arXiv subjects

Haoran Xin

Publications and source records attributed to Haoran Xin.

13 recordsLinked to original sources

arXiv2026

Escaping the KL Agreement Trap in On-Policy Distillation

On-policy distillation (OPD) provides dense token-level supervision by asking a teacher to score student-generated rollouts. However, when the student drifts into an unrecoverable prefix, the teacher may locally agree with the degraded state, producing low reverse KL but little corrective training signal. We identify this persistent regime as a low-KL agreement trap. Further analyses show that tokens during and after such traps produce less useful supervision signals. We propose KAT (KL Agreement Trap Termination), an online OPD termination rule that detects persistent low-KL agreement with a dynamic training-adaptive threshold. By filtering weak supervision from degenerate agreement, KAT improves avg@k accuracy by 2.66% and pass@k by 3.43% across four mathematical benchmarks, while reducing average rollout length by 59.73%.

cs.LG↗

arXiv2026

Decoupling KL and Trajectories: A Unified Perspective for SFT, DAgger, Offline RL, and OPD in LLM Distillation

Knowledge distillation is central to LLM post-training, yet its design space remains poorly understood, especially alongside reinforcement learning (RL). We show that the prevailing paradigms, off-policy distillation and on-policy distillation (OPD), implicitly couple two orthogonal choices: prefix source and token-level KL direction. This follows from decomposing sequence-level KL over autoregressive response distributions: forward KL pairs teacher prefixes with token-level forward KL, and reverse KL pairs student prefixes with token-level reverse KL. We argue this coupling is not intrinsic: decoupling the two axes yields four valid objectives. We establish gradient-level identities showing forward KL gives SFT-style cross-entropy matching with teacher soft targets, whereas reverse KL gives an RL-style policy-gradient objective with a dense teacher-student log-ratio reward, connecting them to off-policy SFT, DAgger-style on-policy SFT, offline-RL-style distillation, and OPD. We conduct an extensive controlled study on math reasoning, evaluating the four objectives both as standalone methods and as initializations for subsequent RL. The results reveal three tradeoffs: KL direction induces an accuracy-entropy tradeoff, prefix source a quality-compute tradeoff, and training length an accuracy-stability tradeoff. Motivated by these findings, we propose KL mixing and an entropy-gated length curriculum. KL mixing shows long-sequence distillation requires substantial forward-KL weight to prevent entropy collapse and length inflation without sacrificing accuracy. The entropy-gated length curriculum improves Avg@k and Pass@k by 3.6 and up to 5.8 points, and cuts average response length by roughly 3x versus fixed long-horizon training. Our results provide a framework and practical methods for designing reasoning distillation objectives that balance accuracy, diversity, compute, and RL behavior.

cs.LG↗

arXiv2026

Tencent Advertising Algorithm Challenge 2025: All-Modality Generative Recommendation

Generative recommender systems are rapidly emerging as a new paradigm for recommendation, where collaborative identifiers and/or multi-modal content are mapped into discrete token spaces and user behavior is modelled with autoregressive sequence models. Despite progress on multi-modal recommendation datasets, there is still a lack of public benchmarks that jointly offer large-scale, realistic and fully all-modality data designed specifically for generative recommendation (GR) in industrial advertising. To foster research in this direction, we organised the Tencent Advertising Algorithm Challenge 2025, a global competition built on top of two all-modality datasets for GR: TencentGR-1M and TencentGR-10M. Both datasets are constructed from real de-identified Tencent Ads logs and contain rich collaborative IDs and multi-modal representations extracted with state-of-the-art embedding models. The preliminary track (TencentGR-1M) provides 1 million user sequences with up to 100 interacted items each, where each interaction is labeled with exposure and click signals, while the final track (TencentGR-10M) scales this to 10 million users and explicitly distinguishes between click and conversion events at both the sequence and target level. This paper presents the task definition, data construction process, feature schema, baseline GR model, evaluation protocol, and key findings from top-ranked and award-winning solutions. Our datasets focus on multi-modal sequence generation in an advertising setting and introduce weighted evaluation for high-value conversion events. We release our datasets at https://huggingface.co/datasets/TAAC2025 and baseline implementations at https://github.com/TencentAdvertisingAlgorithmCompetition/baseline_2025 to enable future research on all-modality generative recommendation at an industrial scale. The official website is https://algo.qq.com/2025.

cs.IR↗

arXiv2025

Who Taught the Lie? Responsibility Attribution for Poisoned Knowledge in Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) integrates external knowledge into large language models to improve response quality. However, recent work has shown that RAG systems are highly vulnerable to poisoning attacks, where malicious texts are inserted into the knowledge database to influence model outputs. While several defenses have been proposed, they are often circumvented by more adaptive or sophisticated attacks. This paper presents RAGOrigin, a black-box responsibility attribution framework designed to identify which texts in the knowledge database are responsible for misleading or incorrect generations. Our method constructs a focused attribution scope tailored to each misgeneration event and assigns a responsibility score to each candidate text by evaluating its retrieval ranking, semantic relevance, and influence on the generated response. The system then isolates poisoned texts using an unsupervised clustering method. We evaluate RAGOrigin across seven datasets and fifteen poisoning attacks, including newly developed adaptive poisoning strategies and multi-attacker scenarios. Our approach outperforms existing baselines in identifying poisoned content and remains robust under dynamic and noisy conditions. These results suggest that RAGOrigin provides a practical and effective solution for tracing the origins of corrupted knowledge in RAG systems. Our code is available at: https://github.com/zhangbl6618/RAG-Responsibility-Attribution

cs.CR↗

arXiv2025

Sculpting Topological Modes on Photonic Chips by Artificial Gauge Fields

Significant efforts have been devoted to manipulating topological states, which often manifest as localized modes at interfaces between distinct topological phases. In this work, we demonstrate a versatile approach to sculpting topological modes (TMs) into any desired shapes by incorporating various artificial gauge fields (AGFs), including scalar, vector, and imaginary gauge potentials, and leveraging the power of artificial neural networks (ANNs). These AGFs enable precise tuning of the dissipation of the TMs across that of bulk modes, facilitating a transition from localized to fully delocalized states. Moreover, ANNs allow precise engineering of these eigenmodes to achieve tailored profiles of topological states, which remain spectrally isolated within the bandgap and exhibit minimal loss compared to other modes. Our theoretical results are experimentally validated on silicon photonic platforms, demonstrating flexible manipulation of TM profiles. This approach enables the design of topological states with customized properties, offering significant potential for diverse applications in photonics and beyond.

physics.optics↗

arXiv2025

LLMs as Better Recommenders with Natural Language Collaborative Signals: A Self-Assessing Retrieval Approach

Incorporating collaborative information (CI) effectively is crucial for leveraging LLMs in recommendation tasks. Existing approaches often encode CI using soft tokens or abstract identifiers, which introduces a semantic misalignment with the LLM's natural language pretraining and hampers knowledge integration. To address this, we propose expressing CI directly in natural language to better align with LLMs' semantic space. We achieve this by retrieving a curated set of the most relevant user behaviors in natural language form. However, identifying informative CI is challenging due to the complexity of similarity and utility assessment. To tackle this, we introduce a Self-assessing COllaborative REtrieval framework (SCORE) following the retrieve-rerank paradigm. First, a Collaborative Retriever (CAR) is developed to consider both collaborative patterns and semantic similarity. Then, a Self-assessing Reranker (SARE) leverages LLMs' own reasoning to assess and prioritize retrieved behaviors. Finally, the selected behaviors are prepended to the LLM prompt as natural-language CI to guide recommendation. Extensive experiments on two public datasets validate the effectiveness of SCORE in improving LLM-based recommendation.

cs.IR↗

arXiv2025

Improving Recommendation Fairness without Sensitive Attributes Using Multi-Persona LLMs

Despite the success of recommender systems in alleviating information overload, fairness issues have raised concerns in recent years, potentially leading to unequal treatment for certain user groups. While efforts have been made to improve recommendation fairness, they often assume that users' sensitive attributes are available during model training. However, collecting sensitive information can be difficult, especially on platforms that involve no personal information disclosure. Therefore, we aim to improve recommendation fairness without any access to sensitive attributes. However, this is a non-trivial task because uncovering latent sensitive patterns from complicated user behaviors without explicit sensitive attributes can be difficult. Consequently, suboptimal estimates of sensitive distributions can hinder the fairness training process. To address these challenges, leveraging the remarkable reasoning abilities of Large Language Models (LLMs), we propose a novel LLM-enhanced framework for Fair recommendation withOut Sensitive Attributes (LLMFOSA). A Multi-Persona Sensitive Information Inference module employs LLMs with distinct personas that mimic diverse human perceptions to infer and distill sensitive information. Furthermore, a Confusion-Aware Sensitive Representation Learning module incorporates inference results and rationales to develop robust sensitive representations, considering the mislabeling confusion and collective consensus among agents. The model is then optimized by a formulated mutual information objective. Extensive experiments on two public datasets validate the effectiveness of LLMFOSA in improving fairness.

cs.IR↗

arXiv2025

Benchmarking Poisoning Attacks against Retrieval-Augmented Generation

Retrieval-Augmented Generation (RAG) has proven effective in mitigating hallucinations in large language models by incorporating external knowledge during inference. However, this integration introduces new security vulnerabilities, particularly to poisoning attacks. Although prior work has explored various poisoning strategies, a thorough assessment of their practical threat to RAG systems remains missing. To address this gap, we propose the first comprehensive benchmark framework for evaluating poisoning attacks on RAG. Our benchmark covers 5 standard question answering (QA) datasets and 10 expanded variants, along with 13 poisoning attack methods and 7 defense mechanisms, representing a broad spectrum of existing techniques. Using this benchmark, we conduct a comprehensive evaluation of all included attacks and defenses across the full dataset spectrum. Our findings show that while existing attacks perform well on standard QA datasets, their effectiveness drops significantly on the expanded versions. Moreover, our results demonstrate that various advanced RAG architectures, such as sequential, branching, conditional, and loop RAG, as well as multi-turn conversational RAG, multimodal RAG systems, and RAG-based LLM agent systems, remain susceptible to poisoning attacks. Notably, current defense techniques fail to provide robust protection, underscoring the pressing need for more resilient and generalizable defense strategies.

cs.CR↗

arXiv2025

Traceback of Poisoning Attacks to Retrieval-Augmented Generation

Large language models (LLMs) integrated with retrieval-augmented generation (RAG) systems improve accuracy by leveraging external knowledge sources. However, recent research has revealed RAG's susceptibility to poisoning attacks, where the attacker injects poisoned texts into the knowledge database, leading to attacker-desired responses. Existing defenses, which predominantly focus on inference-time mitigation, have proven insufficient against sophisticated attacks. In this paper, we introduce RAGForensics, the first traceback system for RAG, designed to identify poisoned texts within the knowledge database that are responsible for the attacks. RAGForensics operates iteratively, first retrieving a subset of texts from the database and then utilizing a specially crafted prompt to guide an LLM in detecting potential poisoning texts. Empirical evaluations across multiple datasets demonstrate the effectiveness of RAGForensics against state-of-the-art poisoning attacks. This work pioneers the traceback of poisoned texts in RAG systems, providing a practical and promising defense mechanism to enhance their security. Our code is available at: https://github.com/zhangbl6618/RAG-Responsibility-Attribution

cs.CR↗

arXiv2024

Observation of topology transition in Floquet non-Hermitian skin effects in silicon photonics

Non-Hermitian physics has greatly enriched our understanding of nonequilibrium phenomena and uncovered novel effects such as the non-Hermitian skin effect (NHSE) that has profoundly revolutionized the field. NHSE is typically predicted in systems with nonreciprocal couplings which, however, are difficult to realize in experiments. Without nonreciprocal couplings, the NHSE can also emerge in systems with coexisting gauge fields and loss or gain (e.g., in Floquet non-Hermitian systems). However, such Floquet NHSE remains largely unexplored in experiments. Here, we realize the Floquet NHSEs in periodically modulated optical waveguides integrated on a silicon photonics platform. By engineering the artificial gauge fields induced by the periodical modulation, we observe various Floquet NHSEs and unveil their rich topological transitions. Remarkably, we discover the transitions between the normal unipolar NHSEs and an unconventional bipolar NHSE which is accompanied by the directional reversal of the NHSEs. The underlying physics is revealed by the band winding in complex quasienergy space which undergoes a topology change from isolated loops with the same winding to linked loops with opposite windings. Our work unfolds a new route toward Floquet NHSEs originating from the interplay between gauge fields and dissipation effects and offers fundamentally new ways for steering light and other waves.

physics.optics↗

arXiv2024

EmotionGesture: Audio-Driven Diverse Emotional Co-Speech 3D Gesture Generation

Generating vivid and diverse 3D co-speech gestures is crucial for various applications in animating virtual avatars. While most existing methods can generate gestures from audio directly, they usually overlook that emotion is one of the key factors of authentic co-speech gesture generation. In this work, we propose EmotionGesture, a novel framework for synthesizing vivid and diverse emotional co-speech 3D gestures from audio. Considering emotion is often entangled with the rhythmic beat in speech audio, we first develop an Emotion-Beat Mining module (EBM) to extract the emotion and audio beat features as well as model their correlation via a transcript-based visual-rhythm alignment. Then, we propose an initial pose based Spatial-Temporal Prompter (STP) to generate future gestures from the given initial poses. STP effectively models the spatial-temporal correlations between the initial poses and the future gestures, thus producing the spatial-temporal coherent pose prompt. Once we obtain pose prompts, emotion, and audio beat features, we will generate 3D co-speech gestures through a transformer architecture. However, considering the poses of existing datasets often contain jittering effects, this would lead to generating unstable gestures. To address this issue, we propose an effective objective function, dubbed Motion-Smooth Loss. Specifically, we model motion offset to compensate for jittering ground-truth by forcing gestures to be smooth. Last, we present an emotion-conditioned VAE to sample emotion features, enabling us to generate diverse emotional results. Extensive experiments demonstrate that our framework outperforms the state-of-the-art, achieving vivid and diverse emotional co-speech 3D gestures. Our code and dataset will be released at the project page: https://xingqunqi-lab.github.io/Emotion-Gesture-Web/

cs.CV↗

arXiv2021

Intelligent Electric Vehicle Charging Recommendation Based on Multi-Agent Reinforcement Learning

Electric Vehicle (EV) has become a preferable choice in the modern transportation system due to its environmental and energy sustainability. However, in many large cities, EV drivers often fail to find the proper spots for charging, because of the limited charging infrastructures and the spatiotemporally unbalanced charging demands. Indeed, the recent emergence of deep reinforcement learning provides great potential to improve the charging experience from various aspects over a long-term horizon. In this paper, we propose a framework, named Multi-Agent Spatio-Temporal Reinforcement Learning (Master), for intelligently recommending public accessible charging stations by jointly considering various long-term spatiotemporal factors. Specifically, by regarding each charging station as an individual agent, we formulate this problem as a multi-objective multi-agent reinforcement learning task. We first develop a multi-agent actor-critic framework with the centralized attentive critic to coordinate the recommendation between geo-distributed agents. Moreover, to quantify the influence of future potential charging competition, we introduce a delayed access strategy to exploit the knowledge of future charging competition during training. After that, to effectively optimize multiple learning objectives, we extend the centralized attentive critic to multi-critics and develop a dynamic gradient re-weighting strategy to adaptively guide the optimization direction. Finally, extensive experiments on two real-world datasets demonstrate that Master achieves the best comprehensive performance compared with nine baseline approaches.

cs.LG↗

arXiv2021

Out-of-Town Recommendation with Travel Intention Modeling

Out-of-town recommendation is designed for those users who leave their home-town areas and visit the areas they have never been to before. It is challenging to recommend Point-of-Interests (POIs) for out-of-town users since the out-of-town check-in behavior is determined by not only the user's home-town preference but also the user's travel intention. Besides, the user's travel intentions are complex and dynamic, which leads to big difficulties in understanding such intentions precisely. In this paper, we propose a TRAvel-INtention-aware Out-of-town Recommendation framework, named TRAINOR. The proposed TRAINOR framework distinguishes itself from existing out-of-town recommenders in three aspects. First, graph neural networks are explored to represent users' home-town check-in preference and geographical constraints in out-of-town check-in behaviors. Second, a user-specific travel intention is formulated as an aggregation combining home-town preference and generic travel intention together, where the generic travel intention is regarded as a mixture of inherent intentions that can be learned by Neural Topic Model (NTM). Third, a non-linear mapping function, as well as a matrix factorization method, are employed to transfer users' home-town preference and estimate out-of-town POI's representation, respectively. Extensive experiments on real-world data sets validate the effectiveness of the TRAINOR framework. Moreover, the learned travel intention can deliver meaningful explanations for understanding a user's travel purposes.

cs.IR↗