arXiv subjects

Zeyu Yang

Publications and source records attributed to Zeyu Yang.

At least 19 recordsLinked to original sources

arXiv2026

When to Use Extra Context: Evidence-Grounded Terminology Adaptation for Simultaneous Speech Translation

Extra context is valuable for simultaneous speech translation of technical talks, but injecting the entire document context into every streaming segment is often too coarse. Through diagnostic experiments, we find that context gains mainly come from paper-specific terminology recovery rather than uniform semantic enhancement. We therefore propose EGTA, an Evidence-Grounded Terminology Adaptation framework that builds a document terminology memory, selects compact candidate terms conditioned on the current streaming state, and adapts ASR/speech-side and decoder-side decision spaces using only the selected terms. EGTA can be instantiated in cascaded, end-to-end, and generation-only SimulST settings without full-model fine-tuning. We evaluate EGTA on an ACL technical-talk SimulST evaluation suite consisting of MCIF-dev and ACL60/60-dev. On MCIF-dev, EGTA-RG improves BLEU by +1.05/+0.59, XCOMET-XL by +0.019/+0.006, named-entity recall by +79\%/+73\% relative, and acronym recall by +0.099/+0.171 on En$\rightarrow$Zh and En$\rightarrow$De. Across MCIF-dev latency settings, EGTA consistently improves XCOMET-XL, named-entity recall, and acronym recall. External validation on ACL60/60-dev further shows consistent terminology-recall gains without additional fine-tuning. Shuffled-memory controls and activation audits provide evidence that the improvements are tied to paper-specific evidence alignment rather than generic context prompting.

cs.CL↗

arXiv2026

Safety-Regulated Transfer Reinforcement Learning with Adaptive Teacher Guidance

We propose Safety-Regulated Adaptive Transfer Reinforcement Learning (SRATRL), a teacher--student framework that combines safety-triggered intervention, safety-adaptive value shaping, and policy-compatibility-based optimization for efficient target-domain adaptation. First, a safety-triggered closed-loop intervention strategy is developed that activates teacher guidance according to the instantaneous safety cost and adaptively adjusts the intervention threshold based on the student policy's recent safety performance, thereby providing timely safety supervision while progressively restoring student autonomy as its safety improves. Next, a safety-adaptive teacher-guided value-shaping scheme is introduced, in which a teacher-consistency signal is incorporated into the critic target, and its contribution is dynamically regulated by the safety-constraint multiplier, enabling stronger teacher guidance under elevated safety risks and gradually weakening such guidance as the safety constraint is better satisfied. In addition, a teacher-student policy-compatibility weighting approach is proposed to alleviate the adverse optimization effects caused by policy mismatch. It reweights teacher-intervened transitions according to the relative likelihood of the executed action under the teacher and student policies, thereby improving policy-update stability. Experimental results demonstrate that compared with a Proximal Policy Optimization with Lagrangian constraint baseline, the proposed method improves the average velocity by 6.90%, and reduces the crash ratio by 75.00%. These results demonstrate that the proposed method can reduce safety costs while maintaining competitive task efficiency.

cs.LG↗

arXiv2026

ConsisGuard: Aligning Safety Deliberation with Policy Enforcement in LLM Guardrails

Reasoning-based LLM guardrails improve safety moderation by generating explicit rationales before issuing final decisions. However, their rationales do not always lead to faithful enforcement: a model may recognize a harmful intent in its reasoning but still predict a safe label, or issue an unsafe decision without policy-grounded justification. We identify this safety-critical failure mode as the deliberation-to-enforcement gap. Unlike general chain-of-thought faithfulness, guardrail reliability requires policy execution consistency: the generated reasoning should be grounded in the safety policy, and the final decision should be entailed by that reasoning. We propose ConsisGuard, a consistency-aware framework for reasoning-based LLM guardrails. ConsisGuard performs Policy-to-Decision Trajectory Distillation and Functional Coupling Alignment, aligning the internal coupling between safety deliberation and decision enforcement. Experiments on prompt and response harmfulness detection benchmarks show that ConsisGuard improves detection performance while reducing policy execution failures. These results suggest that reliable reasoning-based guardrails require accurate faithful execution of safety policies.

cs.CL↗

arXiv2026

Make LLM Learn to Synthesize from Streaming Experiences through Feedback

Large language models (LLMs) have been widely adopted for synthetic data generation, significantly reducing annotation costs. However, most existing studies treat synthesis as a set of isolated tasks and overlook a more fundamental question: whether a model can learn to synthesize by accumulating experience from past tasks and transferring it to future ones. In this work, we introduce StreamSynth, a new setting in which synthesis tasks arrive sequentially and experience from historical tasks provides informative signals for future synthesis. To address this setting, we propose SynLearner, a general framework that enables synthesis models to acquire reusable synthesis experience over a task stream. Instead of generating data independently for each task, SynLearner encourages the model to explore diverse synthesis patterns, learn from feedback, and balance sample quality with set-level diversity as tasks evolve. Extensive experiments across multiple benchmarks show that SynLearner effectively leverages experience from earlier tasks to improve synthesis performance on later ones, exhibiting consistent cross-task transferability. These findings provide evidence for the feasibility of StreamSynth and highlight synthetic data generation as an experience-driven process that can benefit from task streams.

cs.AI↗

arXiv2026

Robust and Generalizable Safety Steering for Text-to-Image Diffusion Transformers

Diffusion Transformers have become a powerful backbone for text-to-image generation, but their layered and cross-modal generation process makes safety control fundamentally different from prompt-level filtering or output-level detection. Harmful semantics may be weakly expressed in text representations, progressively bound to visual latents, and finally entangled with rendering dynamics. As a result, safety steering at a fixed layer can be unstable, and a steering mechanism learned from known risks may not transfer reliably to a shifted target risk domain. We propose SafeDIG, a safety steering framework that formulates DiT safety adaptation as position-aware sparse feature transfer. SafeDIG first constructs Sparse Autoencoders over functionally distinct DiT intervention positions and uses robustness-aware pre-training routing to prioritize intervention sites that are expected to remain stable under source-target risk shift. It then separates transferable safety features from domain-specific activation geometry by freezing the SAE encoder as a reusable sparse safety dictionary and adapting only the decoder to the target-domain activation manifold. During inference, SafeDIG combines Blend and Repel operations to steer unsafe activations toward transferred safety manifolds or away from harmful sparse directions. Experiments on FLUX.1 Dev and Stable Diffusion 3.5 Large show that SafeDIG consistently reduces target-domain and overall unsafe generation rates while preserving source-domain safety and image quality.

cs.AI↗

arXiv2026

Runtime-Orchestrated Second-Order Optimization for Scalable LLM Training

Second-order methods offer an attractive path toward more sample-efficient LLM training, but their practical use is often blocked by the systems cost of maintaining and updating large matrix-based optimizer states. We introduce \textbf{Asteria}, a runtime system designed to remove this bottleneck by separating second-order optimization logic from the critical GPU training path. Rather than keeping all preconditioner state on the accelerator, Asteria dynamically distributes optimizer state across GPU memory, CPU memory, and optional NVMe storage according to architectural constraints and runtime pressure. It further uses training hooks to prepare shadow states in advance, allowing expensive inverse-root computations to proceed asynchronously on the host while GPU computation continues. For distributed training, Asteria employs a bounded-staleness protocol that limits synchronization frequency while preserving optimizer effectiveness through topology-aware coordination. We evaluate Asteria on both memory-constrained and distributed training settings. On a DGX Spark platform with a single GB10 GPU and 128GB unified memory, Asteria supports second-order training for a 1B-parameter language model. On multi-node GH200 systems, it lowers visible optimizer overhead, reduces recurring latency spikes, accelerates convergence in wall-clock time, and maintains the optimization advantages of SOAP and KL-Shampoo in a 7B-parameter language model. Our results suggest that second-order LLM training can be made practical not by simplifying the optimizer alone, but by rethinking how optimizer state, background computation, and distributed synchronization are managed at the runtime level.

cs.DC↗

arXiv2026

Superintelligent Retrieval Agent: The Next Frontier of Agentic Retrieval

Retrieval-augmented agents are increasingly the interface to large knowledge bases, yet most treat retrieval as a black box: they issue exploratory queries, inspect snippets, and reformulate until evidence emerges. This resembles how a newcomer searches an unfamiliar database rather than how an expert navigates it with strong priors about terminology and likely evidence, causing extra retrieval rounds, latency, and poor recall. We introduce \textit{Superintelligent Retrieval Agent} (SIRA), which casts \emph{superintelligence} in retrieval as compressing multi-round exploratory search into a single corpus-discriminative retrieval action. SIRA does not merely ask which terms are relevant; it asks which terms separate the desired evidence from corpus-level confusers. Offline, an LLM enriches each document with missing search vocabulary; at query time, it predicts evidence vocabulary the query omits; and corpus statistics serve as tool calls that filter terms that are absent, overly common, or unlikely to create retrieval margin. The final step is a single weighted BM25 call combining the query with the validated expansion. Across ten BEIR benchmarks, SIRA achieves the strongest average retrieval performance in our comparison, beating dense retrievers, learned sparse retrievers, and LLM search-agent baselines while using no relevance labels or retriever fine-tuning. On downstream QA, its retrieval-only answer coverage exceeds recent RL-trained agentic QA systems on NQ and HotpotQA. We also introduce \textbf{BrowseComp-Wikipedia}, a hard-search benchmark of 232 BrowseComp-derived queries over a 25,587,229-document Wikipedia index. Even without index-time enrichment, using only grounded Wikipedia categories, SIRA outperforms multi-round Perplexity agents at every budget, reaching 9.70% Recall@1, 15.27% Recall@10, and 36.14% Recall@100.

cs.IR↗

arXiv2026

Regularity of the Value Function in Discounted Infinite-Time Mean Field Games

In [17], we introduced the discounted infinite-time mean field games. Subsequently, in [18], we studied the connection between infinite-time mean field FBSDEs and elliptic master equations. In this paper, we further investigate the regularity of the representative player's value function. Specifically, we first prove the strong existence and uniqueness, as well as the uniqueness in law, for an extended class of infinite-time FBSDEs. We then establish the Lions-differentiability for the derivative of the representative player's value function with respect to the measure argument, and provide an explicit characterization for it using solutions to FBSDEs.

math.PR↗

arXiv2026

Slack More, Predict Better: Proximal Relaxation for Probabilistic Latent Variable Model-based Soft Sensors

Nonlinear Probabilistic Latent Variable Models (NPLVMs) are a cornerstone of soft sensor modeling due to their capacity for uncertainty delineation. However, conventional NPLVMs are trained using amortized variational inference, where neural networks parameterize the variational posterior. While facilitating model implementation, this parameterization converts the distributional optimization problem within an infinite-dimensional function space to parameter optimization within a finite-dimensional parameter space, which introduces an approximation error gap, thereby degrading soft sensor modeling accuracy. To alleviate this issue, we introduce KProxNPLVM, a novel NPLVM that pivots to relaxing the objective itself and improving the NPLVM's performance. Specifically, we first prove the approximation error induced by the conventional approach. Based on this, we design the Wasserstein distance as the proximal operator to relax the learning objective, yielding a new variational inference strategy derived from solving this relaxed optimization problem. Based on this foundation, we provide a rigorous derivation of KProxNPLVM's optimization implementation, prove the convergence of our algorithm can finally sidestep the approximation error, and propose the KProxNPLVM by summarizing the abovementioned content. Finally, extensive experiments on synthetic and real-world industrial datasets are conducted to demonstrate the efficacy of the proposed KProxNPLVM.

cs.LG↗

arXiv2026

Scout Before You Attend: Sketch-and-Walk Sparse Attention for Efficient LLM Inference

Self-attention dominates the computational and memory cost of long-context LLM inference across both prefill and decode phases. To address this challenge, we introduce Sketch&Walk Attention, a training-free sparse attention method that determines sparsity with lightweight sketches and deterministic walk. Sketch&Walk applies Hadamard sketching to get inexpensive approximations of attention scores, then aggregates these estimates across layers via a walk mechanism that captures attention influence beyond direct interactions between tokens. The accumulated walk scores are used to select top-k attention blocks, enabling dynamic sparsity with a single training-free algorithm that applies uniformly to both the prefill and decode phases, together with custom sparse attention kernels. Across a wide range of models and tasks, Sketch&Walk maintains near-lossless accuracy at 20% attention density and can slightly outperform dense attention in some settings, while achieving up to 6x inference speedup.

cs.LG↗

arXiv2026

Redefining Machine Simultaneous Interpretation: From Incremental Translation to Human-Like Strategies

Simultaneous Machine Translation (SiMT) requires high-quality translations under strict real-time constraints, which traditional policies with only READ/WRITE actions cannot fully address. We extend the action space of SiMT with four adaptive actions: Sentence_Cut, Drop, Partial_Summarization and Pronominalization, which enable real-time restructuring, omission, and simplification while preserving semantic fidelity. We adapt these actions in a large language model (LLM) framework and construct training references through action-aware prompting. To evaluate both quality and word-level monotonicity, we further develop a latency-aware TTS pipeline that maps textual outputs to speech with realistic timing. Experiments on the ACL60/60 English-Chinese, English-German and English-Japanese benchmarks show that our framework consistently improves semantic metrics and achieves lower delay compared to reference translations and salami-based baselines. Notably, combining Drop and Sentence_Cut leads to consistent improvements in the balance between fluency and latency. These results demonstrate that enriching the action space of LLM-based SiMT provides a promising direction for bridging the gap between human and machine interpretation.

cs.CL↗

arXiv2025

StrengthSense: A Dataset of IMU Signals Capturing Everyday Strength-Demanding Activities

Tracking strength-demanding activities with wearable sensors like IMUs is crucial for monitoring muscular strength, endurance, and power. However, there is a lack of comprehensive datasets capturing these activities. To fill this gap, we introduce \textit{StrengthSense}, an open dataset that encompasses IMU signals capturing 11 strength-demanding activities, such as sit-to-stand, climbing stairs, and mopping. For comparative purposes, the dataset also includes 2 non-strength demanding activities. The dataset was collected from 29 healthy subjects utilizing 10 IMUs placed on limbs and the torso, and was annotated using video recordings as references. This paper provides a comprehensive overview of the data collection, pre-processing, and technical validation. We conducted a comparative analysis between the joint angles estimated by IMUs and those directly extracted from video to verify the accuracy and reliability of the sensor data. Researchers and developers can utilize \textit{StrengthSense} to advance the development of human activity recognition algorithms, create fitness and health monitoring applications, and more.

cs.CV↗

arXiv2025

Cross-Platform Short-Video Diplomacy: Topic and Sentiment Analysis of China-US Relations on Douyin and TikTok

We examine discussions surrounding China-U.S. relations on the Chinese and American social media platforms \textit{Douyin} and \textit{TikTok}. Both platforms, owned by \textit{ByteDance}, operate under different regulatory and cultural environments, providing a unique perspective for analyzing China-U.S. public discourse. This study analyzed 4,040 videos and 338,209 user comments to assess the public discussions and sentiments on social media regarding China-U.S. relations. Through topic clustering and sentiment analysis, we identified key themes, including economic strength, technological and industrial interdependence, cultural cognition and value pursuits, and responses to global challenges. There are significant emotional differences between China and the US on various themes. Since April 2022, the Chinese government has implemented a new regulation requiring all social media accounts to disclose their provincial-level geolocation information. Utilizing this publicly available data, along with factors such as GDP per capita, minority index, and internet penetration rate, we investigate the changes in sentiment towards the U.S. in mainland China. This study links socioeconomic indicators with online discussions, deeply analyzing how regional and economic factors influence Chinese comments on their views of the US, providing important insights for China-U.S. relationship research and policy making.

cs.SI↗

arXiv2025

DPO-Tuned Large Language Models for Segmentation in Simultaneous Speech Translation

Simultaneous speech translation requires accurate segmentation to balance translation quality and latency. Recent studies such as SHAS have introduced pretrained segmentation models, achieving stronger performance than heuristic rules. However, segmentation models such as SHAS, though pretrained and more robust than heuristic methods, are still constrained by supervised learning objectives and do not incorporate human preference alignment, which is crucial for natural real-time interpretation. In this work, we propose a segmentation framework based on large language models (LLMs) trained with Direct Preference Optimization (DPO). By leveraging preference alignment, our method enables LLMs to predict natural segmentation points that better meet the demands of real-time translation. We evaluate the system on the ACL 60/60 corpus across three language pairs (English-Japanese, Chinese, German), using SeamlessM4T v2 as the translation backbone. Experimental results show that our DPO-tuned LLM achieves higher segmentation accuracy than SHAS and yields consistent improvements in translation quality (BLEU, COMET) as well as latency (Average Lagging). Furthermore, our system benefits from IWSLT baselines for direct comparison. These findings highlight the potential of preference-tuned LLMs to surpass existing pretrained segmentation models and advance adaptive, human-aligned simultaneous interpretation.

cs.CL↗

arXiv2025

Infinite-Time Mean Field FBSDEs and the Associated Elliptic Master Equations

This paper presents a further investigation of the properties of infinite-time mean field forward-backward stochastic differential equations (FBSDEs) and the associated elliptic master equations, which were introduced in [18] as mathematical tools for solving discounted infinite-time mean field games. By establishing the continuous dependence of the FBSDE solutions on their initial values, we prove the flow property of the mean field FBSDEs. And then, we prove that, at the Nash equilibrium, the value function of the representative player constitutes a viscosity solution to the corresponding elliptic master equation. In particular, when the coefficients of the equations are distribution-independent, we construct a classical solution to the elliptic partial differential equation (PDE) via fully coupled infinite-time FBSDEs. Furthermore, for classical solutions possessing displacement monotonicity and certain growth conditions, we establish their uniqueness for the elliptic master equation.

math.PR↗

arXiv2025

To Compress or Not? Pushing the Frontier of Lossless GenAI Model Weights Compression with Exponent Concentration

The scaling of Generative AI (GenAI) models into the hundreds of billions of parameters makes low-precision computation indispensable for efficient deployment. We argue that the fundamental solution lies in developing low-precision floating-point formats, which inherently provide numerical stability, memory savings, and hardware efficiency without dequantization overhead. In this paper, we present a theoretical and empirical study of an exponent concentration phenomenon in GenAI weights: exponents consistently exhibit low entropy across architectures and modalities. We show that this arises naturally from $\alpha$-stable distributions induced by stochastic gradient descent, and we prove tight bounds on the entropy of exponents. Our analysis establishes a theoretical compression limit near FP4.67, which motivates the design of a practical FP8 format. Building on these insights, we propose Exponent-Concentrated FP8 (ECF8), a lossless compression framework with entropy-aware encoding and GPU-optimized decoding. Experiments on LLMs and DiTs up to 671B parameters demonstrate up to 26.9% memory savings and 177.1% throughput acceleration, with perfectly lossless computations, i.e., no deviation in model outputs. Our results establish exponent concentration as a statistical law of trained models and open a principled path for lossless low-precision floating-point design in the FP8 era.

cs.LG↗

arXiv2025

SASST: Leveraging Syntax-Aware Chunking and LLMs for Simultaneous Speech Translation

This work proposes a grammar-based chunking strategy that segments input streams into semantically complete units by parsing dependency relations (e.g., noun phrase boundaries, verb-object structures) and punctuation features. The method ensures chunk coherence and minimizes semantic fragmentation. Building on this mechanism, we present SASST (Syntax-Aware Simultaneous Speech Translation), an end-to-end framework integrating frozen Whisper encoder and decoder-only LLM. The unified architecture dynamically outputs translation tokens or symbols to jointly optimize translation timing and content, with target-side reordering addressing word-order divergence. Experiments on CoVoST2 multilingual corpus En-{De, Zh, Ja} demonstrate significant translation quality improvements across languages and validate the effectiveness of syntactic structures in LLM-driven SimulST systems.

cs.CL↗

arXiv2025

EchoAid: Enhancing Livestream Shopping Accessibility for the DHH Community

Livestream shopping platforms often overlook the accessibility needs of the Deaf and Hard of Hearing (DHH) community, leading to barriers such as information inaccessibility and overload. To tackle these challenges, we developed \textit{EchoAid}, a mobile app designed to improve the livestream shopping experience for DHH users. \textit{EchoAid} utilizes advanced speech-to-text conversion, Rapid Serial Visual Presentation (RSVP) technology, and Large Language Models (LLMs) to simplify the complex information flow in live sales environments. We conducted exploratory studies with eight DHH individuals to identify design needs and iteratively developed the \textit{EchoAid} prototype based on feedback from three participants. We then evaluate the performance of this system in a user study workshop involving 38 DHH participants. Our findings demonstrate the successful design and validation process of \textit{EchoAid}, highlighting its potential to enhance product information extraction, leading to reduced cognitive overload and more engaging and customized shopping experiences for DHH users.

cs.HC↗