Marcin Kurdziel

Neural Representations Reveal Distinct Modes of Class Fitting in Residual Convolutional Networks

2022 · arXiv (Cornell University)

We leverage probabilistic models of neural representations to investigate how residual networks fit classes. To this end, we estimate class-conditional density models for representations learned by deep ResNets. We then use these models to characterize …

Neural Representations Reveal Distinct Modes of Class Fitting in Residual Convolutional Networks

2023 · Proceedings of the AAAI Conference on Artificial Intelligence

We leverage probabilistic models of neural representations to investigate how residual networks fit classes. To this end, we estimate class-conditional density models for representations learned by deep ResNets. We then use these models to characterize …