Hongjie Wang

Hongjie Wang, Difan Liu, Yan Kang et al.

Diffusion Models (DMs) have exhibited superior performance in generating high-quality and diverse images. However, this exceptional performance comes at the cost of expensive architectural design, particularly due to the attention module heavily used in leading models. Existing works mainly adopt a retraining process to enhance DM efficiency. This is computationally expensive and not very scalable. To this end, we introduce the Attention-driven Training-free Efficient Diffusion Model (AT-EDM) framework that leverages attention maps to perform run-time pruning of redundant tokens, without the need for any retraining. Specifically, for single-denoising-step pruning, we develop a novel ranking algorithm, Generalized Weighted Page Rank (G-WPR), to identify redundant tokens, and a similarity-based recovery method to restore tokens for the convolution operation. In addition, we propose a Denoising-Steps-Aware Pruning (DSAP) approach to adjust the pruning budget across different denoising timesteps for better generation quality. Extensive evaluations show that AT-EDM performs favorably against prior art in terms of efficiency (e.g., 38.8% FLOPs saving and up to 1.53x speed-up over Stable Diffusion XL) while maintaining nearly the same FID and CLIP scores as the full model. Project webpage: https://atedm.github.io.

Gokcen Y. Dayanikli, Rees R. Hatch, Ryan M. Gerdes et al.

Alleviating range anxiety for electric vehicles (i.e., whether such vehicles can be relied upon to travel long distances in a timely manner) is critical for sustainable transportation. Extremely fast charging (XFC), whereby electric vehicles (EV) can be quickly recharged in the time frame it takes to refuel an internal combustion engine, has been proposed to alleviate this concern. A critical component of these chargers is the efficient and proper operation of power converters that convert AC to DC power and otherwise regulate power delivery to vehicles. These converters rely on the integrity of sensor and actuation signals. In this work the operation of state-of-the art XFC converters is assessed in adversarial conditions, specifically against Intentional Electromagnetic Interference Attacks (IEMI). The targeted system is analyzed with the goal of determining possible weak points for IEMI, viz. voltage and current sensor outputs and gate control signals. This work demonstrates that, with relatively low power levels, an adversary is able to manipulate the voltage and current sensor outputs necessary to ensure the proper operation of the converters. Furthermore, in the first attack of its kind, it is shown that the gate signal that controls the converter switches can be manipulated, to catastrophic effect; i.e., it is possible for an attacker to control the switching state of individual transistors to cause irreparable damage to the converter and associated systems. Finally, a discussion of countermeasures for hardware designers to mitigate IEMI-based attacks is provided.