WaveAttack: Asymmetric Frequency Obfuscation-based Backdoor Attacks Against Deep Neural Networks (Machine Learning Top Conference, CCF-A, Acceptance Ratio: 25%)

Jun Xia, Zhihao Yue, Yingbo Zhou, Zhiwei Ling, Yiyu Shi, Xian Wei, Mingsong Chen

Project

• We obtain image high-frequency features through Discrete Wavelet Transform (DWT) to generate backdoor triggers.

Towards Energy-Aware Federated Learning via MARL: A Dual-Selection Approach for Model and Client (EDA Top Conference, CCF-B, Acceptance Ratio: 24%)

Jun Xia, Yi Zhang, Yiyu Shi

Project

• We propose an energy-aware FL framework named DR-FL, which considers the energy constraints in both clients and heterogeneous deep learning models to enable energy-efficient FL. Unlike Vanilla FL, DR-FL adopts our proposed Muti-Agents Reinforcement Learning (MARL)-based dual-selection method, which allows participated devices to make contributions to the global model effectively and adaptively based on their computing capabilities and energy capacities in a MARL-based manner.

Eliminating backdoor triggers for deep neural networks using attention relation graph distillation (AI Top Conference, CCF-A, Acceptance Ratio: 14%)

Jun Xia, Ting Wang, Jiepin Ding, Xian Wei, Mingsong Chen

Project

• We introduce a novel backdoor defense framework named Attention Relation Graph Distillation (ARGD), which fully explores the correlation among attention features with different orders using our proposed Attention Relation Graphs (ARGs). Based on the alignment of ARGs between both teacher and student models during knowledge distillation, ARGD can eradicate more backdoor triggers than NAD.

Accelerated synthesis of neural network-based barrier certificates using collaborative learning (EDA Top Conference, CCF-A, Acceptance Ratio: 23%)

Jun Xia, Ming Hu, Xin Chen, Mingsong Chen

Project

• We fully exploit the parallel processing capability of underlying hardware to enable quick search for a barrier certificate.

PervasiveFL: Pervasive federated learning for heterogeneous IoT systems (EDA Top Journal, CCF-A, Acceptance Ratio: 23%)

Jun Xia, Tian Liu, Zhiwei Ling, Ting Wang, Xin Fu, Mingsong Chen

Project

• we propose a novel framework named PervasiveFL that enables efficient and effective FL among heterogeneous IoT devices. Without modifying original local models, PervasiveFL installs one lightweight NN model named modellet on each device. By using the deep mutual learning (DML) and our entropy-based decision gating (EDG) method, modellets and local models can selectively learn from each other through soft labels.

Enabling On-Device Large Language Model Personalization with Self-Supervised Data Selection and Synthesis(EDA Top Conference, CCF-A iter,, Acceptance Ratio: 25%)

Ruiyang Qin, Jun Xia, Zhenge Jia, Meng Jiang, Ahmed Abbasi, Peipei Zhou, Jingtong Hu, Yiyu Shi

Project

• In this paper, we propose a novel framework to select and store the most representative data online in a self-supervised way. Such data has a small memory footprint and allows infrequent requests of user annotations for further fine-tuning.

Model-Contrastive Learning for Backdoor Elimination (Multi-Media Top Conference, CCF-A, Acceptance Ratio: 25%)

Zhihao Yue, Jun Xia, Zhiwei Ling, Ming Hu, Ting Wang, Xian Wei, Mingsong Chen

Project

• In this paper, we propose a novel two-stage backdoor defense method, named MCLDef, based on Model-Contrastive Learning (MCL). MCLDef can purify the backdoored model by pulling the feature representations of poisoned data towards those of their clean data counterparts.