Searcharxiv⌕ Search

arXiv subjects

Wei Xu

Publications and source records attributed to Wei Xu.

At least 19 recordsLinked to original sources
arXiv2026

Non-asymptotic Sphere-Packing Bounds for Finite-Blocklength Scalar Gaussian Multiple-Access and Broadcast Channels

This paper investigates non-asymptotic achievable rate regions of Gaussian multiple-access channels (MACs) and broadcast channels (BCs) in the finite-blocklength regime. By employing a proposed volume-uniform random coding (VURC) codebook that distributes codewords uniformly within an $n$-dimensional power sphere, together with successive interference cancellation (SIC) decoding, we derive a closed-form achievability bound, a scheme-specific lower bound on the average error probability, and a fourth-order asymptotic characterization. The proposed SIC achievability bound admits a simple and analytically interpretable closed-form expression, enabling direct performance evaluation without iterative computation. The structure of the proposed SIC asymptotic bound offers a highly extensible framework, allowing advances in single-user finite-blocklength theory to be systematically incorporated into the multiuser setting. Futhermore, to characterize the performance benefit of joint decoding, we further derive numerical and closed-form achievability bounds for the two-user Gaussian MAC under joint nearest-neighbor (JNN) decoding. Numerical results illustrate the performance differences among the proposed SIC and JNN bounds and existing benchmarks.

cs.IT↗
arXiv2026

Functional Critics for Off-Policy Actor-Critic: From Stability to Exploration

Actor-critic methods are central to modern reinforcement learning, yet off-policy actor-critic theory remains far removed from practical algorithms, due in part to the interaction between the deadly triad and the continually changing target policy. We revisit functional critics, i.e., value functions that explicitly condition on the policy, and show that this representation provides a shared foundation for off-policy training stability and policy-space exploration. We first develop a convergent off-policy actor-critic framework under linear function approximation that accommodates target-based TD, partial coverage, and evolving behavior policies. Our analysis combines a policy-evaluation guarantee under partial coverage with policy-gradient generalization of functional critics to yield a dual trust-coverage framework that controls policy-gradient error under evolving behavior policies while enabling greater reuse of off-policy data. We further develop a replay-buffer analysis under i.i.d. sampling, yielding a principled replay-sampling design for minimizing the replay-sample complexity required to reach a prescribed evaluation accuracy. We next show that functional critics provide a natural representation for model-free posterior-sampling-style exploration. Standard critic ensembles do not naturally represent joint uncertainty over return functions across policy space, whereas functional-critic ensembles represent policy-conditioned return functions directly. This motivates a particle-based, model-free analogue of posterior-sampling-style exploration without explicit model learning. Finally, we instantiate these ideas in a deep functional actor-critic algorithm that, despite omitting several standard off-policy AC heuristics, achieves strong performance across the full Dog and Humanoid task series of the DeepMind Control Suite.

cs.LG↗
arXiv2026

Local Limit Theorems for Heavy-Tailed Lévy Processes with Negative Drift

This paper establishes uniform local asymptotics for the first-passage time below zero and the stopped maximum of a Lévy process with negative drift and a locally regularly varying Lévy measure. We show that their probabilities of lying in distant intervals are asymptotic to the expected first-passage time multiplied by the corresponding Lévy masses, uniformly over interval widths bounded away from zero, including unbounded intervals. Although the two local events have asymptotically equivalent probabilities, their conditional overlap vanishes for fixed-width intervals. Nevertheless, we establish conditional concentration of the first-passage time and stopped maximum around their deterministic linear relation on scales dominating the fluctuations of the centered process. We further prove conditional functional limit theorems yielding a single-step process with random jump time on the original time scale and a deterministic linear descent on the accelerated scale. Finally, we identify the joint limiting distribution of the unscaled pre-passage position and terminal overshoot, distinguishing continuous passage from passage by a downward jump.

math.PR↗
arXiv2026

Less is more: learning what to profile for GPU kernel optimization

Large language models have recently shown strong potential for automatically optimizing GPU kernels, yet their effectiveness depends critically on the hardware information available to the kernel optimization model. Existing systems typically rely on predefined profiling strategies, leaving what to profile largely outside the optimization process. We show that effective profiling is not simply about collecting more information: increasing the number of profiling metrics does not monotonically improve kernel optimization, while the most useful profiling information varies across kernels. Our theoretical analysis further shows that, under such heterogeneous profiling utility, kernel-conditioned selection can outperform any single fixed profiling configuration. Motivated by this observation, we propose AutoProfiler, a feedback-driven framework for learning a deterministic profiling tree from optimization feedback. The tree is iteratively refined through REWRITE and SPLIT operations, with an LLM generating candidate modifications and empirical evaluation determining their acceptance. Once constructed, the tree quickly gathers the required profiling information for any kernel, allowing the LLM rewriter to be invoked only once. Experiments on KernelBench show that our approach improves kernel optimization over fixed profiling configurations under constrained profiling budgets. These results suggest that hardware profiling should evolve from static metric collection toward profiling reasoning, making profiling itself a new dimension for improving kernel optimization. The codes are available at https://github.com/Satissss/AutoProfiler.

cs.LG↗
arXiv2026

Investigating and Alleviating Harm Amplification in LLM Interactions

Large language models (LLMs) can serve as helpful assistants, yet they can equally function as harm amplifiers that enable malicious users to achieve harmful outcomes beyond their capabilities through extended interactions. This risk manifests along two axes, i.e., democratizing domain expertise that allows novices to produce specialized harmful content, and scaling harmful operations at volumes that manual effort cannot match. Existing works, however, often overlook how LLMs compound harm across multi-turn conversations. We introduce HarmAmp, a new benchmark for multi-turn harm amplification scenarios spanning twelve risk categories. Each scenario is grounded in real-world threats and satisfies rigorous criteria, i.e., substantive amplification, operational specificity, and multi-turn necessity. We further propose TrajSafe, a proactive monitor that anticipates harmful trajectories and intervenes through actions such as probing users' genuine intents and steering the models towards safer completion. Our extensive experiments demonstrate that TrajSafe significantly reduces the harmfulness incurred in multi-turn interactions while preserving a low over-refusal rate and the target model's general capabilities. Our work offers a promising paradigm to alleviate the nuanced safety risks in LLM interactions.

cs.CL↗
arXiv2026

Parallel Universes for Temporal Evaluation and Knowledge Updating in LLMs

Large language models (LLMs) rely on static pretraining corpora, resulting in outdated knowledge over time. Existing approaches for evaluating knowledge edits either suffer from rapid contamination or rely on counterfactual edits that conflict with rigid existing knowledge. In this work, we propose a simulation-driven approach for evaluating and training LLMs on knowledge insertions. We introduce PARALLELEVENTS, a benchmark that simulates coherent trajectories of fictional yet realistic future events, enabling controlled evaluation without data contamination or conflicts with existing knowledge. We further propose SYNAPSE, a training framework that leverages model-generated data to integrate new knowledge into model parameters via instruction tuning. This synthetic pipeline enables scalable knowledge integration without costly human-curated data. SYNAPSE improves QA accuracy by 14.23 points over prior methods, indicating that simulation-based synthetic training yields robust, coherent knowledge insertion.

cs.CL↗
arXiv2026

Evaluating DeepSeek V4.1 Flash-1017 for loss of control and harmful manipulation risks

As frontier language models grow increasingly capable and autonomous, independent third-party evaluations are critical for characterizing their systemic safety risks. While closed-source releases are routinely accompanied by comprehensive system cards, open-weight models receive less scrutiny regarding frontier risks. To address this gap, we present an independent safety evaluation of DeepSeek V4.1 Flash-1017, the latest version of V4.1 Flash that departs from V4 with a new architecture, scaled-up training, and strong agentic capabilities. Drawing on the EU General-Purpose AI Code of Practice, we evaluate the model across 34 benchmarks, including 5 in-house suites, spanning evaluation integrity, loss of control, and harmful manipulation. Conforming to the AEF-1 standard, we compare V4.1 Flash-1017 against V4.1 Flash, earlier DeepSeek V4 models, and leading closed-source systems (GPT-6 Astra, GPT-5.6 Sol, and Gemini 3.8 Flash). Overall, V4.1 Flash-1017 shows substantial safety improvements over the V4 generation and approaches or matches leading closed-source baselines on several dimensions, while still exhibiting notable behaviors that warrant continued attention. Building on our findings, we highlight three key observations and recommendations. First, because safety can shift and even regress across minor updates, developers should integrate safety evaluation into incremental releases. Second, as final-turn outputs and end-state checks alone cannot reliably detect loss-of-control behaviors, deployments require process-level observability across both reasoning traces and action logs. Third, because public benchmarks elicit strong evaluation awareness and face data contamination risks, developing private benchmarks is essential for trustworthy frontier evaluation.

cs.CL↗
arXiv2026

Large Excursions of Reflected Lévy Processes: Local Asymptotics

This paper establishes uniform local large-deviation asymptotics for excursions of a Lévy process $X$ with negative drift $-β$ and $α$-locally regularly varying Lévy measure $ν$, reflected at its running infimum. Under the excursion measure $\underline n$, we first prove that the lifetime $ζ$ and height $\overlineε$ satisfy \begin{equation*} \underline{n}\big(ζ\in(t,t+δ] \big) \sim \underline{n}(\overlineε\in\big (βt,βt+βδ]\big) \sim\underline{n}(ζ)\cdotν\big((βt,βt+βδ]\big), \quad \mbox{as } t\to\infty, \end{equation*} uniformly in $δ\in[δ_0,\infty]$ for every $δ_0>0$. In contrast to the asymptotic equivalence of the corresponding tail events, fixed-width lifetime and height windows have asymptotically negligible relative overlap. Nevertheless, for any positive function $θ$ satisfying $|X_t+βt|/θ_t\to0$ in probability, \begin{equation*} \underline n\big( |\overlineε-βt| \leq θ_t \,\big|\, ζ\in (t,t+δ] \big) \sim \underline n\big( |ζ-t| \leq θ_t \,\big|\, \overlineε\in (βt,βt+βδ] \big) \to 1 . \end{equation*} For conditioning windows of sublinear width bounded away from zero, both conditional excursion laws converge under spatial rescaling to a single-jump process with deterministic jump size and random arrival time determined by the integrated excursion survival function. An additional time rescaling yields the same deterministic linear descent killed at zero. Finally, we identify the joint limiting distribution of the unscaled pre-terminal position and terminal overshoot, including continuous and discontinuous termination.

math.PR↗
arXiv2026

A Framework for Learning and Reliability Certification in Neural Polar Decoding

This paper investigates a framework that connects network training, independently certified population risk and sequential reliability in neural polar decoding. Network outputs determine information-bit decisions by their signs. Posterior mean-squared error (MSE) controls bit-error rate (BER), and a first-error identity connects errors evaluated with true preceding bits to sequential block-error rate (BLER). Three methods exploit sequential decoding structure to certify changes to trained receivers. First, distances from trained outputs to the zero decision threshold, termed learned margins, yield reliability bounds for over-parameterized neural network (ONN) updates subject to separate mean-squared output-change constraints at information positions. Second, a shared loss takes the largest block-error indicator difference between candidate decoders and successive cancellation (SC) on each frame. Its expectation bounds the corresponding BLER differences, termed excess BLER, for all fixed combinations of information-position implementations. An exact expression quantifies the gap between this shared bound and the largest BLER difference among fixed decoders. Third, verified linear programming (LP) and candidate-set subdivision tighten these guarantees with controllable computation at fixed confidence. ONN experiments assess conditional fitting and weight updates. Polar factor-graph experiments demonstrate reusable combination guarantees and storage reduction under prescribed excess-BLER allowances. Component timing identifies the dominant graph-propagation cost.

eess.SP↗
arXiv2026

Spark-Audio 1.0 Technical Report

This paper introduces Spark-Audio 1.0, a speech foundation model trained on domestically produced computing infrastructure. The model accepts text and speech inputs and generates text outputs. It supports multilingual (99 languages) and multi-dialect (202 dialects) speech recognition, speech translation, audio perception and understanding, and both text- and speech-based dialogue, marking a shift from merely ``hearing clearly'' to genuinely ``understanding'' speech. Architecturally, Spark-Audio 1.0 couples a 0.65B-parameter dense audio encoder with a 30B-A3B Mixture-of-Experts (MoE) large language model (LLM). The audio encoder is pretrained with representation-enhanced BEST-RQ and then refined through multi-task supervised training. The audio--LLM is trained through staged pretraining, followed by post-training that combines supervised fine-tuning with on-policy distillation. In total, the model is trained on 13 million hours of audio data together with large-scale text data. We evaluate Spark-Audio 1.0 on mainstream audio and text benchmarks, as well as on in-house test sets drawn from real-world application scenarios. Despite using substantially less audio training data than Qwen3.5-Omni-Flash, Spark-Audio 1.0 achieves comparable performance on most audio and text tasks, outperforms it in multilingual speech recognition, spoken dialogue, and real-world application scenarios, and approaches the performance of larger closed-source models. Nevertheless, there remains room for improvement in instruction following and complex audio reasoning. Finally, the successful training of Spark-Audio 1.0 on domestic hardware, including Huawei Ascend 910B and Hygon BW1000 chips, demonstrates the feasibility of a domestic technical pathway for building speech foundation models.

eess.AS↗
arXiv2026

Budgeted Attack Scheduling for LLM Red-Teaming: The Roles of Static Priors, Semantic Guidance, and Online Feedback

As jailbreak attacks against large language models (LLMs) become increasingly diverse, red-teaming has come to rely on large libraries of jailbreak prompts and attack strategies to evaluate the robustness of victim models and their defense settings. However, attack effectiveness is highly condition-dependent: an attack that succeeds under one victim-defense setting may fail under another, while exhaustively evaluating the full attack library for each setting incurs substantial token and monetary costs. This motivates studying how jailbreak attacks should be prioritized under a limited attempt budget for efficient robustness evaluation. In this work, we formalize this problem as budgeted attack scheduling over a fixed attack library and investigate three complementary signals under a fixed victim-defense condition: (1) static historical attack success rate (ASR) priors; (2) semantic guidance from LLMs; and (3) online feedback from earlier queries. Building on these signals, we design four scheduling strategies and evaluate them across three victim models and three defense settings on GPT-5.2. Extensive experiments clearly characterize the conditions under which each signal is most effective. Historical ASR provides a strong initialization, but its reliability degrades under defense-induced distribution shifts. Semantic guidance consistently improves or matches static prior-based scheduling in the cold-start regime, while online feedback becomes particularly valuable when transferred priors become stale. Semantic guidance and online feedback can also partially compensate for missing historical priors. Taken together, these findings characterize budgeted red-teaming as an attack scheduling problem in which historical priors, semantic guidance, and online feedback play complementary roles.

cs.CR↗
arXiv2026

Grouped Sparse Bayesian Estimation for Near-Field Massive MIMO Channels

This paper studies near-field massive MIMO channel estimation based on an electromagnetic channel model with limited pilots. Starting from Maxwell's equations, we derive the homogeneous background response and the scattering response in the spatial and wavenumber domains. It is found that a compact scattering cluster produces a localized receive--transmit transverse-wavenumber block across finite apertures, thereby yields a block-sparse NLOS channel coefficient matrix in the wavenumber domain. Considering the spatial non-stationarity of large-scale antenna arrays, we introduce visibility factors at the transmit and receive sides of the channel model. To solve the non-stationary sparse channel estimation problem in the wavenumber domain, we propose a visibility-aware block-sparse Bayesian channel estimation algorithm. Simulation results demonstrate that the proposed method achieves lower channel-estimation NMSE than benchmark schemes.

eess.SP↗
arXiv2026

Robust Design of Sensing-Assisted Secure Transmission for RIS-Aided ISAC System under Eavesdropper 3D Positioning Uncertainty

Localization errors in estimating the position of a potential eavesdropper (Eve) pose a critical challenge to secure transmission in reconfigurable intelligent surface (RIS)-assisted integrated sensing and communication (ISAC) systems. To address this issue, this paper proposes a sensing-assisted robust secure transmission framework that exploits ISAC sensing to characterize the positioning uncertainty induced by these errors. The RIS is partitioned into a reflection region and a sensing reception region, supporting sensing and communication through spatial separation. By jointly exploiting observations collected at the base station (BS) and the RIS sensing reception region, maximum likelihood estimation (MLE) is employed to estimate the arrival angles, from which Eve's three-dimensional (3D) position is obtained through cross localization. The resulting localization Cramer-Rao lower bound (CRLB) is then used to construct an ellipsoidal uncertainty region around the estimated position. A two-stage optimization strategy is developed. In Stage I, the sensing beamformer and RIS reflection coefficients are jointly optimized to improve localization accuracy. In Stage II, the resulting positioning uncertainty is incorporated into a robust secure transmission design, where the communication beamformer and RIS reflection coefficients are jointly optimized to maximize the worst-case secrecy rate. Simulation results demonstrate that the proposed approach outperforms benchmark schemes without RIS assistance and with random RIS configurations in terms of secrecy performance. The proposed framework effectively mitigates secrecy performance degradation caused by localization errors in Eve's position.

eess.SP↗
arXiv2026

Beyond Expert Rankings: Contrast Calibration for Mixed-Precision MoE Quantization

Mixture-of-experts (MoE) language models activate only a few experts per token, yet all expert weights must be stored. Mixed-precision quantization reduces this storage cost by assigning higher bit-widths to sensitive experts and lower bit-widths to resilient ones, but expert rankings alone do not determine how strongly to weight their errors. To address this gap, we propose NatScale, a contrast-calibrated allocation framework for mixed-precision MoE quantization. It uses a model-specific power transformation to calibrate gradient-based expert weights to an information-theoretic contrast target while preserving rankings. We jointly evaluate each expert's gate, up and down configurations to estimate its output error. We then combine these errors and calibrated weights with an existing prior based on weight spectra, variance and precision decay to select configurations under a storage budget. Across three models, NatScale achieves the lowest native-to-quantized KL divergence and the highest top-token agreement in all nine settings against five allocation baselines. It also yields the lowest perplexity in seven settings. These findings highlight the importance of objective-aware precision allocation.

cs.LG↗
arXiv2026

SiMDex: Mining Similar Egocentric Videos for Cross-Embodiment Dexterous Manipulation

Recent years have witnessed an explosive trend of scaling ego-centric human videos for robot manipulation, yet it remains unclear which data actually benefits dexterous manipulation. We present SiMDex, a similarity-based data mining framework that casts human data selection for VLA post-training in dexterous manipulation as a recommendation problem. For each robot demonstration, SiMDex runs a three-stage recall-ranking-re-ranking cascade over ~32M egocentric human samples: recall and ranking select a task-relevant subset in a morphology-agnostic action space, and re-ranking verifies it on pixels, with no change to VLA architecture or training. Against a strong baseline trained with an equal amount of randomly sampled human data, SiMDex uses only ~1.49M mined samples (<5% of the pool) yet improves the overall success rate from 47.7% to 61.7%, showing that selective curation outperforms indiscriminate data mixing.

cs.RO↗
arXiv2026

Manifold4D: Denoising on Point Cloud Rendered Manifolds for Video Re-shooting

Video re-shooting is the task of re-rendering a monocular video of a dynamic scene along a user-specified camera trajectory. A dominant paradigm explicitly provides the target geometry by lifting the source video into a 4D point cloud using per-frame depth, and rasterizing this representation along the target trajectory to obtain point-cloud renders as geometric conditions for trajectory-controlled generation. The point cloud render and source video may provide conflicting visual cues throughout denoising, as both are directly used as conditions. We refer to this conflict as a trust dilemma, which can compromise trajectory control, 3D consistency of dynamic content, and visual quality, particularly under distribution shifts. We argue that because a point-cloud render is pixel-aligned with the target view, it does not need to be continuously provided as a conditioning signal throughout denoising. We propose Manifold4D, which injects the render directly into the initial noise of flow matching, so that generation no longer departs from standard Gaussian noise but from a geometry-bearing noise distribution that we term the point cloud rendered manifold, leaving the source video as the only visual condition. The render is thus used exactly once, and the network is never asked to learn how to read it; in subsequent denoising steps the model can focus on the source video. In our experiments, Manifold4D attains the best camera-control accuracy on every metric, lowering rotation error by 25--27% and translation error by up to 32% over the strongest baseline, while matching it in video fidelity and attaining the best overall photometric quality and 3D consistency of dynamic content.

cs.CV↗
arXiv2026

Humanize MLE Agents: Multi-Agent Session Alternation Beyond Single-Agent Limits

Existing LLM-based agent approaches for long-horizon machine learning engineering (MLE) often use context engineering, implemented through custom harnesses that manage tools, skills, and accumulated experience. However, we find that as single-session execution lengthens, the agent's reasoning intensity declines alongside slowing progress, summarized as long-horizon deceleration, and the benefits of these custom harnesses vanish as the underlying model improves, summarized as harness staleness. These findings suggest that their benefits may not persist across versions, which motivates a workflow-level approach in which agents build on one another's work across sessions. We model these alternating agent sessions as a context-reset semi-Markov decision process (CR-SMDP) and establish sufficient conditions for alternation to exceed single-agent reference limits, derive an explicit experiment-budget bound, and characterize when session caps preserve this advantage. Guided by this analysis, we introduce Humanize MLE Agents (HMA), which alternates two agents on a shared workspace, limits each session to at most k experiments and resets the context between sessions, without additional task-specific harness or prior knowledge supplements. Across 75 Kaggle tasks in MLE-bench, HMA reaches a state-of-the-art any-medal rate of 78.2%, exceeding the strongest open-source baseline by 8.0 points with a quarter of its time budget. Moreover, among six configurations spanning five heterogeneous agent pairs, HMA improves any-medal rate over the constituent baselines by 6.4 points on average, demonstrating cross-pair generality. A supplementary case on kernel optimization reaches 878 cycles, within 1.6% of the community leaderboard leader, demonstrating applicability beyond MLE.

cs.AI↗
arXiv2026

Training User Simulators with Latent Intrinsic Feedback

User simulators are gaining attention as environments to train LLM assistants that can better cope with the nuances of multi-turn conversations with real users. Prior work on user simulation has emphasized generating realistic user utterances, leaving unexplored their potential as a source of direct feedback on the assistant's responses. Such evaluation signals may not be explicitly communicated in language, but can instead reside in the latent reasoning of users. In this work, we introduce user simulators with intrinsic feedback, namely, models equipped with latent evaluation states: internal thoughts, reactions, and scalar ratings that evaluate the assistant from a first-person user perspective at each conversation turn. We train LLM-based simulators using interaction logs between real users and assistants and demonstrate that intrinsic feedback improves simulation realism and critiquing behaviors versus prompted simulator baselines, while also providing turn-level evaluations that existing fine-tuned simulators lack. Further, evaluations generated by our trained simulators correlate better with human judgments than those of prompted simulators of the same size. Lastly, our experiments reveal that simulators with intrinsic feedback provide effective reward signals for multi-turn RL training, matching setups that rely on external judge LLMs for rewards while cutting RL compute by more than half.

cs.CL↗