates the resulting text with a separate frozen question-answering model, yielding an explicit and inspectable intermediate readout. We further optimize the inverter with direct preference optimization (DPO), using rewards that capture both semantic recoverability and lexical fidelity. Across six text families, AVPO improves gist- and detail-level information recovery over the strongest baseline by up to 17.1 and 9.3 percentage points, respectively. Crucially, the gains arise from preference optimization rather than fine-tuning on selected reconstructions alone, enabling compact cross-model inverters to surpass donor-matched question-conditioned verbalizers while improving both semantic recoverability and lexical fidelity. Moreover, out-of-distribution case study shows that AVPO better recovers high-level semantics while fabricating fewer details.

cs.CL↗

arXiv2026

Persistent Delivery Optimization for Streaming Speech-to-Text Translation with Revisions

Revision-capable streaming speech-to-text translation (S2TT) can correct earlier drafts, but process rewards based on visible text may credit content later withdrawn. Persistent Delivery Optimization (PDO) assigns intermediate reward only to content that survives revisions while scoring final quality separately. With 7.49 h of task-specific FLEURS adaptation, PDO achieves the best BLEU on four of five directions and higher COMET than every external streaming baseline in all five directions. Relative to its History-SFT initialization, PDO reduces mean/P90 finalization-aware latency by 10.8\%/11.3\% and normalized erasure by 15.8\%, while emitting at the first permitted 2-s update and improving macro BLEU. Zero-shot evaluation on Europarl-ST and CoVoST 2 confirms that these gains are not confined to the FLEURS training domain.

eess.AS↗

arXiv2026

Helpful but Fallible: Developer Experiences of AI Tools Under a Coordinated Industrial Roll-out

AI-enabled software development tools (AI-devtools) are being industrially adopted under strong expectations of productivity gains, yet developers' experiences of such roll-outs are underexplored. Organizations commit budgets, evaluate staff, and revise practice on a partial picture, since the evidence base is mainly tool evaluations, productivity metrics, and surveys, with few qualitative in-situ accounts of ongoing, coordinated roll-outs. We report a case study of a coordinated roll-out of AI-devtools at a large Swedish telecommunications company, investigating how developers experience the roll-out and how they anticipate their profession will change. We conducted semi-structured interviews with 12 software professionals across three sites, analyzed with process coding and thematic analysis, and interpreted through the extended Technology Acceptance Model (TAM2) as a post-hoc analytical lens. Our findings on use cases, productivity, frustrations, and tool limitations corroborate prior survey work. Beyond corroboration, the interviews surface a management-developer expectation gap that maps onto the TAM2 constructs of subjective norm and voluntariness, and show that participants weigh perceived risk heavily, a factor that TAM2 and similar acceptance models do not represent. AI-devtools emerge as helpful but fallible assistants whose value is shaped by organizational expectations, system scale, and developers' skills.

cs.SE↗

arXiv2026

Learning CNN Filters via Generalized Stein's Method

Convolutional Neural Networks (CNNs) have undoubtedly revolutionized image data analysis and the field of computer vision. As the cornerstone of CNNs, the convolution operation enables the networks to extract abstract features and uncover hidden relationships in the image data. This paper considers the problem of estimating convolution filters from a statistical perspective using a classical tool --- Stein's formula. We first formulate CNNs into a general index model with matrix-valued input, where convolution filters can be viewed as index vectors. Furthermore, we propose a novel singular value decomposition (SVD) based approach to accurately learn the convolution filters based on a generalized version of the first-order Stein's formula. Theoretical analysis suggests that our estimation achieves an optimal convergence rate, comparable to that of generalized linear models where the link function is known. Extensive simulation studies and real data analyses demonstrate that our approach outperforms popular deep learning algorithms, such as Adam. Notably, our method extends beyond filter estimation and can be applied to nonlinear dimension reduction, providing a viable pathway for representation learning.

stat.AP↗

arXiv2026

Answer Probing-Guided Search for Diverse Solution Exploration of LLMs

Generating multiple diverse and high-quality solutions is valuable for many applications, such as code-test generation and drug discovery. However, Large Language Models (LLMs) tend to converge on a single high-confidence solution during inference, limiting exploration of alternative valid solution paths. Existing test-time methods promote diversity through tree-like search and prune semantically similar branches using response-level semantic embeddings. However, we find that such embeddings are easily confounded by linguistic and stylistic similarities, making it difficult to distinguish genuinely distinct solution paths. To address this, we introduce Answer Probing, which probes the potential answer an LLM would reach from an intermediate reasoning path. We demonstrate that the hidden states of probed answers more effectively differentiate distinct solution paths than semantic embeddings, and the perplexity of probed answers serves as a practical proxy for reasoning correctness. Based on these findings, we propose Answer Probing-Guided Tree Search (APTS), which guides the tree search by the probed answers' hidden state similarity and perplexity. Experiments on three reasoning tasks across two LLMs show that APTS consistently enhances solution diversity, demonstrating its effectiveness and robustness.

cs.AI↗

arXiv2026

Reconstructing High-Fidelity Light Yield Maps for Surface LArTPCs Using Crossing Cosmic Muons

Photon detection systems in liquid argon time projection chambers provide prompt scintillation light information that can improve triggering, timing, calorimetry, and interaction reconstruction. These applications require an understanding of the spatial dependence of the detected light yield (LY), which is not adequately described by a single detector-wide average. We present a method for reconstructing voxelized 3D light yield maps in surface LArTPCs using crossing cosmic muons. For each selected muon, the path length through every crossed voxel is converted to deposited energy using a minimum ionizing particle approximation, producing a linear system relating the unknown voxel light yields to the total detected photon signal. The resulting inverse problem is solved using nonnegative least squares, with an additional $L_2$ smoothness penalty used to stabilize weakly constrained voxel values. The method is studied in simulation using the ProtoDUNE-VD detector geometry and photon detection system. Reconstructed maps recover the dominant spatial structure of a visibility-based truth reference and respond as expected to changes in the photon detector configuration. Smoothness regularization reduces zero-valued voxels and localized fluctuations while leaving the detector-average light yield approximately unchanged at the selected regularization strengths. The voxel-wise RMSE relative to the truth reference is reduced by approximately 44% for the regularized reconstruction.

physics.ins-det↗

arXiv2026

LLM-Enhanced Commit Message Generation via Issue Information: An Exploratory Study

Commit messages help developers understand code changes, support collaboration, and improve long-term maintenance. However, the use of issue information alone as the external context for LLM-based CMG has not been systematically studied. We propose an ISsue-Augmented framework for Commit message generation (ISAC) by combining code diffs with issue information as LLM input. To support the evaluation, we construct ApacheCM-Issue, a commit-issue aligned dataset built upon ApacheCM by linking commits with issues from GitHub and Apache Jira. Using samples from Scala, Java, and C++ projects, we evaluate four input configurations using two representative LLMs, GPT-5.5 and DeepSeek-V4-Flash in different reasoning configurations. The results show that incorporating issue information consistently improves LLM-based CMG across all evaluated model configurations and metrics, with the largest gains observed for CIDEr. Incorporating a similar historical commit further improves automatic metric scores, while replacing full issue information with a structured issue summary decreases them. ISAC also outperforms the four reproduced state-of-the-art (SOTA) CMG baselines across all five automatic metrics on the experimental dataset. The human evaluation further shows that structured issue summaries may improve perceived completeness, although replacing the original issue information can sacrifice contextual details and lead to worse results on automatic metrics.

cs.SE↗

arXiv2026

DREAM Technical Report

Industrial recommender systems commonly use cascaded retrieval, ranking, and re-ranking pipelines. Although efficient, these pipelines fragment information and objectives across modules, rely on rigid rules, and have limited awareness of real-time intent, leaving session-level shifts among browsing, comparison, and purchase insufficiently addressed. We present DREAM (Developing Recommender Engine with Agentic Methods), an autonomous optimization control architecture that adds a perception-aware, orchestrable, and auditable policy layer atop existing pipelines without replacing them. DREAM has two core components. First, a three-tier Intent Engine fuses on-device signals into structured L0/L1/L2 intent representations; its edge-cloud trigger chain reduces reporting volume to approximately 8.7%. Second, a Meta Engine uses a MetaModel for layered M1-to-M2-to-M3 reasoning: intent summarization, strategy planning informed by Strategy Memory, and parameter translation. It dispatches the resulting parameters through a unified outlet with safety guardrails. A Reward Dual Loop continuously optimizes both components by combining offline simulation for strategy-space exploration with online feedback for outcome calibration, forming a cycle of generation, execution, evaluation, and experience accumulation. Large-scale A/B tests on Taobao's homepage feed show that re-ranking control alone improves IPV by 2.06%, Core IPV by 2.39%, and GMV by 0.88%. Extending control to fine ranking raises these gains to 2.71%, 3.06%, and 1.31%, respectively, while consistently improving PV by more than 1%. These gains require neither replacement of pipeline models nor compromise of serving stability, supporting agentic meta-control as a viable paradigm for industrial recommendation.

cs.IR↗

arXiv2026

Integrated Energy System Optimal Operation in Coal District With Hydrogen Heavy Trucks

The coal industry contributes significantly to the social economy, but the emission of greenhouse gases puts huge pressure on the environment during mining, transportation, and power generation. In the integrated energy system (IES), current research on power-to-gas (P2G) technology mainly focuses on injecting hydrogen generated from renewable-energy-powered electrolysis into natural gas pipelines. However, hydrogen injection may cause irreversible hydrogen embrittlement of pipelines. In this paper, sufficient hydrogen is produced through both P2G technology and coal-to-hydrogen (C2H) based on coal gasification, considering the typical scenario where coal-producing regions are rich in both coal and renewable energy resources. To transport mined coal, hydrogen heavy-duty trucks provide a promising solution by consuming hydrogen in a timely manner, thereby avoiding potentially dangerous hydrogen injection into pipelines and reducing the need for relatively expensive hydrogen storage. An optimal scheduling model for an electric-gas IES is proposed based on second-order cone programming (SOCP). The proposed model innovatively studies a closed industrial loop including coal mining, hydrogen production, coal transportation by hydrogen trucks, and integrated energy systems to enhance renewable energy utilization and coordinate multiple energy carriers. Finally, a case study based on the IEEE 30-bus power system and the Belgium 24-node natural gas network is conducted. The results show that the proposed hydrogen production technology reduces typical daily operating costs by 7.7%. Under China's carbon emissions trading system, the operating costs of hydrogen heavy-duty trucks are reduced by 0.95% and 4.68% compared with electric vehicles and diesel trucks, respectively. Under Europe's stricter carbon emissions trading system, ...

eess.SY↗

arXiv2026

Enhanced Reconstruction of Sub-GeV Neutrinos Charged Current Interactions in LArTPC

This paper presents a comprehensive study of the reconstruction of sub-GeV neutrino charged-current interactions within a Liquid Argon Time Projection Chamber (LArTPC). We demonstrate that traditional charge-based calorimetry is fundamentally limited at sub-GeV scales by significant fluctuations in recombination and missing hadronic energy. We show that energy reconstruction using energy deposited as scintillation light ($L$) partially benefits from the self-compensating light effect. At neutrino energies above 400 MeV, the light-only reconstruction still outperforms charge-only methods that can separate EM and hadronic objects. The performance of the two remains comparable below 300 MeV. Using the energy-deposit information from both detector signals, we demonstrate at least 70\% efficiency can be achieved in tagging electron neutrinos and antineutrinos, respectively. By using a proximity-based algorithm coupled with a geometric lepton-exclusion cone, we also demonstrate the ability to isolate neutron-induced energy depositions from background. This enables an improvement of sub-GeV direction reconstruction by about 20 degrees for antineutrinos. This study provides new insights into how to enhance the physics reach of future LArTPC atmospheric neutrino analyses.

hep-ex↗

arXiv2026

Prompt Generation Technical Report

Generative retrieval has become an increasingly adopted paradigm for industrial search, recommendation, and advertising systems, delivering significant online gains. Most existing work combines user behavior sequences with large language models (LLMs) to model user preferences. In practice, feature engineering remains critical to model effectiveness, yet its complexity slows offline iteration and makes online deployment heavy and hard to reuse, all under tight online latency budgets. The root cause is a tight coupling between feature-processing logic and model architecture, where every feature change touches the training and serving code and resists reuse across scenarios. To break this coupling, we present Prompt Generation (PG), a high-level tokenizer and configuration-driven framework that decouples feature-processing logic from model architecture through two declarative JSON files, which serve as the single source of truth for both offline training and online serving, ensuring feature consistency across the two stages. Organizing features under four types with three composable processing components to assemble and compress heterogeneous features, PG delivers acceleration at three levels: (1)fast training iteration: feature experiments require only configuration changes, with built-in token compression for ultra-long sequences; (2)fast deployment: a new scenario only needs to conform to the PG schema and plug into a universal pipeline, with no scenario-specific engineering; (3)fast online inference: engine applies unified optimizations over the standardized configuration, reducing PG's overhead to a negligible level. PG has been deployed on Taobao Search with statistically significant online A/B uplifts of +0.47% in transaction count and +0.51% in GMV, and has been applied across multiple Taobao search and recommendation teams as the iteration framework for generative retrieval.

cs.IR↗

arXiv2026

UniRec-0.1B: Unified Text and Formula Recognition with 0.1B Parameters

Text and formulas constitute the core informational components of many documents. Accurately and efficiently recognizing both is crucial for developing robust and generalizable document parsing systems. Recently, vision-language models (VLMs) have achieved impressive unified recognition of text and formulas. However, they are large-sized and computationally demanding, restricting their usage in many applications. In this paper, we propose UniRec-0.1B, a unified recognition model with only 0.1B parameters. It is capable of performing text and formula recognition at multiple levels, including characters, words, lines, paragraphs, and documents. To implement this task, we first establish UniRec40M, a large-scale dataset comprises 40 million text, formula and mixed samples, enabling the training of a powerful yet lightweight model. Secondly, we identify two challenges when building such a lightweight but unified expert model. They are: structural variability across levels and semantic entanglement between textual and formulaic content. To tackle these, we introduce a hierarchical supervision training that explicitly guides structural comprehension, and a semantic-decoupled tokenizer that separates text and formula representations. Finally, we develop a comprehensive evaluation benchmark covering Chinese and English documents from multiple domains and with multiple levels. Experimental results on this and public benchmarks demonstrate that UniRec-0.1B outperforms both general-purpose VLMs and leading document parsing expert models, while achieving 2-9x speedup, validating its effectiveness and efficiency. Codebase and Dataset: https://github.com/Topdu/OpenOCR.

cs.CV↗

arXiv2026

RASFT: Rollout-Adaptive Supervised Fine-Tuning for Reasoning

Supervised fine-tuning (SFT) is a prevailing method for adapting large language models to reasoning tasks by imitating offline expert demonstrations, often treating a single expert trajectory as the target behavior. However, reasoning is not simple path imitation: rigidly following one demonstrated solution may overfit to surface forms and suppress the model's own reasoning distribution. We propose Rollout-Adaptive Supervised Fine-Tuning (RASFT), a policy-aware SFT framework that calibrates expert supervision according to problem-level solvability estimated from verified on-policy rollouts. For each problem, RASFT strengthens expert guidance when the current policy struggles, while relaxing rigid imitation and incorporating correct self-generated trajectories when the model already exhibits reliable reasoning behavior. To preserve useful reasoning priors, RASFT further introduces a clipped inverse ratio between the frozen reference model and the current policy to constrain excessive policy drift. Experiments across multiple models on six mathematical reasoning benchmarks and two code reasoning benchmarks show that RASFT achieves better overall performance than SFT, SFT variants, and representative RL methods. The code is available at https://github.com/zjd1sq/RASFT.

cs.LG↗