Searcharxiv⌕ Search

arXiv subjects

Xin Liu

Publications and source records attributed to Xin Liu.

At least 19 recordsLinked to original sources
arXiv2026

Near-Zero Token Embeddings Silently Stall Training in Gated Sequence Models

Among 291 surveyed open-weight language model repositories (280 distinct embedding tables), 52.6% of untied gated and hybrid checkpoints (20/38) contain reachable input embeddings with norms below $10^{-6}$, compared with 0% of weight-tied models. Such near-zero rows can arise during pretraining when unseen rare or template tokens receive zero input gradients while AdamW weight decay shrinks untied embeddings. In tested Transformer configurations, non-initial positions or a BOS token prevent this failure mode. In gated sequence models (Mamba, Gated DeltaNet, xLSTM), a near-zero token can amplify input-embedding gradients by more than 30 orders of magnitude (up to 35). At zero input, current-token readout factors (SiLU gates or linear queries) vanish, suppressing historical-state readout. Residual RMS values can then fall below the square root of the RMSNorm floor across consecutive layers, causing Pre-RMSNorm to amplify backward gradients by up to $10^3$ per layer. When a gradient entry exceeds $1.84 \times 10^{19}$, FP32 global-norm clipping in PyTorch, Hugging Face Trainer, and FSDP2 can overflow to infinity and zero the current batch's gradients. In Telugu continual pretraining of Nemotron-3-Nano-4B, this discards updates on approximately 20% of steps and leaves triggered rows frozen. We propose two remedies. For released checkpoints, GroupClip decouples input-embedding clipping using a 64-bit or rescaled norm; its rescaled implementation eliminates update loss and heals triggered rows in one step with negligible measured end-to-end overhead. For pretraining, GateFloor ($\mathrm{SiLU}(z) + δ$, $δ\in [0.01, 0.1]$) keeps the gate positive at zero input, reducing near-zero gradients by 28 orders of magnitude in experiments up to 370M parameters, with perplexity comparable to the baseline.

cs.LG↗
arXiv2026

The effect of local stacking fault fluctuations on dislocation glide and the interplay with solid solution strengthening in Al-Mg alloys

Concentrated alloys are characterized by a random arrangement of solute atoms of different sizes, which generates spatially fluctuating displacement and stress fields with non-trivial correlations. This chemical disorder also induces local fluctuations in stacking-fault (SF) energies, giving rise to fluctuating SF forces acting on the partial dislocations bounding the SF. In this work, we investigate the characteristics of the SF noise, including spatial correlations and the impact of such SF energy fluctuations on dislocation glide. Specifically, the gliding behavior of screw, edge, and mixed dislocations in a model concentrated Al-Mg random alloy with varying Mg content is analyzed. The interplay between SF energy fluctuations and solid-solution noise is also discussed. We show that SF forces are spatially correlated, in close analogy with long-range elastic stresses, and exhibit a robust negative correlation along the slip direction. We identify the physical origin of these correlations and demonstrate that they significantly contribute to the critical resolved shear stress for dislocations to glide. In contrast, correlations in other directions are weaker but can still be non-negligible, depending on the underlying interatomic interactions.

cond-mat.mtrl-sci↗
arXiv2026

Chain of Faith: Policy-Internal Rewards for Reasoning Reliance

Chain-of-thought (CoT) explanations can sound plausible while having little influence on a model's final answer. We study four rewards that use the policy's own signals to encourage reliance on generated reasoning: gradient sensitivity, confidence gaps, progressive confidence gains, and reasoning--prompt swaps. We add each reward to group relative policy optimization (GRPO) and evaluate two small instruction-tuned models on GSM8K and StrategyQA using reasoning edits, paraphrases, and prompt hints. Several configurations increase answer reliance to reasoning edits, while effects on accuracy and paraphrase robustness vary. We also observe answer copying at a high reward weight and high reliance in decomposition settings with poor task accuracy. The study compares practical objectives for reasoning reliance, confirms the trade-off between reliance and performance, and proposes four lightweight rewards that can improve reasoning reliance.

cs.AI↗
arXiv2026

Learning structural response operators across varying geometries and loads with global and local geometry encoding

Predicting mechanical responses across complex geometries and varying loading conditions is central to simulation-driven engineering, yet capturing the influence of geometric boundaries remains challenging for data-driven models. Here we propose a Load- and Geometry-adaptive Deep Operator Network (LG-DON) that integrates global geometric representations with local, boundary-aware information to predict structural responses. By distinguishing distances to different boundary categories rather than relying on the nearest boundary alone, the framework captures spatially varying geometric influences that conventional distance-based representations may overlook. Across benchmark problems involving diverse non-convex geometries, LG-DON reduces the mean relative L2 error from 12.17% for a modified MIONet baseline to 7.41%. Controlled ablation studies further demonstrate the advantages of category-resolved boundary distances and show that combining global and local geometric representations outperforms either representation alone in 15 of 18 evaluated configurations. These findings highlight the value of integrating complementary geometric information for accurate operator learning across complex structural domains.

cs.CE↗
arXiv2026

Robust Principal Component Analysis via $\ell_0$ Alternating Minimization

Robust principal component analysis concerns the recovery of a low-rank matrix $Z$ and a sparse matrix $S$ from their observed sum $D = Z + S$. In this paper, we study this recovery problem under an $\ell_0$ sparsity penalty and a rank constraint on the low-rank component. By exploiting the entrywise separable structure of the sparse subproblem, we obtain an exact closed-form solution via hard thresholding with box clipping, avoiding the shrinkage bias of $\ell_1$-based methods. The low-rank subproblem is solved by truncated SVD. We prove block-coordinate convergence of the resulting alternating minimization scheme, establish uniqueness of block minimizers under non-degeneracy and singular value gap conditions, establish local linear convergence under a contraction condition on the Fréchet derivative of the truncated SVD operator restricted to the sparse support, and provide a second-order optimality analysis via the Riemannian Hessian. Experiments on synthetic data, images, and video show that the proposed algorithm outperforms the existing convex and nonconvex algorithms, most notably in support recovery and in the accuracy-efficiency trade-off.

math.OC↗
arXiv2026

Pure annihilation decay $B_c^+ \to ST$ with improved perturbative QCD formalism

We study the pure annihilation decay $B_c^+ \to ST$ for the first time in the standard model with $S$ and $T$ representing the states of the light scalar ($1^3\!P_0$) and tensor ($1^3\!P_2$) correspondingly. Utilizing the assignment $q\bar q$ for light scalars with two-scenarios classification, we calculate the branching ratios of $B_c^+ \to ST$ with the improved perturbative QCD (iPQCD) formalism at leading order in the strong coupling $α_s$. The large branching ratios around $10^{-6}$ within uncertainties in the Cabibbo-Kobayashi-Maskawa-favored channels, such as $B_c^+ \to a_0(1450) a_2(1320)$, $B_c^+ \to a_0(980, 1450)^+ f_2(1270)$, $B_c^+ \to K_0^*(700, 1430) K_2^*(1430)$, and $B_c^+ \to f_0(500, 1370, 1500) a_2(1320)^+$, are obtained in the iPQCD formalism, which would be tested at the Large Hadron Collider-beauty detector in the (near) future. The relevant tests to these theoretical predictions of large branching ratios associated with typical relative ratios could provide important opportunities for understanding the QCD dynamics of the involved modes or light scalars, as well as for exploring the reliability of the iPQCD formalism. Moreover, the direct charge-parity violating asymmetries in the decay $B_c^+ \to ST$ are naturally zero, since only tree-diagram contributions are involved.

hep-ph↗
arXiv2026

Estimation and Inference for One-pass Streaming Function-on-Function Regression

In many applications, functional pairs are collected sequentially from a data stream. Existing function-on-function regression methods are primarily designed for batch data, and bootstrap inference has become one of the most popular approaches. However, this procedure may be infeasible when observations cannot be stored. In this paper, we propose a one-pass functional SGD estimator that updates upon each observation's arrival without data storage, and a weighted rolling validation rule is provided for step size selection for implementation. Further, an online multiplier bootstrap is developed for inference that updates perturbed stochastic gradient paths in parallel, bypassing estimating an infinite-dimensional covariance operator. Asymptotic behaviors are established for the proposed method, and conditional bootstrap validity is justified for the bias-centered pointwise estimation error. Simulation studies and two real data applications confirm the empirical effectiveness of the proposed procedures. To the best of our knowledge, this is the first theoretically justified one-pass framework for bootstrap inference on a bivariate coefficient surface with streaming functional data.

stat.ME↗
arXiv2026

Towards Generalized Image Manipulation Localization via Score-based Model

With the rapid evolution of synthetic media, Image Manipulation Localization (IML) has emerged as a critical component in multimedia forensics for ensuring the integrity of digital content. However, generalization remains a core challenge, as existing discriminative methods typically learn a fixed decision boundary that tends to overfit to specific training artifacts and fails to adapt to unseen manipulation types. To address this, we propose DiffIML, a novel framework that introduces score-based generative modeling to IML. Diverging from the direct estimation of hard boundaries, DiffIML approximates the score function, the gradient of the log-likelihood, to capture the intrinsic geometric topology of mask distributions. This paradigm leverages structural priors to iteratively recover coherent masks from noise, thereby circumventing the brittleness associated with discriminative models. Under this formulation, diffusion models serve as an effective numerical solver for the learned score function.To ensure practicality, we respectively resolve the efficiency and stability bottlenecks of standard diffusion by: (1) utilizing a Lightweight Mask-Specific VAE for fast latent-space process and a decoupled architecture with a lightweight denoising UNet, (2) edge supervision and error prior to mitigate error accumulation during sampling. Extensive experiments of two distinct protocols on eight non-generative and three generative benchmarks demonstrate that DiffIML consistently outperforms state-of-the-art methods, yielding remarkable generalization improvements on diverse unseen datasets. The code is publicly available at https://github.com/scu-zjz/DiffIML.

cs.CV↗
arXiv2026

AudioFM: Towards General Purpose Audio Representation Learning

Learning audio representations that effectively bridge holistic global soundscapes with fine-grained acoustic moments is a long-standing goal in self-supervised learning. Recording-level contrastive objectives encourage agreement between pooled audio representations, but do not explicitly supervise local time-frequency representations. We introduce AudioFM, a self-supervised framework that combines recording-level contrast with patch-wise masked latent discrimination. Using visible acoustic context, a student predicts representations of masked patches and identifies their corresponding full-view teacher targets among candidates from other recordings. We further incorporate weaker supervision from eligible temporal neighbors to encourage local temporal consistency. Extensive evaluations show that AudioFM achieves state-of-the-art performance across diverse audio domains in linear probing, fine-tuning, and zero-shot transfer, demonstrating superior event detection, robust scene-level understanding, and strong domain generalizability.

cs.SD↗
arXiv2026

GS-MoE: Training-Free MoE Compression via Groupwise Subspace Sharing and Pruning

Mixture-of-Experts (MoE) models reduce computation through sparse activation, but storing their expert parameters remains a major memory bottleneck. We introduce GS-MoE, a training-free compression framework that addresses redundancy across and within experts through two compression axes: shared-subspace rank and expert intermediate width. Groupwise low-rank representations capture shared structure across experts while retaining expert-specific projections, and Nyström-based neuron compression reduces intermediate width within each expert. For fixed shared subspaces, we derive per-expert reconstruction-error reductions from the reconstruction objective and use them to guide expert grouping. Budgets are allocated hierarchically using a gradient-free layer-importance score to guide the allocation of rank and width budgets across layers, which eigenvalues and leverage then distribute as non-uniform ranks across groups and intermediate widths across experts. Experiments demonstrate the complementary benefits of compression along both axes, particularly at high compression rates on fine-grained MoE models. On DeepSeek-MoE-16B-Base, at a 60% reduction in total model parameters, GS-MoE retains approximately 90% of the uncompressed model's average reasoning performance, compared with 73% for D$^2$-MoE, while achieving 20% lower out-of-domain perplexity. We provide a code implementation at: https://github.com/cbacary/GS-MoE

cs.LG↗
arXiv2026

Beyond Finite-Sum Optimization: A Primal-Dual Method for Decentralized Finite-Max Optimization

Many decentralized systems are judged by their worst-performing agents rather than their average performance. We study finite-max optimization over a connected undirected network, where each agent knows only its own smooth, possibly nonconvex objective. Quadratic regularization yields a smooth finite-sum representation with a scalar dual variable and locally computable clipped weights. We develop an alternating primal-dual gradient tracking (APDGT) method, using neighbor communication without any global or centralized projections. Under smoothness and level-boundedness assumptions, a localized Lyapunov analysis with vanishing regularization establishes bounded primal iterates and Clarke stationarity of every accumulation point of their network average. The averaged sequence contains an $\varepsilon$-stationary point, measured by near-active gradients, within $O (\varepsilon^{-3})$ iterations. Synthetic tests and decentralized logistic regression on Fashion-MNIST illustrate convergence behavior and potential improvements in worst-agent performance.

math.OC↗
arXiv2026

Harness-Delta Attribution: Dissecting Gains From Harness Evolution

Harness evolution automatically optimizes the prompts, skills, tools, and orchestration code around a fixed-weight model, offering a way to improve agentic systems without updating model weights. The agent harness is iteratively optimized to achieve higher performance on the search set. But how much of this gain reflects benchmark-specific overfitting or additional inference-time compute, and how much may generalize beyond the search set? In this work, we introduce Harness-Delta Attribution (HDA), an evaluation protocol that dissects an evolution gain into artifact-driven overfitting, gains explained by test-time scaling, and the remainder, which gives an upper bound on residual generalizable improvement. We apply HDA to 11 executor models and three harness-evolution frameworks across math, coding, creativity, and agentic tasks. In 21 of these settings that show statistically significant search-set gains, we find that a large share of the improvement is explained by overfitting or test-time scaling. Specifically, among these 21 settings, overfitting and test-time scaling account for 54% and 14% of the gain on average; their combined share exceeds 50% in 14 settings. The residual and held-out transfer vary substantially across tasks, models, and frameworks. Importantly, search-set gains often fail to transfer to held-out tasks, and provide limited evidence that they will persist under practical distribution shifts. Moreover, validation-based signals are insufficient to prevent overfitting. These results motivate attribution-aware evaluation in future harness evolution research.

cs.AI↗
arXiv2026

AlphaBet: Reinforcement Learning for LLM Event Forecasting Agents

Forecasting an unresolved event requires finding relevant evidence and deciding how strongly it supports each outcome. Yet outcome-based training of Large Language Model (LLM) forecasters often begins with preassembled evidence at a single prediction date, leaving evidence acquisition outside the learning loop. We introduce AlphaBet, an agentic reinforcement learning (RL) framework that trains a single LLM policy to search, reason, and report probabilities across multiple pre-resolution dates. Each dated forecast is scored by a proper scoring rule against the eventual outcome, allowing one resolved event to supervise decisions under different information states. We decompose forecasting regret into an evidence-acquisition gap and a probability-reporting gap, and show that Group Relative Policy Optimization (GRPO)'s standard-deviation normalization can distort the reporting target under multi-class proper scores. On GLM-4.5-Air, AlphaBet reduces the Brier score from 0.731 to 0.626 on Forecast-Dojo and from 0.612 to 0.528 on FutureX. Controlled ablations attribute these gains to multi-date training and show that they come from improvements in both information acquisition and probability reporting. Our empirical results validate that GRPO's std normalization degrades proper-scoring-rule performance while leaving top-outcome accuracy largely unchanged.

cs.LG↗
arXiv2026

ForceBench: Benchmarking Force Integration for Robotic Manipulation

Force feedback provides vision-language-action (VLA) policies with physical observations that can guide action selection and execution. However, differences in tasks, demonstrations, and control settings across existing systems hinder reliable comparisons of force integration designs. We introduce ForceBench, a simulation benchmark for evaluating six-axis wrist-force integration under a unified protocol. ForceBench combines eight tasks spanning physical-property inference, resistance-dependent interaction, and contact-rich assembly with 1,600 synchronized demonstrations and task-specific success criteria. We compare one no-force baseline and six force integration designs on a shared $π_{0.5}$ backbone. Prompt Fusion achieves the highest observed mean simulation success of 35.31%, compared with 23.25% for the baseline, and improves success rates across all three task categories. Predictive and temporal gains in simulation concentrate on physical-property inference, while both temporal designs underperform the baseline on assembly. Real-world evaluation of eight systems across five tasks supports the main qualitative trends: Prompt Fusion and Dual-Path Fusion achieve 63.00% mean success, compared with 40.00% for the baseline, but only Prompt Fusion improves every task. These findings support ForceBench's practical value for identifying the benefits and limitations of force integration across manipulation demands. The project page is available at https://forcebench-platform.github.io/.

cs.RO↗
arXiv2026

E2DRL: End-to-End Differentiable Reinforcement Learning for 3D Obstacle Avoidance in Quadruped Robots

Differentiable reinforcement learning shows strong promise for robot policy learning, but its application to legged locomotion has remained limited to blind walking or simulated 2D tasks. Extending it to end-to-end visual obstacle avoidance is challenging because hard collision constraints and early termination upon contact corrupt first-order gradients, trapping optimizers in local minima. We present E2DRL, which addresses this with a soft differentiable obstacle-interaction formulation: penetration penalties across multiple body keypoints with velocity-conditioned stop-gradient shaping to provide the first-order training signals for 3D obstacle avoidance. To realize a deployable perception-control system, a privileged teacher trained via Soft Analytic Policy Optimization (SAPO) in a differentiable simulator is distilled into a recurrent depth-based student via DAgger, then blended with a base locomotion policy using a terrain-aware gating module for deployment. In simulation, E2DRL achieves higher sample efficiency than PPO and demonstrates superior velocity-tracking performance and 3D obstacle-avoidance capability compared with both PPO and Duality-based DCBF NMPC. We present the first sim-to-real validation of end-to-end differentiable RL for 3D obstacle avoidance on a Unitree Go2 in static and dynamic environments.

cs.RO↗
arXiv2026

Interaction-Limited Safe Continuous-Time RL for Dynamical Medical Treatment

Dynamic medical treatment requires deciding treatment intensity and intervention timing, while patient states evolve continuously and adverse events may occur between clinical interactions. Most existing treatment learning methods assume fixed schedules or enforce safety only at discrete decision points. We propose Interaction-Limited Safe Continuous-Time Reinforcement Learning, a framework that jointly optimizes treatment administration and clinical interaction timing under trajectory-level safety constraints. Our key idea is to reformulate the continuous time treatment problem as an option-based semi-Markov decision process, where each option specifies a continuous-time treatment policy and its duration. We develop a safety-tightening mechanism showing that suitably constructed constraints at interaction times guarantee safety over the full continuous-time trajectory with high probability. We further establish finite-sample guarantees for policy learning from logged treatment trajectories and introduce a practical data-driven conservative surrogate. Experiments show that the proposed adaptive interaction-timing mechanism improves both safety and treatment effectiveness over equidistant interaction schemes across different safe policy optimization methods.

cs.LG↗
arXiv2026

How Narrative Wrapping Affects LLM Refusal: A Cross-Language Benchmark and Defense

Safety-aligned language models often refuse a harmful request stated directly but answer the same request inside a role-play or narrative wrapper. We measure this vulnerability across languages and registers: attack success on Qwen3-1.7B is already 89.4% in English and 93.0% in modern Chinese, and reaches 95.7% in Classical Chinese. We build GUISE, a benchmark for systematically studying this vulnerability. It includes parallel requests in English, modern Chinese, and Classical Chinese, matched harmful and benign pairs, wrapper types held out for evaluation, and a stricter criterion that counts warn-then-answer responses as attack successes. Representation analysis shows that language and register move harmful-request representations only slightly away from the model's refusal direction, whereas narrative wrappers move them much farther away. We propose AXIS, which combines preference optimisation with a rotation objective that aligns harmful-request representations with the refusal direction and a commitment objective that trains the model to refuse completely rather than produce a warn-then-answer response. Across Qwen3-1.7B, Qwen3-4B and GLM-4-9B, AXIS achieves the highest combined safety and usability score among the compared methods.

cs.AI↗
arXiv2026

Observation of stacked vortex structure with number controllable Majorana zero mode in Bi$_{2}$Te$_{3}$/FeTe heterojunction

Creation and control of Majorana zero modes (MZMs) in solid devices is the key challenge in realizing fault tolerant topological quantum computation. Here we report spectroscopic evidence for number controllable MZMs in stoichiometric one quintuple layer Bi$_{2}$Te$_{3}$/FeTe heterostructures, together with the observation of stacked vortices in superconducting islands. Low temperature scanning tunnelling microscopy and spectroscopy reveal a long period moire superlattice arising from lattice mismatch, which induces nanoscale modulation of superconducting gaps. Magnetic vortices exhibit a stacked core structure with two layer derived bound state branches: a dispersive branch evolving into an X-shaped pattern away from the vortex center, and a quantum-limit branch characterized by a spatially unsplit MZMs together with discrete integer quantized Caroli-de Gennes-Matricon states. Notably, MZMs are observed in all measured vortices, and their number can be tuned by magnetic field and the size of Bi$_{2}$Te$_{3}$ islands. These results establish stoichiometric 1QL-Bi$_{2}$Te$_{3}$/FeTe as a patterned superconducting interface that enables isolated and few-vortex Majorana configurations, providing a controlled platform for fuse MZMs.

cond-mat.supr-con↗