arXiv subjects

Ziming Li

Publications and source records attributed to Ziming Li.

At least 37 records · Page 2Linked to original sources

arXiv2026

OneReason Technical Report

Generative recommendation models in the OneRec family have been widely deployed in many real-world services, such as short-video, live-streaming, advertising, and e-commerce. However, these generative models can only benefit from the scaling advantage, while their reasoning ability is hard to activate, since we cannot construct meaningful Chain-of-Thought (CoT) sequences consisting of itemic tokens only. Inspired by the success of the reasoning-style ``think before answer'' paradigm in the LLM field, we conduct preliminary studies (i.e., OneRec-Think, OpenOneRec) to explore reasoning capability in generative recommendation. Nevertheless, we notice an unexpected phenomenon: the thinking mode does not show advantages over the non-thinking mode. Drawing insights from recent findings on CoT robustness in multi-modal language models, we argue that effective reasoning in recommendation rests on two factors: perception, the ability to ground itemic tokens in their underlying language semantics, and cognition, the ability to reorganize a user's behavior sequence into coherent latent interest points. We therefore propose OneReason, which includes: (1) strong itemic token perception in pre-training, (2) a three-level cognition-enhanced CoT format for recommendation tasks in SFT, and (3) a specialize-then-unify training recipe in RL to enhance the thinking ability.

cs.IR↗

arXiv2026

Reinforcement Learning with Robust Rubric Rewards

While Reinforcement Learning with Verifiable Rewards (RLVR) is effective for deterministically checkable tasks, many vision-language tasks are partially verifiable, demanding multi-criteria supervision (e.g., perceptual details, reasoning steps, and constraints). Rubrics provide a natural interface for this fine-grained supervision, but their effectiveness depends on the execution accuracy during online RL. We propose Reinforcement Learning with Robust Rubric Rewards ($\text{RLR}^3$), extending RLVR from task-level verification to criterion-level verification. $\text{RLR}^3$ routes instance-specific rubrics through two execution paths: an LLM-as-an-extractor paired with a deterministic verifier, or an LLM-as-a-Judge for non-verifiable criteria. To ensure faithful scoring, $\text{RLR}^3$ introduce a minimal exposure strategy that masks ground truths from extractors and images from judges. Furthermore, $\text{RLR}^3$ employs hierarchical aggregation to prioritize essential criteria over additional criteria, and mitigates score saturation within rollout groups. Evaluated on Qwen3-VL-30B-A3B across 15 benchmarks, $\text{RLR}^3$ consistently outperforms RLVR, yielding a 4.7-point improvement over the base model and exceeding the official instruct-to-thinking model gap. Controlled audits confirm our deterministic verification and minimal exposure significantly reduce exploitable false positives.

cs.CV↗

arXiv2026

Counterfactual Graph for Multi-Agent LLM Calibration

Multi-agent LLM systems often treat agreement as evidence: when many agents in a panel give the same answer, that answer is assumed to be more reliable. We show that this assumption can fail after agents communicate. Communication can induce correlated failures and false consensus, so the same vote share may reflect reliable agreement in one topology but over-confidence in another. We propose CAGE-CAL, a counterfactual agent-graph calibration framework for multi-agent LLMs. For each query, CAGE-CAL compares an observed post-communication agent graph with a matched counterfactual no-communication graph, capturing both pairwise failure correlations and group-level dependencies. Rather than simply counting how many agents agree, CAGE-CAL estimates the counterfactual shift between observed and no-communication dependence, and calibrates confidence accordingly. Across five benchmarks, CAGE-CAL improves reliability discrimination with competitive ECE, and its calibrated confidence further improves topology selection over the best fixed-topology strategy.

cs.CL↗

arXiv2026

Graph is a Substrate Across Data Modalities

Graphs provide a natural representation of relational structure that arises across diverse domains. Despite this ubiquity, graph structure is typically learned in a modality- and task-isolated manner, where graph representations are constructed within individual task contexts and discarded thereafter. As a result, structural regularities across modalities and tasks are repeatedly reconstructed rather than accumulated at the level of intermediate graph representations. This motivates a representation-learning question: how should graph structure be organized so that it can persist and accumulate across heterogeneous modalities and tasks? We adopt a representation-centric perspective in which graph structure is treated as a structural substrate that persists across learning contexts. To instantiate this perspective, we propose G-Substrate, a graph substrate framework that organizes learning around shared graph structures. G-Substrate comprises two complementary mechanisms: a unified structural schema that ensures compatibility among graph representations across heterogeneous modalities and tasks, and an interleaved role-based training strategy that exposes the same graph structure to multiple functional roles during learning. Experiments across multiple domains, modalities, and tasks show that G-Substrate outperforms task-isolated and naive multi-task learning methods. The codebase, model, and datasets are available at https://github.com/zmli6/G-Substrate.

cs.LG↗

arXiv2026

On the Safety of Graph Representation Learning

Graph representation learning (GRL) has evolved from topology-only graph embeddings to task-specific supervised GNNs, and more recently to reusable representations and graph foundation models (GFMs). However, existing evaluations mainly measure clean transfer, adaptation, and task coverage. It remains unclear whether GRL methods stay reliable when deployment stresses affect graph signals, graph contexts, label support, structural groups, or predictive evidence. We introduce GRL-Safety, a multi-axis safety evaluation benchmark for GRL. GRL-Safety evaluates twelve representative methods, spanning topology-only embedding methods, supervised GNNs, self-supervised graph models, and GFMs, on twenty-five graph datasets under standardized evaluation conditions while preserving method-native adaptation. The evaluation covers five safety axes: corruption robustness, OOD generalization, class imbalance, fairness, and interpretation, with per-axis and sub-condition reporting rather than a single aggregate score. Our analysis yields three cross-axis insights that can inspire future research. First, safety behavior is shaped by the interaction between representation design and the stressed graph factor, rather than by method family alone. Second, foundation-era methods show axis-specific strengths rather than broad safety dominance. Third, several deployment regimes remain difficult even for the best evaluated method, revealing capability gaps that require new robustness, adaptation, or training objectives beyond model selection. The benchmark, evaluation protocols, and code are available at: https://github.com/GXG-CS/GRL-Safety.

cs.LG↗

arXiv2026

Visual Preference Optimization with Rubric Rewards

The effectiveness of Direct Preference Optimization (DPO) depends on preference data that reflect the quality differences that matter in multimodal tasks. Existing pipelines often rely on off-policy perturbations or coarse outcome-based signals, which are not well suited to fine-grained visual reasoning. We propose rDPO, a preference optimization framework based on instance-specific rubrics. For each image-instruction pair, we create a checklist-style rubric of essential and additional criteria to score responses from any possible policies. The instruction-rubric pool is built offline and reused during the construction of on-policy data. On public reward modeling benchmarks, rubric-based prompting massively improves a 30B-A3B judge and brings it close to GPT-5.4. On public downstream benchmarks, rubric-based filtering raises the macro average to 82.69, whereas outcome-based filtering drops it to 75.82 from 81.14. When evaluating scalability on a comprehensive benchmark, rDPO achieves 61.01, markedly outperforming the style-constrained baseline (52.36) and surpassing the 59.48 base model. Together, these results show that visual preference optimization benefits from combining on-policy data construction with instance-specific criterion-level feedback.

cs.CV↗

arXiv2026

Not All Tokens See Equally: Perception-Grounded Policy Optimization for Large Vision-Language Models

While Reinforcement Learning from Verifiable Rewards (RLVR) has advanced reasoning in Large Vision-Language Models (LVLMs), prevailing frameworks suffer from a foundational methodological flaw: by distributing identical advantages across all generated tokens, these methods inherently dilute the learning signals essential for optimizing the critical, visually-grounded steps of multimodal reasoning. To bridge this gap, we formulate \textit{Token Visual Dependency}, quantifying the causal information gain of visual inputs via the Kullback-Leibler (KL) divergence between visual-conditioned and text-only predictive distributions. Revealing that this dependency is highly sparse and semantically pivotal, we introduce Perception-Grounded Policy Optimization (PGPO), which is a novel fine-grained credit assignment framework that dynamically reshapes advantages at the token level. Through a threshold-gated, mass-conserving mechanism, PGPO actively amplifies learning signals for visually-dependent tokens while suppressing gradient noise from linguistic priors. Extensive experiments based on the Qwen2.5-VL series across seven challenging multimodal reasoning benchmarks demonstrate that PGPO boosts models by 18.7% on average. Both theoretical and empirical analyses confirm that PGPO effectively reduces gradient variance, prevents training collapse, and acts as a potent regularizer for robust, perception-grounded multimodal reasoning. Code will be released on https://github.com/Yzk1114/PGPO.

cs.AI↗

arXiv2026

Semantic Refinement with LLMs for Graph Representations

Graph-structured data exhibit substantial heterogeneity in where their predictive signals originate: in some domains, node-level semantics dominate, while in others, structural patterns play a central role. This structure-semantics heterogeneity implies that no graph learning model with a fixed inductive bias can generalize optimally across diverse graph domains. However, most existing methods address this challenge from the model side by incrementally injecting new inductive biases, which remains fundamentally limited given the open-ended diversity of real-world graphs. In this work, we take a data-centric perspective and treat node semantics as a task-adaptive variable. We propose a Graph-Exemplar-guided Semantic Refinement (GES) framework for graph representation learning which -- unlike existing LLM-enhanced methods that generate node descriptions without graph context -- leverages structurally and semantically similar nodes from the graph itself to guide semantic refinement. Specifically, a GNN is first trained to produce predictive states, which along with structural and semantic similarity are used to retrieve in-graph exemplars that inform an LLM in refining node descriptions. We evaluate our approach on both text-rich and text-free graphs. Results show consistent improvements on semantics-rich and structure-dominated graphs, demonstrating the effectiveness of data-centric semantic refinement under structure-semantics heterogeneity.

cs.CL↗

arXiv2026

FlexiCamAR: Enhancing Everyday Camera Interactions on AR Glasses with a Flexible Additional Viewpoint

The recent emergence and popularity of consumer-grade augmented reality (AR) glasses from major technology companies highlight their potential to become the next daily computing platform. A dominant design trend in this context is the integration of a front-facing camera to deliver a first-person perspective. While this approach is intuitive, there is limited evidence that it is optimal (or sufficient) for supporting users in daily tasks. This paper explores a more effective camera interaction technique for AR glasses, which we term ``FlexiCamAR." This novel method aims to enhance both efficiency and the range of applications for AR glasses by offering flexible and comfortable secondary camera viewpoints. To investigate the applicability and usability of this approach, we developed a ring camera prototype that can be attached to users' fingers. We then conducted a user study with 12 participants, comparing FlexiCamAR against the baseline, a traditional front-facing AR camera setup, across two common tasks: taking photos and scanning QR codes. Our findings show that FlexiCamAR significantly reduces physical load. We also explore potential scenarios where the additional viewpoint afforded by FlexiCamAR proves valuable, such as capturing low-angle perspectives or navigating confined spaces. Participant feedback further suggests strong potential for additional applications, including selfie taking, video conferencing, and object scanning. Overall, FlexiCamAR presents a novel interaction approach that can serve as a powerful supplement or alternative to the first-person perspective, significantly improving the adaptability of AR glasses for everyday use.

cs.HC↗

arXiv2026

VRSL:Exploring the Comprehensibility of 360-Degree Camera Feeds for Sign Language Communication in Virtual Reality

This study explores integrating sign language into virtual reality (VR) by examining the comprehensibility and user experience of viewing American Sign Language (ASL) videos captured with body-mounted 360-degree cameras. Ten participants identified ASL signs from videos recorded at three body-mounted positions: head, shoulder, and chest. Results showed the shoulder-mounted camera achieved the highest accuracy (85%), though differences between positions were not statistically significant. Participants noted that peripheral distortion in 360-degree videos impacted clarity, highlighting areas for improvement. Despite challenges, the overall comprehension success rate of 83.3% demonstrates the potential of video-based ASL communication in VR. Feedback emphasized the need to refine camera angles, reduce distortion, and explore alternative mounting positions. Participants expressed a preference for signing over text-based communication in VR, highlighting the importance of developing this approach to enhance accessibility and collaboration for Deaf and Hard of Hearing (DHH) users in virtual environments.

cs.HC↗

arXiv2026

Kelix Technical Report

Autoregressive large language models (LLMs) scale well by expressing diverse tasks as sequences of discrete natural-language tokens and training with next-token prediction, which unifies comprehension and generation under self-supervision. Extending this paradigm to multimodal data requires a shared, discrete representation across modalities. However, most vision-language models (VLMs) still rely on a hybrid interface: discrete text tokens paired with continuous Vision Transformer (ViT) features. Because supervision is largely text-driven, these models are often biased toward understanding and cannot fully leverage large-scale self-supervised learning on non-text data. Recent work has explored discrete visual tokenization to enable fully autoregressive multimodal modeling, showing promising progress toward unified understanding and generation. Yet existing discrete vision tokens frequently lose information due to limited code capacity, resulting in noticeably weaker understanding than continuous-feature VLMs. We present Kelix, a fully discrete autoregressive unified model that closes the understanding gap between discrete and continuous visual representations.

cs.CV↗

arXiv2026

OpenOneRec Technical Report

While the OneRec series has successfully unified the fragmented recommendation pipeline into an end-to-end generative framework, a significant gap remains between recommendation systems and general intelligence. Constrained by isolated data, they operate as domain specialists-proficient in pattern matching but lacking world knowledge, reasoning capabilities, and instruction following. This limitation is further compounded by the lack of a holistic benchmark to evaluate such integrated capabilities. To address this, our contributions are: 1) RecIF Bench & Open Data: We propose RecIF-Bench, a holistic benchmark covering 8 diverse tasks that thoroughly evaluate capabilities from fundamental prediction to complex reasoning. Concurrently, we release a massive training dataset comprising 96 million interactions from 160,000 users to facilitate reproducible research. 2) Framework & Scaling: To ensure full reproducibility, we open-source our comprehensive training pipeline, encompassing data processing, co-pretraining, and post-training. Leveraging this framework, we demonstrate that recommendation capabilities can scale predictably while mitigating catastrophic forgetting of general knowledge. 3) OneRec-Foundation: We release OneRec Foundation (1.7B and 8B), a family of models establishing new state-of-the-art (SOTA) results across all tasks in RecIF-Bench. Furthermore, when transferred to the Amazon benchmark, our models surpass the strongest baselines with an average 26.8% improvement in Recall@10 across 10 diverse datasets (Figure 1). This work marks a step towards building truly intelligent recommender systems. Nonetheless, realizing this vision presents significant technical and theoretical challenges, highlighting the need for broader research engagement in this promising direction.

cs.IR↗

arXiv2026

Scaling Rough Terrain Locomotion with Automatic Curriculum Reinforcement Learning

Curriculum learning has demonstrated substantial effectiveness in robot learning. However, it still faces limitations when scaling to complex, wide-ranging task spaces. Such task spaces often lack a well-defined difficulty structure, making the difficulty ordering required by previous methods challenging to define. We propose a Learning Progress-based Automatic Curriculum Reinforcement Learning (LP-ACRL) framework, which estimates the agent's learning progress online and adaptively adjusts the task-sampling distribution, thereby enabling automatic curriculum generation without prior knowledge of the difficulty distribution over the task space. Policies trained with LP-ACRL enable the ANYmal D quadruped to achieve and maintain stable, high-speed locomotion at 2.5 m/s linear velocity and 3.0 rad/s angular velocity across diverse terrains, including stairs, slopes, gravel, and low-friction flat surfaces--whereas previous methods have generally been limited to high speeds on flat terrain or low speeds on complex terrain. Experimental results demonstrate that LP-ACRL exhibits strong scalability and real-world applicability, providing a robust baseline for future research on curriculum generation in complex, wide-ranging robotic learning task spaces.

cs.RO↗

arXiv2026

SIN-Bench: Tracing Native Evidence Chains in Long-Context Multimodal Scientific Interleaved Literature

Evaluating whether multimodal large language models truly understand long-form scientific papers remains challenging: answer-only metrics and synthetic "Needle-In-A-Haystack" tests often reward answer matching without requiring a causal, evidence-linked reasoning trace in the document. We propose the "Fish-in-the-Ocean" (FITO) paradigm, which requires models to construct explicit cross-modal evidence chains within native scientific documents. To operationalize FITO, we build SIN-Data, a scientific interleaved corpus that preserves the native interleaving of text and figures. On top of it, we construct SIN-Bench with four progressive tasks covering evidence discovery (SIN-Find), hypothesis verification (SIN-Verify), grounded QA (SIN-QA), and evidence-anchored synthesis (SIN-Summary). We further introduce "No Evidence, No Score", scoring predictions when grounded to verifiable anchors and diagnosing evidence quality via matching, relevance, and logic. Experiments on eight MLLMs show that grounding is the primary bottleneck: Gemini-3-pro achieves the best average overall score (0.573), while GPT-5 attains the highest SIN-QA answer accuracy (0.767) but underperforms on evidence-aligned overall scores, exposing a gap between correctness and traceable support.

cs.CL↗

arXiv2025

Say it or AI it: Evaluating Hands-Free Text Correction in Virtual Reality

Text entry in Virtual Reality (VR) is challenging, even when accounting for the use of controllers. Prior work has tackled this challenge head-on, improving the efficiency of input methods. These techniques have the advantage of allowing for relatively straightforward text correction. However, text correction without the use of controllers is a topic that has not received the same amount of attention, even though it can be desirable in several scenarios, and can even be the source of frustration. Large language models have been adopted and evaluated as a corrective methodology, given their high power for predictions. Nevertheless, their predictions are not always correct, which can lead to lower usability. In this paper, we investigate whether, for text correction in VR that is hands-free, the use of AI could surpass in terms of usability and efficiency. We observed better usability for AI text correction when compared to voice input.

cs.HC↗

arXiv2025

ReVul-CoT: Towards Effective Software Vulnerability Assessment with Retrieval-Augmented Generation and Chain-of-Thought Prompting

Context: Software Vulnerability Assessment (SVA) plays a vital role in evaluating and ranking vulnerabilities in software systems to ensure their security and reliability. Objective: Although Large Language Models (LLMs) have recently shown remarkable potential in SVA, they still face two major limitations. First, most LLMs are trained on general-purpose corpora and thus lack domain-specific knowledge essential for effective SVA. Second, they tend to rely on shallow pattern matching instead of deep contextual reasoning, making it challenging to fully comprehend complex code semantics and their security implications. Method: To alleviate these limitations, we propose a novel framework ReVul-CoT that integrates Retrieval-Augmented Generation (RAG) with Chain-of-Thought (COT) prompting. In ReVul-CoT, the RAG module dynamically retrieves contextually relevant information from a constructed local knowledge base that consolidates vulnerability data from authoritative sources (such as NVD and CWE), along with corresponding code snippets and descriptive information. Building on DeepSeek-V3.1, CoT prompting guides the LLM to perform step-by-step reasoning over exploitability, impact scope, and related factors Results: We evaluate ReVul-CoT on a dataset of 12,070 vulnerabilities. Experimental results show that ReVul-CoT outperforms state-of-the-art SVA baselines by 16.50%-42.26% in terms of MCC, and outperforms the best baseline by 10.43%, 15.86%, and 16.50% in Accuracy, F1-score, and MCC, respectively. Our ablation studies further validate the contributions of considering dynamic retrieval, knowledge integration, and CoT-based reasoning. Conclusion: Our results demonstrate that combining RAG with CoT prompting significantly enhances LLM-based SVA and points out promising directions for future research.

cs.SE↗

arXiv2025

Debiasing LLMs by Masking Unfairness-Driving Attention Heads

Large language models (LLMs) increasingly mediate decisions in domains where unfair treatment of demographic groups is unacceptable. Existing work probes when biased outputs appear, but gives little insight into the mechanisms that generate them, leaving existing mitigations largely fragile. In this paper, we conduct a systematic investigation LLM unfairness and propose DiffHeads, a lightweight debiasing framework for LLMs. We first compare Direct-Answer (DA) prompting to Chain-of-Thought (CoT) prompting across eight representative open- and closed-source LLMs. DA will trigger the nature bias part of LLM and improve measured unfairness by 534.5%-391.9% in both one-turn and two-turn dialogues. Next, we define a token-to-head contribution score that traces each token's influence back to individual attention heads. This reveals a small cluster of bias heads that activate under DA but stay largely dormant with CoT, providing the first causal link between prompting strategy and bias emergence. Finally, building on this insight, we propose DiffHeads that identifies bias heads through differential activation analysis between DA and CoT, and selectively masks only those heads. DiffHeads reduces unfairness by 49.4%, and 40.3% under DA and CoT, respectively, without harming model utility.

cs.CL↗

arXiv2025

OneRec-V2 Technical Report

Recent breakthroughs in generative AI have transformed recommender systems through end-to-end generation. OneRec reformulates recommendation as an autoregressive generation task, achieving high Model FLOPs Utilization. While OneRec-V1 has shown significant empirical success in real-world deployment, two critical challenges hinder its scalability and performance: (1) inefficient computational allocation where 97.66% of resources are consumed by sequence encoding rather than generation, and (2) limitations in reinforcement learning relying solely on reward models. To address these challenges, we propose OneRec-V2, featuring: (1) Lazy Decoder-Only Architecture: Eliminates encoder bottlenecks, reducing total computation by 94% and training resources by 90%, enabling successful scaling to 8B parameters. (2) Preference Alignment with Real-World User Interactions: Incorporates Duration-Aware Reward Shaping and Adaptive Ratio Clipping to better align with user preferences using real-world feedback. Extensive A/B tests on Kuaishou demonstrate OneRec-V2's effectiveness, improving App Stay Time by 0.467%/0.741% while balancing multi-objective recommendations. This work advances generative recommendation scalability and alignment with real-world feedback, representing a step forward in the development of end-to-end recommender systems.

cs.IR↗