arXiv subjects

Yujian Ma

Publications and source records attributed to Yujian Ma.

6 recordsLinked to original sources

arXiv2026

M3-Tele: A Unified Multimodal Teleoperational Framework for Compliant Whole-Body Mobile Manipulation

Executing contact-rich tasks efficiently requires the seamless integration of whole-body coordination and physical compliance regulation. However, existing teleoperation and data-collection frameworks often overlook the joint consideration of multimodal perception and coordinated whole-body operation. This limitation can reduce the efficiency and quality of demonstration collection, thereby affecting the effectiveness of downstream policy learning. In this work, we present \textbf{M3-Tele}: A Unified \underline{M}ultimodal \underline{Tele}operational Framework for Compliant Whole-Body \underline{M}obile \underline{M}anipulation, enabling stable physical interaction and capturing aligned visual, tactile, force, and proprioceptive observations during task execution. Extensive experiments demonstrate that the proposed framework significantly improves contact-rich teleoperation performance. The proposed controller reduces the force tracking error from 4.132~N to 0.346~N, the contact loss from 2.46 to 0.02 events per trial and the tactile deformation error by 65\%. User studies across four mobile manipulation tasks also verify the reliability and usability of the proposed system. Furthermore, Diffusion Policy experiments highlight the value of joint tactile and force sensing.

cs.RO↗

arXiv2026

Training-Free VLM Personalization via Calibrated Residual Decoding

Vision-language models can be personalized in a training-free manner by directly providing user profiles, preferences, or visual references at inference time, without updating model parameters. However, direct personalized prompting does not guarantee that the model will reliably exploit such evidence. The predictive distribution under the positive user profile often mixes two sources: personalized signals genuinely supported by the current profile, and the model's generic visual or linguistic priors. As a result, from the positive-profile response alone, it is difficult to determine whether a high-confidence answer is supported by the user profile or merely reflects the model's default preference. To address this problem, we propose a training-free calibrated residual decoding framework. Given the same image and question, we construct three evidence conditions: a positive profile , a counterfactual profile , and an empty profile . Our method keeps the prediction under as the anchored base, and explicitly estimates the marginal contribution of personalization from score differences across the three conditions. We further introduce normalized-entropy-based uncertainty calibration, allowing the strength of personalized enhancement to adapt to the reliability of the residual signal. Experiments on MMPB, YoLLaVA, and MyVLM show that the proposed method improves personalized multimodal understanding without fine-tuning, with consistent gains on identity-sensitive visual personalization tasks. Additional analysis shows that entropy calibration stabilizes residual decoding when the contrastive personalization signal is uncertain.

cs.CV↗

arXiv2026

From Semantics to Readout: Mechanistic Understanding of Audio Tokens after Fine-Tuning for Temporal Audio Grounding

Large audio-language models (LALMs) convey acoustic evidence to language decoders through native audio tokens, yet the internal roles of these tokens remain poorly understood. Using temporal audio grounding as a diagnostic setting, we examine how language-model fine-tuning affects the layerwise semantics, decoder accessibility, and temporal output alignment of native audio-token states through four complementary analyses: query-conditioned token semantics, calibrated token readout, temporal-window probes, and residual-delta erasure during generation. Alongside substantial improvements in temporal localization, semantic analysis of Qwen2.5-Omni shows that latent evidence for queried events is already present before fine-tuning and that the audio tokens most strongly aligned with the queried event appear at similar temporal positions before and after fine-tuning. After fine-tuning, event-related information in audio tokens becomes more accessible to the decoder, especially in early and middle layers, and a cross-checkpoint control shows that this improvement arises primarily from decoder adaptation. Temporal probes show that the base checkpoint already contains recoverable information about annotated windows and that fine-tuning mainly improves alignment with each checkpoint's own predicted temporal support. Residual-delta erasure further shows that removing audio-token updates within predicted windows harms timestamp generation more than removing the same number of randomly selected updates. The same broad improvements in decoder readability and prediction alignment also appear in Qwen2-Audio. Together, these results support a semantics-to-readout account in which grounding fine-tuning helps the decoder read existing event evidence and connect it more reliably to temporal outputs.

cs.SD↗

arXiv2025

A Lightweight Fourier-based Network for Binaural Speech Enhancement with Spatial Cue Preservation

Binaural speech enhancement faces a severe trade-off challenge, where state-of-the-art performance is achieved by computationally intensive architectures, while lightweight solutions often come at the cost of significant performance degradation. To bridge this gap, we propose the Global Adaptive Fourier Network (GAF-Net), a lightweight deep complex network that aims to establish a balance between performance and computational efficiency. The GAF-Net architecture consists of three components. First, a dual-feature encoder combining short-time Fourier transform and gammatone features enhances the robustness of acoustic representation. Second, a channel-independent globally adaptive Fourier modulator efficiently captures long-term temporal dependencies while preserving the spatial cues. Finally, a dynamic gating mechanism is implemented to reduce processing artifacts. Experimental results show that GAF-Net achieves competitive performance, particularly in terms of binaural cues (ILD and IPD error) and objective intelligibility (MBSTOI), with fewer parameters and computational cost. These results confirm that GAF-Net provides a feasible way to achieve high-fidelity binaural processing on resource-constrained devices.

eess.AS↗

arXiv2025

Behind the Scenes: Mechanistic Interpretability of LoRA-adapted Whisper for Speech Emotion Recognition

Large pre-trained speech models such as Whisper offer strong generalization but pose significant challenges for resource-efficient adaptation. Low-Rank Adaptation (LoRA) has become a popular parameter-efficient fine-tuning method, yet its underlying mechanisms in speech tasks remain poorly understood. In this work, we conduct the first systematic mechanistic interpretability study of LoRA within the Whisper encoder for speech emotion recognition (SER). Using a suite of analytical tools, including layer contribution probing, logit-lens inspection, and representational similarity via singular value decomposition (SVD) and centered kernel alignment (CKA), we reveal two key mechanisms: a delayed specialization process that preserves general features in early layers before consolidating task-specific information, and a forward alignment, backward differentiation dynamic between LoRA's matrices. Our findings clarify how LoRA reshapes encoder hierarchies, providing both empirical insights and a deeper mechanistic understanding for designing efficient and interpretable adaptation strategies in large speech models. Our code is available at https://github.com/harryporry77/Behind-the-Scenes.

cs.SD↗

arXiv2023

Camera-Based HRV Prediction for Remote Learning Environments

In recent years, due to the widespread use of internet videos, remote photoplethysmography (rPPG) has gained more and more attention in the fields of affective computing. Restoring blood volume pulse (BVP) signals from facial videos is a challenging task that involves a series of preprocessing, image algorithms, and postprocessing to restore waveforms. Not only is the heart rate metric utilized for affective computing, but the heart rate variability (HRV) metric is even more significant. The challenge in obtaining HRV indices through rPPG lies in the necessity for algorithms to precisely predict the BVP peak positions. In this paper, we collected the Remote Learning Affect and Physiology (RLAP) dataset, which includes over 32 hours of highly synchronized video and labels from 58 subjects. This is a public dataset whose BVP labels have been meticulously designed to better suit the training of HRV models. Using the RLAP dataset, we trained a new model called Seq-rPPG, it is a model based on one-dimensional convolution, and experimental results reveal that this structure is more suitable for handling HRV tasks, which outperformed all other baselines in HRV performance and also demonstrated significant advantages in computational efficiency.

cs.CV↗