Caiwen Ding

Multi-task Convolutional Neural Network (CNN) inference on Energy Harvesting (EH) devices has received limited attention, particularly in scenarios where energy availability fluctuates significantly. This poses substantial implementation challenges, especially in deploying and running inference under varying energy conditions while performing multiple correlated tasks. In this work, we present an end-to-end framework that enables multi-task inference to inherently conserve computation under dynamic energy conditions. To facilitate this adaptation in multi-task settings, we propose inter-task and intra-task shared-weight design strategies. The inter-task design removes redundancy across related tasks, while the intra-task design targets redundancy within weight matrices. Finally, to enable efficient inference on resource-constrained devices, we compress the model and propose novel algorithms that support energy-aware adaptation and compression-tolerant computation without expensive weight reconstruction. Experimental results show that our proposed framework can enable up to 1.8x inference latency reduction, 48% energy efficiency improvement, and 3.5x memory conservation.

Recent advances in adversarial machine learning have shown that defenses previously considered robust are actually susceptible to adversarial attacks which are specifically customized to target their weaknesses. However, whether the adversarial examples generated by customized attacks, are effective on other defenses, is an open question. In this work we seek to explore three important security questions: First, do different defense strategies exhibit the same low transferability properties as different model architectures and, if so, how can this low transferability be utilized to improve robustness? Second, how can a white-box adversary design attacks to specifically thwart multi-defense based setups? Last, how can game theoretic analysis further improve the robustness against an adversary capable of implementing multiple state-of-the-art attacks? To this end we provide multiple contributions, including the first transferability study between multiple defense strategies, three new attack algorithms designed to break random transform and ensemble defenses, and two game theoretic frameworks for analyzing and optimizing robustness over a combination of adversarial attacks and defenses. Empirically, we show our framework is 18% more robust on CIFAR-10 and is 27% more robust on Tiny-ImageNet than the best single state-of-the-art defense that we analzye.

Integration of memristors into neuromorphic systems is receiving substantial attention due to their potential to facilitate energy-efficient and highly parallel in-memory computation. In this paper, a memristor-bridge based design for the convolution operation of convolution neural network (CNN) and its crossbar realization are developed. A LeNet-5 network is realized using the proposed design and tested on the MNIST dataset. The architecture includes circuit configurations for activation and pooling operations. The weight-mapping procedure for the memristor-bridges is developed in relation to the exact physics of conduction mechanism of memristor. Efficient modeling of the devices results in excellent performance of the network, achieving up to 99.08% inference accuracy. Isolation among the bridges and parallelization of the convolution operation leads to a rapid mapping within 0.11[ ][ ] and fast response in less than 20[ ][ ]. The overall energy consumption by the memristor units during mapping and inference remains well below [ ].