Speaker: Xiuwen Liu
Date: Sep 1, 2023
Abstract: Pretrained large foundation models play a central role in the recent surge of artificial intelligence, resulting in finetuned models with remarkable abilities when measured on benchmark datasets, standard exams, and applications. Due to their inherent complexity, these models are poorly understood. While small adversarial inputs to such models are well known, the structures of the representation space are not well characterized despite their fundamental importance. In this talk, I will discuss the representation space of transformers and show their vulnerabilities to adversarial attacks. Based on local directional Lipschitz constant estimation techniques, we propose an effective framework to characterize and explore the embedding spaces of deployed large models. More specifically, using the vision transformers as an example due to the continuous nature of their input space, we show via analyses and systematic experiments that the representation space consists of large piecewise linear subspaces where there exist very different inputs sharing the same representations, and at the same time, local normal spaces where there are visually indistinguishable inputs having very different representations. The empirical results are further verified using the local directional estimations of the Lipschitz constants of the underlying models. The work is done jointly with Shaeke Salman and Md Montasir Bin Shams.
Links and comments:For the zoom option, please use https://fsu.zoom.us/j/8506440050.