Searcharxiv⌕ Search

arXiv subjects

Yiming Chen

Publications and source records attributed to Yiming Chen.

At least 19 recordsLinked to original sources
arXiv2026

Do Agentic Policies Need to Look at Every Step? Geometric Abstraction for Reasoning and Action

Recent multimodal language models can reason directly about robot scenes, but it is unclear whether they need to repeatedly reinterpret raw visual observations at every step of an interaction. We propose shifting routine spatial reasoning from images to text and symbols through a 3D geometric abstraction that is refreshed locally from observations and read directly by the language models. Specifically, our interface represents scene geometry via typed, evidence-aware symbolic references to object parts, grounds them from calibrated RGB-D, and returns compact textual checkpoints with metric residuals after each action. Symbolic task structure can thus be reused across decisions while metric values are rebound to the current scene, with direct visual reasoning invoked only on demand. Across offline policy synthesis and continuous online execution, reasoning over grounded text and symbols improves multi-step manipulation completion while substantially reducing repeated multimodal queries, positioning explicit geometric state as an effective substrate for agentic robot policies.

cs.RO↗
arXiv2026

$TF^2$-Steering: Action Refinement with Tactile Forecasts and Feedback via Velocity-Field Steering

Tactile sensing plays a crucial role in everyday human interaction. Consider inserting a plug into a tight socket, vision alone provides limited guidance during the final insertion stage, and we instead rely on both anticipated contact and realized tactile feedback to ajust the insertion motion. The same capability is equally important for robots, particularly in dexterous manipulation tasks. Yet many contemporary robot policies remain vision-dominant. This visual bias, compounded by limited tactile data and sparse contact events, can cause directly incorporated tactile signals to be ignored or even compromise the policy's pretrained capabilities, yielding degrading performance as observed in our experiments. We present $TF^2$-Steering, which refines actions via learned residual steering of the flow-matching action expert's velocity field, using a tactile mixture-of-experts architecture to integrate both tactile forecasts and real-time tactile feedback. Together with specific designed three stages training recipe, $TF^2$-Steering enables tactile signals to correct and, when necessary, redirect the flow-matching action-generation process, while building on the strong action-generation capacity of the vision-based foundation policy. Extensive experiments across six simulated and six real-world manipulation tasks demonstrate the effectiveness of $TF^2$-Steering. It outperforms the strongest evaluated baselines by 10.7 percentage points in average simulation success rate and 9.1 percentage points in average real-world progress success rate. Project page: https://yiming-cc.github.io/tf2-steering/.

cs.RO↗
arXiv2026

Mixing time under monotone censoring

We prove the conjecture of Ding and Mossel~\cite{DM}: the lazy random walk on the discrete cube, censored to any increasing set of fixed positive density, mixes in time $O(n\log n)$. More precisely, for every nonempty increasing set $A\subseteq\{0,1\}^n$, with $a=μ(A)$ and $μ$ uniform on the cube, we prove $$ t_{\mathrm{mix}}(P)\le C_a n\log(en), $$ where $C_a$ depends only on $a$. The proof uses hypercontractivity on the ambient cube to strengthen the Poincaré inequality on coordinate sections. An optimal stopping time occupation inequality for increasing sets converts the resulting local bound into a uniform bound on hitting times of large sets. An appendix by Y.~Peres improves the density dependence to $C_a=13a^{-1}\log(e/a)$, within a logarithmic factor of the conjectured $a^{-1}$ dependence.

math.PR↗
arXiv2026

X-World: Transferring Video Priors to Variable-View Robot Control

Video diffusion models provide a source of interaction representations for robot control, but transferring these representations requires reconciling variable camera inputs with heterogeneous action interfaces and closed-loop execution. We present X-World, a framework that transfers a video prior through a fixed-length token sequence to a shared action expert. A residual cross-view attention branch extends a pretrained video backbone to one, two, or three views while retaining its original self-attention path. A layer-aligned Q-Former aggregates backbone features, and dataset-specific modules map between native state and action interfaces and the shared action expert. Training separates video adaptation from action-only pretraining by freezing the video backbone in the latter stage, followed by joint target-domain post-training. The policy extracts video features with a single backbone pass and uses asynchronous action chunking for execution. On LIBERO, X-World reaches 98.5\% mean success; its LIBERO-trained checkpoint obtains 76.5\% on LIBERO-Plus without further adaptation. The seven-shift breakdown reveals gains in robot, language, lighting, and layout perturbations, alongside weaknesses under camera, background, and noise shifts relative to OpenVLA-OFT. On six dual-arm real-world tasks, it achieves 81.7\% mean success over 20 trials per task. Removing video adaptation or cross-view fusion produces the largest observed success drops among the tested removals.

cs.RO↗
arXiv2026

Training in Deployment: Human-in-the-loop Online Post-Training for Robot Manipulation

Large pretrained robot foundation models provide strong priors for manipulation, yet adapting them to target tasks with offline teleoperated demonstrations often fails to cover the contact-rich, recovery, and precision-critical states encountered during deployment. Existing methods mitigate this issue by augmenting offline demonstrations with human-in-the-loop data or online reinforcement learning. Despite this, offline post-training adapts slowly because it cannot immediately absorb deployment experience, while online RL typically relies on value functions or auxiliary actor-critic models, increasing both training complexity and instability. We present TIDE (Training In DEployment), a human-in-the-loop online post-training system. TIDE introduces a provenance-conditioned supervised learning framework that updates the policy online from human interventions and policy-generated rollouts collected during deployment, without a learned critic, reward model, or action scorer. Across four real-world bimanual manipulation tasks, TIDE improves milestone-normalized task progress and binary full-task completion over the warm-start policy and matched additional offline demonstrations, and performs comparably to the fixed-policy collection baseline while making updated checkpoints available during deployment. Each task-specific online post-training session, from the start of deployment rollouts until the final checkpoint is available, completes in less than one hour while learning and checkpoint handling run asynchronously.

cs.RO↗
arXiv2026

Frontier-OPD: Discovering the Edge of Agent Capability and Teaching Agents to Go Beyond

On-policy distillation aligns teacher feedback with a student's current behavior, yet acquiring useful supervision for multi-turn agents requires identifying both where guidance is needed and whether the student can sustain the interaction afterward. We introduce Frontier-OPD, which addresses this problem through the student's learning frontier: decision situations that remain challenging but admit useful autonomous continuation after brief teacher guidance. To operationalize this frontier, Frontier-OPD organizes on-policy experience into a shared-state trajectory graph, where trajectory-balanced aggregation of teacher-student disagreement identifies distinct intervention candidates while preserving the histories needed for replay. Short teacher bridges are then assessed by comparing agreement over the bridge and the ensuing student continuation with an equally long segment of the original trajectory, providing a behavioral proxy for guidance that the current student can build on. Accepted teacher responses augment ordinary distillation data, and fresh rollouts renew the frontier estimate as the policy changes. Experiments on ALFWorld, WebShop, and ScienceWorld report higher mean autonomous performance than the compared distillation baselines across the evaluated student scales and teacher configurations, with ablations indicating complementary effects of candidate selection and guidance filtering. In one ALFWorld timing setting, the additional collection cost is offset by a shorter time to a target success rate.

cs.CL↗
arXiv2026

SLM-SS: Token-Based Speech Separation with Speech Language Models

Speech separation (SS) has witnessed substantial progress in recent years with the advent of neural network-based methods, achieving notable improvements on signal-level evaluation metrics. Existing speech separation systems are mostly optimized with signal-level objectives, and often yield outputs that are acceptable in waveform metrics but suboptimal for intelligibility-sensitive downstream tasks. To address this gap, we propose SLM-SS, a token-based generative speech separation framework that models multi-talker mixtures with speech language models. We first discretize clean speech into multi-codebook token sequences with neural codecs and serialize multi-speaker targets via Serialized Output Training (SOT). An encoder-decoder autoregressive speech language model predicts the zero-order sequence to establish speaker ordering and coarse linguistic content, and a non-autoregressive module then refines higher-order residual codebooks for efficient reconstruction. We further introduce a per-talker non-autoregressive strategy that simplifies high-order modeling and improves the stability. On Libri2Mix, SLM-SS with per-talker NAR achieves $6.92\%$ WER, approaching codec-only reconstruction. The WavTokenizer variant eliminates NAR refinement and achieves the highest mean subjective quality score among evaluated separation systems. Its UTMOS and DNSMOS advantages over discriminative baselines persist on unseen noisy and Chinese mixtures without fine-tuning.

cs.SD↗
arXiv2026

RiSE: Residual Subspace Expert for Generalizable Text-Centric Image Forgery Localization

As AI-assisted image editing becomes increasingly prevalent, Text-Centric Image Forgery Localization (TFL) is essential for protecting trust in financial and legal records. However, we observe that existing TFL detectors predominantly rely on Full-Parameter Fine-Tuning (FPFT) and fail to generalize to unseen forgery types because FPFT can induce low-rank feature collapse and overfit training artifacts. Meanwhile, prior detectors capture diverse forgery traces with a single unified representation, making subtle traces entangled and difficult to learn. To address these challenges, we propose RiSE, a Residual Subspace Expert model that preserves pre-trained priors by freezing principal SVD components and adapting only the residual subspaces with a ViT-only segmentor to avoid feature collapse. RiSE introduces forgery-aware residual experts, where each expert is trained independently on a training subset with a particular forgery type, and we show that cross-domain localization performance improves consistently as more experts are added. At inference, we introduce Latent Perturbation Confidence (LPC), which selects the most confident expert by analytically computing the latent-perturbed output. LPC enables efficient confidence estimation with a single forward pass, avoiding repeated forward passes on perturbed samples and improving expert selection. By training on synthetic forgery data, RiSE with 1 expert outperforms prior methods on real-world forgery images by 20.4% while reducing training steps by 10 times. Scaling the number of experts to 9 yields a 35.8% gain. RiSE remains effective with DCT feature, even when the majority of the parameters is frozen with RGB-only pretraining. Code: https://github.com/KahimWong/RiSE.

cs.CV↗
arXiv2026

Lossless Inference Acceleration for ASR via EAGLE-Style Speculative Decoding

Autoregressive Transformer and Conformer models have achieved state-of-the-art performance in automatic speech recognition (ASR). However, their sequential token generation mechanism imposes a significant bottleneck during decoding, resulting in substantial inference latency. While speculative decoding has emerged as a promising lossless acceleration technique, existing ASR-oriented approaches either suffer from low draft acceptance rates due to suboptimal draft quality or rely on specially designed external draft models, which limit their generality and practicality. To overcome these limitations, we propose EAGLE-ASR, a novel framework that accelerates ASR inference via EAGLE-style speculative decoding while strictly preserving lossless output quality. Specifically, we introduce Speech-Draft, a lightweight autoregressive module that serves as a universal draft model by jointly leveraging encoder acoustic features and hierarchical decoder representations from the target ASR model. Furthermore, we adopt a multi-step generation strategy with self-distilled data to train the Speech-Draft, enabling it to better approximate the target model's output distribution and maintain causal consistency during multi-token generation. Extensive experiments on representative Whisper and Conformer AED models demonstrate that EAGLE-ASR achieves significant lossless speedups of up to $1.9\times$ and $2.4\times$, respectively, consistently outperforming existing speculative decoding baselines for ASR. Our proposed framework further generalizes effectively to multilingual ASR and speech translation tasks, establishing EAGLE-ASR as a practical and universally applicable solution for accelerating modern ASR systems without compromising recognition accuracy.

cs.SD↗
arXiv2026

VISTA: Internalizing Collective Visual Experience via On-Policy Distillation for Active Multimodal Agents

Active multimodal agents use visual tools to acquire task-relevant evidence while reasoning. Although reinforcement learning samples multiple interaction trajectories per input, outcome-based objectives primarily use the group to estimate scalar advantages, leaving complementary visual discoveries underused. We introduce VISTA, which internalizes collective visual experience through on-policy distillation by turning observations from same-input rollouts into shared supervision. Collective visual experience distillation (CVED) organizes these observations with their interaction context and aligns them with individual decisions, while heterogeneity-aware policy improvement (HAPI) reinforces successful trajectories and provides experience-guided distillation for unsuccessful attempts. An experience-conditioned teacher evaluates the student's sampled response prefixes, allowing discoveries from one trajectory to guide learning in another without replacing the student's original history or generating new target trajectories. The trained agent retains its visual tools and acts using its own interaction history. VISTA achieves the strongest average performance among the evaluated active multimodal agents of comparable size and consistently outperforms same-backbone training baselines across fine-grained perception and general reasoning tasks, demonstrating the value of collective experience for active multimodal learning.

cs.CV↗
arXiv2026

An Exponential Lower Bound for the Permanent of Random Bernoulli Matrix

Let $M_n$ be an $n\times n$ matrix with independent uniform sign entries. We prove that there exist absolute constants $C,c>0$ such that, for all sufficiently large $n$, \[ \mathbb{P}\!\left( \left|\operatorname{Per}(M_n)\right| \ge e^{-Cn}\sqrt{n!} \right) \ge 1-n^{-c}. \] This establishes the exponential scale lower bound suggested by Tao and Vu. The proof bounds the cumulative logarithmic loss of the sum of squared permanents of minors under row exposure.

math.PR↗
arXiv2026

CMAMBADEPTH: Self-supervised Monocular Depth Estimation with Channel Mamba and Hybrid Attention

Accurate monocular depth estimation serves as a core enabler for single camera scene understanding. However, existing self-supervised monocular depth estimation methods generally suffer from the bottleneck of inefficient cross-scale information interaction and difficulty in balancing local and global spatial modeling. In this paper, we propose CMambaDepth, a self-supervised framework that achieves efficient multi-scale feature fusion and fine-grained contextual modeling via channel-wise selective state propagation. Specifically, Bidirectional Channel Mamba (Bi-CMamba) aligns encoder features across scales and enables bidirectional information exchange among ordered scale groups. Unidirectional Channel Mamba (Uni-CMamba) progressively aggregates decoder features and retains fine-grained scale groups through a group selection mechanism for subsequent fusion. Furthermore, a Hybrid Attention Module (HAM) is introduced to combine large-kernel local context and Manhattan self-attention for complementary spatial modeling. Experimental results demonstrate that our method achieves highly competitive performance. Specifically, our model achieves an AbsRel of 0.094 and an RMSE of 4.156 on KITTI, and an AbsRel of 0.140 on DDAD. In the zero-shot cross-dataset generalization test on NYUv2, it attains an AbsRel of 0.232, outperforming the baseline RA-Depth by 7.2%.

cs.CV↗
arXiv2026

High-Moment Stability and Error Analysis of a Fully Discrete LDG-IMEX Method for High Dimensional Nonlinear Stochastic Convection-Diffusion Equations

A fully discrete local discontinuous Galerkin (LDG) method coupled with an implicit-explicit (IMEX) Euler time discretization is presented and analyzed for a class of high dimensional nonlinear stochastic convection-diffusion equations driven by multiplicative $\mathcal Q$-Wiener noise. The model allows nonlinear leading coefficients, nonlinear convection terms, dissipative source terms, and gradient-dependent noise. The diffusion operator is treated implicitly through the LDG formulation, while the nonlinear convection, lower-order drift, and stochastic terms are evaluated explicitly. The main contribution is a high-moment stability and error analysis for the fully discrete scheme. A central difficulty is that the nonlinear terms lead to pathwise growth factors that cannot be controlled uniformly on the full sample space. To provide the stability and error estimate, we introduce recursively defined nested subsets adapted to the numerical solution. Under the stated stochastic parabolicity and refinement conditions, we prove that these subsets have probabilities converging to one. On these subsets, the numerical solution satisfies high-moment stability, and the fully discrete error converges with order arbitrarily close to $r+1$ in space and $1/2$ in time. We also derive a pathwise error estimate by combining the high-moment error bound with a discrete Kolmogorov argument. Numerical experiments for stochastic Burgers' and Allen-Cahn equations confirm the theoretical rates and demonstrate the robustness of the proposed method for nonlinear stochastic models.

math.NA↗
arXiv2026

Chaos of Berry curvature for BPS microstates

We expect black hole microstates to differ in their chaotic properties from states associated with other geometries. For supersymmetric black holes, ordinary level statistics cannot diagnose this distinction, since their energy levels are exactly degenerate. We propose that there is an intrinsic probe of chaos, encoded in the mixing of the microstates under changes in the couplings of the theory, as determined by the non-Abelian Berry curvature of the BPS states under certain deformations. For states dual to horizonless geometries in holographic systems, such as 1/2-BPS states in the D1/D5 CFT and 1/4-BPS states in $\mathcal{N}=4$ SYM, we find that the Berry curvature for marginal deformations is non-random and often exactly zero at generic couplings. By contrast, for states dual to supersymmetric black holes, we show through computations in $\mathcal{N}=2$ super-JT gravity and explicit numerics in the $\mathcal{N}=2$ SYK model that the Berry curvature resembles a random matrix. We also uncover interesting topological features of the $\mathcal{N}=2$ SYK moduli space, as probed by Chern numbers. These results suggest that the Berry curvature sharply distinguishes black hole microstates from smooth horizonless states and provides a robust diagnostic of chaos in supersymmetric sectors.

hep-th↗
arXiv2026

PSMP-CLIP: Patch-Prompt SAM and Multi-Semantic Prompting for CLIP-Based Zero-Shot Anomaly Detection

Zero-shot anomaly detection aims to localize anomalies without target-domain samples. Existing CLIP-based methods suffer from coarse anomaly maps and limited semantic prompts. We propose PSMP-CLIP, integrating patch-prompt SAM2 segmentation (PPSS) and multi-semantic guided prompt regularization (MSGPR). PPSS samples prompts directly from intermediate patch features, avoiding threshold drift and guiding SAM2 to produce precise masks. MSGPR uses multiple learnable prompts constrained by semantic anchors to preserve generalization. Experiments on 14 datasets show highly competitive performance, achieving the best pixel-level AUROC on MVTec AD, BTAD, DTD-Synthetic, CVC-ClinicDB, TN3K, Endo, and Kvasir.

cs.CV↗
arXiv2026

Understanding the Limits of Agentic ICD Coding

ICD-10-CM codes are alphanumeric codes used in the US to classify diagnoses and injuries for medical billing and epidemiological reporting. Standard ICD-10-CM benchmarks report aggregate metrics that obscure performance on complex coding scenarios. We evaluate neural, workflow, and agentic systems on a rarity-stratified set of MIMIC-IV discharge summaries and identify two orthogonal failure modes. Neural classifiers exhibit a 0.43 micro-F1 gap between rare and common codes. Workflow systems handle rare codes well but score near zero on injury and external cause codes that require multi-step guideline following. A tool-augmented agentic configuration with structured access to official ICD-10-CM reference materials recovers up to 0.34 micro-F1 on this subset. No single system dominates across all conditions.

cs.CL↗
arXiv2026

A Fully Discrete Local Discontinuous Galerkin Method for Quasilinear Stochastic Convection-Diffusion-Type Equations

In this paper, we develop and analyze a fully discrete local discontinuous Galerkin (LDG) method with IMEX-Euler time discretization for a class of multi-dimensional quasilinear stochastic convection-diffusion-type equations driven by multiplicative $\mathcal Q$-Wiener noise. The leading diffusion matrix may depend on the solution as well as the spatial and temporal variables, while the lower-order drift and noise coefficients may depend on both the solution and its gradient. Under a suitable stochastic parabolicity condition, we establish unconditional high-moment stability estimates for the fully discrete scheme in the quasilinear setting. In the semilinear setting, where the leading diffusion matrix is independent of the solution but may vary in space and time, we further prove optimal high-moment strong error estimates of order $\mathcal O(h^{r+1})$ in space and $\mathcal O(k^{1/2})$ in time. A pathwise error estimate is then derived by combining the high-moment error bound with a discrete Kolmogorov argument. Numerical experiments are presented to illustrate the stability and convergence properties of the proposed method.

math.NA↗
arXiv2026

A semicircle law for the normalized Laplacian of sparse random graphs

We study the limiting spectral distribution of the normalized Laplacian $\mathcal L$ of an Erdős-Rényi graph $G(n,p)$. To account for the presence of isolated vertices in the sparse regime, we define $\mathcal L$ using the Moore-Penrose pseudoinverse of the degree matrix. Under this convention, we show that the empirical spectral distribution of a suitably normalized $\mathcal L$ converges weakly in probability to the semicircle law whenever $np\to\infty$, thereby providing a rigorous justification of a prediction made in (Akara-pipattana and Evnin, 2023). Moreover, if $np>\log n+ω(1)$, so that $G(n,p)$ has no isolated vertices with high probability, the same conclusion holds for the standard definition of $\mathcal L$. We further strengthen this result to almost sure convergence when $np=Ω(\log n)$. Finally, we extend our approach to the Chung-Lu random graph model, where we establish a semicircle law for $\mathcal L$ itself, improving upon (Chung, Lu, and Vu 2003), which obtained the semicircle law only for a proxy matrix.

math.PR↗