arXiv subjects

Donghua Cai

Publications and source records attributed to Donghua Cai.

3 recordsLinked to original sources

arXiv2026

Beyond Memory Construction: Rethinking Memory Access for LLM-based Conversational Agents

Memory is a core component of conversational agents, enabling coherent and context-aware behavior over long interactions. Recent approaches commonly rely on LLM-based memory construction, where raw interactions are rewritten into structured memory units and later retrieved via a RAG pipeline. While effective in controlled settings, we show that this paradigm breaks down in long-horizon, high-entropy conversations: memory construction becomes increasingly lossy and unstable as context length and information complexity grow, and incurs prohibitive cost due to repeated LLM invocation. To address these limitations, we propose Threader, a memory system that shifts the focus from memory construction to efficient, structure-aware access over raw interactions. Instead of rewriting interactions, Threader preserves them as first-class memory, organizes them into topic-coherent segments via lightweight incremental segmentation, and enables accurate retrieval through multi-view representation. At query time, it performs multi-signal retrieval that combines segment-level access with localized evidence matching, ensuring both completeness and coherence. Extensive experiments demonstrate that Threader consistently improves answer accuracy and evidence recall, while significantly reducing the memory construction overhead.

cs.CL↗

arXiv2026

OmniEcho: Audio-Visual Spatial Understanding for Omni-Modal Embodied Agents

Humans can effortlessly localize the direction of a sound source and integrate it with visual cues for reasoning, yet this remains challenging for embodied agents. In particular, it is still unclear how to effectively evaluate and model spatial audio understanding in embodied settings. To address this gap, we introduce \textbf{OmniEchoBench}, a unified benchmark for spatial audio-visual perception and audio-vision-language navigation. OmniEchoBench comprises six tasks over 197 real-world spatial audio-visual scenes, 2,972 question-answer pairs, and 900 navigation samples with first-order ambisonics (FOA) audio collected from 30 real-world environments. To enable scalable training supervision, we develop a controllable rendering pipeline for spatial audio. It preserves geometric consistency among sound sources, visual observations, and agent trajectories. Building on this, we propose \textbf{OmniEcho}, a spatially aware omni-modal model. It introduces an FOA spatial encoder alongside a pretrained semantic audio pathway. Extensive experiments show that OmniEcho achieves state-of-the-art performance on spatial audio-visual perception. For our sound-guided navigation, OmniEcho reaches a performance level close to that of traditional vision-language navigation. These results demonstrate that spatial audio can serve as a valuable signal for embodied scene reasoning and navigation, while also highlighting fine-grained spatial localization and distance estimation as important open challenges. Our code and data will be available in https://github.com/PKU-VaLuE-Lab/OmniEcho/tree/main

cs.SD↗

arXiv2025

CSVAR: Enhancing Visual Privacy in Federated Learning via Adaptive Shuffling Against Overfitting

Although federated learning preserves training data within local privacy domains, the aggregated model parameters may still reveal private characteristics. This vulnerability stems from clients' limited training data, which predisposes models to overfitting. Such overfitting enables models to memorize distinctive patterns from training samples, thereby amplifying the success probability of privacy attacks like membership inference. To enhance visual privacy protection in FL, we present CSVAR(Channel-Wise Spatial Image Shuffling with Variance-Guided Adaptive Region Partitioning), a novel image shuffling framework to generate obfuscated images for secure data transmission and each training epoch, addressing both overfitting-induced privacy leaks and raw image transmission risks. CSVAR adopts region-variance as the metric to measure visual privacy sensitivity across image regions. Guided by this, CSVAR adaptively partitions each region into multiple blocks, applying fine-grained partitioning to privacy-sensitive regions with high region-variances for enhancing visual privacy protection and coarse-grained partitioning to privacy-insensitive regions for balancing model utility. In each region, CSVAR then shuffles between blocks in both the spatial domains and chromatic channels to hide visual spatial features and disrupt color distribution. Experimental evaluations conducted on diverse real-world datasets demonstrate that CSVAR is capable of generating visually obfuscated images that exhibit high perceptual ambiguity to human eyes, simultaneously mitigating the effectiveness of adversarial data reconstruction attacks and achieving a good trade-off between visual privacy protection and model utility.

cs.CR↗