Asynchronous Federated Learning (AFL) enhances the efficiency of edge collaborative learning systems by asynchronously aggregating client updates to prevent slowdowns from slow clients. However, in non-IID scenarios, such operations magnify uneven learning among various local data, degrading global model generalization performance on various local data. Prior solutions work by increasing the contribution of certain slow clients in global aggregation but fail to balance efficiency and generalization performance. In this paper, to deal with such a contradiction, we propose an adaptive fine-tuning-based efficient AFL framework, called AFLTuning, which enables certain clients to contribute high-quality local updates more frequently, thereby facilitating rapid and balanced global learning without introducing additional delay. Specifically, we first propose a data distribution-driven underrepresented client identification mechanism to recognize clients that may have highly skewed data distributions or infrequent participation in global aggregation due to long local processing times, causing the global model to poorly learn their local data. Then, we develop a prediction calibration-based adaptive fine-tuning mechanism to improve these clients’ training efficiency and quality, and a frequency-aware weighted aggregation mechanism to strategically increase their weights during global aggregation. These designs allow underrepresented clients to participate more frequently and contribute more to the global model learning without introducing additional delay, thus mitigating the uneven learning and improving the generalization of AFL while maintaining system efficiency. Extensive experiments demonstrate AFLTuning’s superiority to state-of-the-art methods in both performance and efficiency.
Xiaoyi Pang, Yajie Zhou, Zhibo Wang et al.· IEEE Transactions on Mobile...· 0 citations
Collaborative inference deploys Large Vision-Language Models (LVLMs) by partitioning computation between edge devices and the cloud. While withholding raw inputs supposedly ensures privacy, transmitting intermediate hidden states exposes a critical attack surface. However, it remains unclear whether deep-layer LVLM hidden states retain recoverable private information, given that visual content has been projected into the language embedding space. To address this concern, we theoretically analyze LVLM hidden-state recoverability and show that, under regularity assumptions and a positive semantic--nuisance margin, privacy-relevant visual semantics remain identifiable and stably recoverable. Motivated by this analysis, we propose RASR, a novel coarse-to-fine multimodal reconstruction attack. RASR obtains initial image and text reconstructions through modality-specific inverse paths that follow their respective forward processing pipelines in reverse, and then uses hidden-state consistency to refine both reconstructions. Evaluations on Qwen3-VL-8B-Instruct and LLaVA-1.5-7B across five datasets demonstrate that RASR reduces image reconstruction MSE by \(\sim\)50\% compared to the strongest baselines, while achieving up to 99\% token accuracy for text recovery. These results show that privacy-sensitive visual and textual information can be recovered even from deep-layer LVLM hidden states, exposing the privacy risks of collaborative inference.
Shuaifan Jin, Zhibo Wang, Qiyuan Wang et al.· 0 citations
DeePCAP introduces a query-driven fidelity framework spanning packet- and flow-level queries to tackle the fidelity disconnection, and proposes a novel dimensionality reduction approach using frequency domain encoding to improve cost-fidelity trade-off.
Fenghao Dong, Yucheng Yin, Yajie Zhou et al.· Conference on Applications,...· 0 citations
This position paper outlines a common abstraction layer that can substantially lower the effort for designing agents with constrained creativity, and demonstrates early promise of this paradigm for two SysOps use cases: Root Cause Analysis and Network Configuration Generation.
†. SayanSinha, Vipul Harsh, Yajie Zhou et al.· 0 citations
TabQueryBench is a query-centric benchmark that uses SQL-shaped analytical queries as structural assessors for synthetic data fidelity, and provides an extensible foundation for query-centric synthetic-data evaluation.
