Searcharxiv⌕ Search

arXiv subjects

Zhitao Chen

Publications and source records attributed to Zhitao Chen.

7 recordsLinked to original sources

MAST: Medoid-Anchored Subspace Steering for Visual Grounding in Multimodal LLMs

Existing activation steering methods for multimodal large language models (MLLMs) use the global activation mean as the steering vector, which tends to cause cross-task interference and unstable performance. We trace this failure to the multi-centered geometry of task activations. The global mean lands in the empty region between clusters and is faithful to no single task. Moreover, the steering signal in each task-pure local cluster is low-rank, and different query tasks benefit from emphasizing different families of local-basis directions under the shared routing protocol. To address this, we propose MAST (Medoid-Anchored Subspace Steering), a training-free, input-adaptive activation steering framework. MAST groups samples by task, performs multi-center clustering within each group, and selects one representative medoid per cluster. At inference time, each query is routed to multiple representative medoids. For each matched cluster, a low-rank local steering direction is extracted and aggregated across clusters to produce the final steering vector. Building on MAST, MAST-PR (Probe-Reweighted) uses a lightweight task-aware probe to reweight task-specialized local-direction families, yielding a further average improvement of +8.09 points over MAST.

cs.CV↗

Memory-Augmented Online Test-Time Adaptation for EEG Motor Imagery via Perturbation Consistency and Confidence Calibration

Recently, electroencephalography (EEG)-based brain--computer interfaces (BCIs) have gained traction as portable neural measurement systems. However, the practical deployment of these instruments is often hindered by nonstationarity and cross-subject distribution shifts, which lead to significant intersubject inconsistency. Existing adaptation methods remain impractical for real-time measurement scenarios, as they typically require offline access to source or aggregated target data. To address these challenges, this article proposes memory-augmented online test-time adaptation (OTTA) with perturbation consistency and confidence calibration (MOPC), a source-free framework designed to enhance the reliability of BCI instrumentation. MOPC enables the measurement system to update its decoding parameters on the fly during the inference stage, facilitating single-pass adaptation without accessing sensitive source data. We introduce random-mask perturbations to enforce prediction consistency against inherent EEG signal noise and acquisition artifacts. Furthermore, a confidence-calibrated reweighting loss is designed to integrate feature-space similarity, effectively mitigating unreliable pseudolabels and class imbalance during continuous signal acquisition. To ensure system stability and prevent the catastrophic forgetting of established neural patterns, a lightweight memory module is incorporated to regularize parameter updates within a tight latency budget. Evaluated on the BCI IV IIb and OpenBMI datasets, MOPC achieves state-of-the-art performance, outperforming prior methods by 2.22~\% and 3.67~\%, respectively, while keeping the per-trial adaptation latency to approximately 40--60 ms. This study provides a robust, recalibration-free solution for real-time EEG decoding, marking a significant advancement in the precision and usability of practical BCI measurement instruments.

cs.LG↗

AlignedServe: Orchestrating Prefix-aware Batching to Build a High-throughput and Computing-efficient LLM Serving System

High-throughput inference serving is essential for applications built on large language models (LLMs). Existing serving frameworks reduce request-level and batch-level bubbles through batching and scheduling, but often overlook bubbles within each decode iteration. Tokens generated in the same iteration may incur different costs because they depend on KV caches of different lengths; tokens with long KV caches can become bottlenecks and delay the next iteration. We propose AlignedServe, an LLM serving framework built around prefix-aware batching. It groups requests with similar KV-cache lengths into the same batch to reduce iteration-level bubbles. To support this policy efficiently, AlignedServe uses large CPU memory to maintain sufficient in-flight requests for batching and applies a batch-level scheduling policy to reduce batch-level bubbles. It also introduces a GPU-Prefetch-For-GPU architecture, where one GPU prefetches KV cache for another to reduce CPU-to-GPU transfer latency. Experiments on synthetic and application workloads show that AlignedServe improves decoding throughput by up to 1.98 times and reduces latency by up to 7.4 times over state-of-the-art systems.

cs.DC↗

Active nematic defects in compressible and incompressible flows

We study the dynamics of active nematic films on a substrate driven by active flows with or without the incompressible constraint.Through simulations and theoretical analysis, we show that arch patterns are stable in the compressible case, whilst they become unstable under the incompressibility constraint. For compressible flows at high enough activity, stable arches organize themselves into a smectic-like pattern, which induce an associated global polar ordering of $+1/2$ nematic defects. By contrast, divergence-free flows give rise to a local nematic order of the $+1/2$ defects, consisting of anti-aligned pairs of neighboring defects, as established in previous studies.

cond-mat.soft↗

Odd moduli of disordered odd elastic lattices

We study the effects of bond disorder on triangular and honeycomb lattices where each spring has a probability p to be odd elastic. Using an effective medium theory and numerical simulations, we uncover the behavior of odd moduli in the presence of disorder, which we interpret as a crossover between the affine response of the passive elastic backbone, and a rigidity percolation transition in the odd elastic components. Though oddness is generally robust against disorder even at low p, we find that fine-tuned features of an odd elastic honeycomb lattice are not robust against disorder.

cond-mat.soft↗

Spontaneous tilt of single-clamped thermal elastic sheets

Very thin elastic sheets, even at zero temperature, exhibit nonlinear elastic response by virtue of their dominant bending modes. Their behavior is even richer at finite temperature. Here we use molecular dynamics (MD) to study the vibrations of a thermally fluctuating two-dimensional elastic sheet with one end clamped at its zero-temperature length. We uncover a tilt phase in which the sheet fluctuates about a mean plane inclined with respect to the horizontal, thus breaking reflection symmetry. We determine the phase behavior as a function of the aspect ratio of the sheet and the temperature. We show that tilt may be viewed as a type of transverse buckling instability induced by clamping coupled to thermal fluctuations and develop an analytic model that predicts the tilted and untilted regions of the phase diagram. Qualitative agreement is found with the MD simulations. Unusual response driven by control of purely geometric quantities like the aspect ratio, as opposed to external fields, offers a very rich playground for two-dimensional mechanical metamaterials.

cond-mat.soft↗

The role of fluid flow in the dynamics of active nematic defects

We adapt the Halperin-Mazenko formalism to analyze two-dimensional active nematics coupled to a generic fluid flow. The governing hydrodynamic equations lead to evolution laws for nematic topological defects and their corresponding density fields. We find that $\pm 1/2$ defects are propelled by the local fluid flow and by the nematic orientation coupled with the flow shear rate. In the overdamped and compressible limit, we recover the previously obtained active self-propulsion of the +1/2 defects. Non-local hydrodynamic effects are primarily significant for incompressible flows, for which it is not possible to eliminate the fluid velocity in favor of the local defect polarization alone. For the case of two defects with opposite charge, the non-local hydrodynamic interaction is mediated by non-reciprocal pressure-gradient forces. Finally, we derive continuum equations for a defect gas coupled to an arbitrary (compressible or incompressible) fluid flow.

cond-mat.soft↗