arXiv subjects

Jinghao Zhang

Publications and source records attributed to Jinghao Zhang.

At least 19 recordsLinked to original sources

arXiv2026

Robust Bimanual Vision-Language-Action Models via Embarrassingly Simple Modality Masking

Query-based Vision-Language-Action (VLA) models offer low-latency inference that is attractive for bimanual robotic manipulation, but we observe that they can still exhibit discontinuous actions and execution failures in complex dual-arm tasks. We hypothesize that unstable multi-view and language fusion is one contributing factor in these failures, often coinciding with attention spreading to distracting regions. To improve robustness, we introduce the Modality Masking Mechanism (M3), an embarrassingly simple, training-only strategy that requires no architectural changes or large-scale robot pretraining. M3 stochastically masks subsets of modality channels during training, exposing the policy to controlled partial observations and encouraging it to rely less on distracting cues and more on evidence that remains reliable. We evaluate M3 on ten bimanual tasks from RoboTwin 2.0 and on three long-horizon real-world tasks. Compared with the Adapter baseline, M3 improves average success by 21.7% in the Clean setting and 11.4% in Clean2Rand, where policies are trained on clean demonstrations and evaluated on randomized scenes, while also improving averaged real-world full-task success by over 30%. These results suggest that structured training-time masking is a practical way to improve the robustness of query-based VLA policies for bimanual manipulation.

cs.RO↗

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

Not only where, But when: Temporal Scheduling for RLVR

Reinforcement learning with verifiable rewards (RLVR) has become a core technique for post-training of Large Language Models (LLMs). While policy optimization is driven by all sampled tokens under a globally broadcast scalar reward, the heterogeneous policy behaviors exhibited along trajectories are largely overlooked without differentiation. Existing works address this by credit allocation, including token-level advantage reweighting, and selective token optimization, however, the allocation criterion are principally stagnant throughout training, limiting resilient policy evolution. In this work, we argue that \textit{when} learning signals are scheduled can be as important as \textit{where} they are allocated across tokens, and introduce the temporal dimension that scheduling the credit allocation criteria over the course of RLVR optimization. We find that prioritizing targeted tokens emphasized with specific policy behaviors, and gradually attenuating toward general optimization leads to more stable and efficient learning dynamics. Furthermore, we show that simple trajectory percentiles provide a natural perspective for distinguishing policy behaviors, and works effectively with temporal scheduling. Our analysis reveals that standard optimization substantially sacrifices policy entropy when simultaneously accommodating heterogeneous behaviors, whereas temporal scheduling yields healthier policy evolution dynamics. Experiments across mathematical and general reasoning benchmarks demonstrate consistent improvements, suggesting that temporal scheduling constitutes a promising optimization dimension.

cs.LG↗

arXiv2026

Kwai Summary Attention Technical Report

Long-context ability, has become one of the most important iteration direction of next-generation Large Language Models, particularly in semantic understanding/reasoning, code agentic intelligence and recommendation system. However, the standard softmax attention exhibits quadratic time complexity with respect to sequence length. As the sequence length increases, this incurs substantial overhead in long-context settings, leading the training and inference costs of extremely long sequences deteriorate rapidly. Existing solutions mitigate this issue through two technique routings: i) Reducing the KV cache per layer, such as from the head-level compression GQA, and the embedding dimension-level compression MLA, but the KV cache remains linearly dependent on the sequence length at a 1:1 ratio. ii) Interleaving with KV Cache friendly architecture, such as local attention SWA, linear kernel GDN, but often involve trade-offs among KV Cache and long-context modeling effectiveness. Besides the two technique routings, we argue that there exists an intermediate path not well explored: {Maintaining a linear relationship between the KV cache and sequence length, but performing semantic-level compression through a specific ratio $k$}. This $O(n/k)$ path does not pursue a ``minimum KV cache'', but rather trades acceptable memory costs for complete, referential, and interpretable retention of long distant dependency. Motivated by this, we propose Kwai Summary Attention (KSA), a novel attention mechanism that reduces sequence modeling cost by compressing historical contexts into learnable summary tokens.

cs.CL↗

arXiv2026

SEA-Eval: A Benchmark for Evaluating Self-Evolving Agents Beyond Episodic Assessment

Current LLM-based agents demonstrate strong performance in episodic task execution but remain constrained by static toolsets and episodic amnesia, failing to accumulate experience across task boundaries. This paper formalizes the Self-Evolving Agent (SEA) from the perspective of digital embodiment and continuous cross-task evolution, introduces the Evolutionary Flywheel as its minimal sufficient architecture, and presents SEA-Eval -- the first benchmark designed specifically for evaluating SEAs. Grounded in Flywheel theory, SEA-Eval establishes SR and T as primary metrics and, through sequential task stream design, is designed to quantify evolutionary gain, evolutionary stability, and implicit alignment convergence. Empirical evaluation reveals that, under comparable success rates, token consumption differs by up to 31.2 times between frameworks on individual tasks, with divergent evolutionary trajectories emerging under sequential analysis -- demonstrating that success rate alone creates a capability illusion and that the sequential convergence of $T$ is the key criterion for distinguishing genuine evolution from pseudo-evolution.

cs.AI↗

arXiv2026

Novae breves from magnetar giant flares: Potential probes of neutron star crusts

Matter ejected from the magnetar crust during giant flares (GFs) may undergo $r$-process nucleosynthesis, producing short-lived optical transients termed "novae breves". Although intrinsically much fainter than kilonovae from compact binary mergers, novae breves may occur within or near the Galaxy, making them promising observational targets. We aim to investigate how the neutron star (NS) equation of state (EOS) and the mass of the central magnetar affect the ejecta properties following GFs and the resulting nova brevis emission. We employ a semi-analytical ejecta model combined with nuclear reaction network calculations to compute nucleosynthesis yields and multi-band light curves for different EOSs and magnetar masses, and assess their detectability with current and future facilities. We find that variations in the EOS and magnetar mass modify the ejecta mass and its density and velocity distributions, etc., leading to observable differences in nova brevis light curves. In particular, both the peak luminosity and the characteristic peak timescale are EOS-dependent. Assuming a fixed Galactic magnetar mass of 1.4 solar mass and taking the $u$ band as an example, we find that the minimum apparent AB magnitudes range from 7 mag (H4 EOS) to 8.5 mag (WFF EOS) with peak timescales of 100-1000 s. A more massive magnetar produces fainter emission with a shorter peak timescale. For a magnetar mass of 1.4 solar mass, novae breves associated with known magnetars may reach peak luminosities of 1e37-1e39 erg/s, enabling targeted searches, particularly following high-energy GF alerts. Moreover, a detection horizon of 10 Mpc or beyond is achievable with current and future facilities, allowing searches for novae breves from previously unknown magnetars in the Local Volume. Although challenging, detection of such rapidly evolving transients is feasible.

astro-ph.HE↗

arXiv2026

Experimental Asynchronous Measurement-Device-Independent Quantum Cryptographic Conferencing

The quantum cryptographic conferencing (QCC) protocol, which distributes identical secure keys to user groups, is a crucial component of the quantum network. Previous experimental works have implemented the measurement-device-independent (MDI) QCC, of which the key rate in an $N$-user network scales down as $R\sim O(\eta^N)$, respectively. Building on the MDI QCC protocol, the asynchronous MDI (AMDI) QCC protocol theoretically integrates the mode pairing scheme into QCC, significantly boosting the key rate to $R\sim O(\eta)$, which is independent of the number of users, and thus demonstrating greater application potential. Experimentally, in this work, we implement the three-user AMDI QCC network without global phase tracking by adopting the fast Fourier transform-based frequency difference estimation and the phase drift compensation technique. Finally, we achieve a key rate of about $4.470\times10^{-9}$ bits per pulse under a maximum overall loss of about 59.6 dB. This work provides a scalable solution for the development of large-scale quantum communication networks in the future.

quant-ph↗

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

ToolWeaver: Weaving Collaborative Semantics for Scalable Tool Use in Large Language Models

Prevalent retrieval-based tool-use pipelines struggle with a dual semantic challenge: their retrievers often employ encoders that fail to capture complex semantics, while the Large Language Model (LLM) itself lacks intrinsic tool knowledge from its natural language pretraining. Generative methods offer a powerful alternative by unifying selection and execution, tasking the LLM to directly learn and generate tool identifiers. However, the common practice of mapping each tool to a unique new token introduces substantial limitations: it creates a scalability and generalization crisis, as the vocabulary size explodes and each tool is assigned a semantically isolated token. This approach also creates a semantic bottleneck that hinders the learning of collaborative tool relationships, as the model must infer them from sparse co-occurrences of monolithic tool IDs within a vast library. To address these limitations, we propose ToolWeaver, a novel generative tool learning framework that encodes tools into hierarchical sequences. This approach makes vocabulary expansion logarithmic to the number of tools. Crucially, it enables the model to learn collaborative patterns from the dense co-occurrence of shared codes, rather than the sparse co-occurrence of monolithic tool IDs. We generate these structured codes through a novel tokenization process designed to weave together a tool's intrinsic semantics with its extrinsic co-usage patterns. These structured codes are then integrated into the LLM through a generative alignment stage, where the model is fine-tuned to produce the hierarchical code sequences. Evaluation results with nearly 47,000 tools show that ToolWeaver significantly outperforms state-of-the-art methods, establishing a more scalable, generalizable, and semantically-aware foundation for advanced tool-augmented agents.

cs.AI↗

arXiv2025

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↗

arXiv2025

WaterWave: Bridging Underwater Image Enhancement into Video Streams via Wavelet-based Temporal Consistency Field

Underwater video pairs are fairly difficult to obtain due to the complex underwater imaging. In this case, most existing video underwater enhancement methods are performed by directly applying the single-image enhancement model frame by frame, but a natural issue is lacking temporal consistency. To relieve the problem, we rethink the temporal manifold inherent in natural videos and observe a temporal consistency prior in dynamic scenes from the local temporal frequency perspective. Building upon the specific prior and no paired-data condition, we propose an implicit representation manner for enhanced video signals, which is conducted in the wavelet-based temporal consistency field, WaterWave. Specifically, under the constraints of the prior, we progressively filter and attenuate the inconsistent components while preserving motion details and scenes, achieving a natural-flowing video. Furthermore, to represent temporal frequency bands more accurately, an underwater flow correction module is designed to rectify estimated flows considering the transmission in underwater scenes. Extensive experiments demonstrate that WaterWave significantly enhances the quality of videos generated using single-image underwater enhancements. Additionally, our method demonstrates high potential in downstream underwater tracking tasks, such as UOSTrack and MAT, outperforming the original video by a large margin, i.e., 19.7% and 9.7% on precise respectively.

cs.CV↗

arXiv2025

EtCon: Edit-then-Consolidate for Reliable Knowledge Editing

Knowledge editing aims to update specific facts in large language models (LLMs) without full retraining. Prior efforts sought to tune the knowledge layers of LLMs, achieving improved performance in controlled, teacher-forced evaluations. However, they still encounter challenges in real-world autoregressive generation scenarios, which greatly limit their practical applicability. Our empirical analysis reveals two issues: (1) Most methods degrade pre-trained capabilities after injecting new knowledge; (2) They may exhibit a discrepancy between stored parametric knowledge and inference-time autoregressive generation behavior. To this end, we propose EtCon, an edit-then-consolidate paradigm that couples targeted edits with post-edit consolidation. Specifically, our framework comprises two stages: (1) Targeted Proximal Supervised Fine-Tuning (TPSFT) performs a constrained targeted edit to update parametric knowledge while controlling policy drift. (2) Group Relative Policy Optimization (GRPO) consolidates the edit by aligning autoregressive trajectories with the intended fact. Extensive experiments demonstrate that our EtCon improves editing reliability and real-world generalization, while better preserving pre-trained capabilities.

cs.CL↗

arXiv2025

Mini-supernovae from white dwarf-neutron star mergers: Viewing-angle-dependent spectra and lightcurves

Unstable mass transfer may occur during white dwarf-neutron star (WD-NS) mergers, in which the WD can be tidally disrupted and form an accretion disk around the NS. Such an accretion disk can produce unbound wind ejecta, with synthesized $^{56}\mathrm{Ni}$ mixed in. Numerical simulations reveal that this unbound ejecta should be strongly polar-dominated, which may cause the following radioactive-powered thermal transient to be viewing-angle-dependent. This issue has so far received limited investigation. We investigate how the intrinsically non-spherical geometry of WD-NS wind ejecta affects the viewing-angle dependence of the thermal transients. Using a two-dimensional axisymmetric ejecta configuration and incorporating heating from the radioactive decay of $^{56}\mathrm{Ni}$, we employ a semi-analytical discretization scheme to simulate the observed viewing-angle-dependent photospheric evolution, as well as the resulting spectra and lightcurves. The observed photosphere evolves over time and depends strongly on the viewing angle: off-axis observers can see deeper, hotter inner layers of the ejecta and larger projected photospheric areas compared to on-axis observers. For a fiducial WD-NS merger producing 0.3 solar mass of ejecta and 0.01 solar mass of synthesized $^{56}\mathrm{Ni}$, the resulting peak optical absolute magnitudes of the transient span from ~ -12 mag along the polar direction to ~ -16 mag along the equatorial direction, corresponding to luminosities of $10^{40}$-$10^{42}$ erg s$^{-1}$. The typical peak timescales are expected to be 3-10 d. We for the first time explore the viewing-angle effect on WD-NS merger transients. Since their ejecta composition and energy sources resemble those of supernovae, yet WD-NS merger transients are dimmer and evolve more rapidly, we propose using "mini-supernovae" to describe the thermal emission following WD-NS mergers.

astro-ph.HE↗

arXiv2025

OneRec-Think: In-Text Reasoning for Generative Recommendation

The powerful generative capacity of Large Language Models (LLMs) has instigated a paradigm shift in recommendation. However, existing generative models (e.g., OneRec) operate as implicit predictors, critically lacking the capacity for explicit and controllable reasoning-a key advantage of LLMs. To bridge this gap, we propose OneRec-Think, a unified framework that seamlessly integrates dialogue, reasoning, and personalized recommendation. OneRec-Think incorporates: (1) Itemic Alignment: cross-modal Item-Textual Alignment for semantic grounding; (2) Reasoning Activation: Reasoning Scaffolding to activate LLM reasoning within the recommendation context; and (3) Reasoning Enhancement, where we design a recommendation-specific reward function that accounts for the multi-validity nature of user preferences. Experiments across public benchmarks show state-of-the-art performance. Moreover, our proposed "Think-Ahead" architecture enables effective industrial deployment on Kuaishou, achieving a 0.159\% gain in APP Stay Time and validating the practical efficacy of the model's explicit reasoning capability.

cs.IR↗

arXiv2025

RLFR: Extending Reinforcement Learning for LLMs with Flow Environment

Reinforcement Learning with Verifiable Rewards (RLVR) has recently emerged as a promising framework for improving reasoning abilities in Large Language Models (LLMs). However, policy optimized with binary verification prone to overlook potential valuable exploration in reasoning trajectory. In view of heavy annotation cost of golden Process Reward Models (PRMs), recent works attempt using auxiliary signals for reward shaping of process tokens, involving entropy and likelihood collected from logit space. In this work, we offer a novel perspective on shaping RLVR with flow rewards derived from latent space, and propose RLFR, where the flow fields of model latents are constructed from either off-policy high-quality data and on-policy rejection sampling data, and the velocity deviations of policy latents within it are quantified to serve as a reward signal. RLFR first demonstrates that a well-established flow field can be a sound environment for reward signal collection, highlighting the expressive latent space is much underexplored. Moreover, RLFR is able to compress any off-policy expert data as reference for constituting reward signals, and we show that the efficient context dependence compressed within the hidden states are utilized, rather than individual token-level denotation for context comprehending. Experiments on both language and multimodal reasoning benchmarks demonstrate the reliability of flow rewards, and suggesting a promising paradigm for reward shaping with auxiliary signals.

cs.LG↗

arXiv2025

Unleashing the Potential of the Semantic Latent Space in Diffusion Models for Image Dehazing

Diffusion models have recently been investigated as powerful generative solvers for image dehazing, owing to their remarkable capability to model the data distribution. However, the massive computational burden imposed by the retraining of diffusion models, coupled with the extensive sampling steps during the inference, limit the broader application of diffusion models in image dehazing. To address these issues, we explore the properties of hazy images in the semantic latent space of frozen pre-trained diffusion models, and propose a Diffusion Latent Inspired network for Image Dehazing, dubbed DiffLI$^2$D. Specifically, we first reveal that the semantic latent space of pre-trained diffusion models can represent the content and haze characteristics of hazy images, as the diffusion time-step changes. Building upon this insight, we integrate the diffusion latent representations at different time-steps into a delicately designed dehazing network to provide instructions for image dehazing. Our DiffLI$^2$D avoids re-training diffusion models and iterative sampling process by effectively utilizing the informative representations derived from the pre-trained diffusion models, which also offers a novel perspective for introducing diffusion models to image dehazing. Extensive experiments on multiple datasets demonstrate that the proposed method achieves superior performance to existing image dehazing methods. Code is available at https://github.com/aaaasan111/difflid.

cs.CV↗

arXiv2025

CultureScope: A Dimensional Lens for Probing Cultural Understanding in LLMs

As large language models (LLMs) are increasingly deployed in diverse cultural environments, evaluating their cultural understanding capability has become essential for ensuring trustworthy and culturally aligned applications. However, most existing benchmarks lack comprehensiveness and are challenging to scale and adapt across different cultural contexts, because their frameworks often lack guidance from well-established cultural theories and tend to rely on expert-driven manual annotations. To address these issues, we propose CultureScope, the most comprehensive evaluation framework to date for assessing cultural understanding in LLMs. Inspired by the cultural iceberg theory, we design a novel dimensional schema for cultural knowledge classification, comprising 3 layers and 140 dimensions, which guides the automated construction of culture-specific knowledge bases and corresponding evaluation datasets for any given languages and cultures. Experimental results demonstrate that our method can effectively evaluate cultural understanding. They also reveal that existing large language models lack comprehensive cultural competence, and merely incorporating multilingual data does not necessarily enhance cultural understanding. All code and data files are available at https://github.com/HoganZinger/Culture

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↗