We propose a backward-mode AD proxy using only forward passes applying to Hamiltonian recurrent units and stacks thereof (namely, SSMs) with theoretical guarantees and experimental evidence

We propose to treat LLM activations as images and detect hallucinations more accurately (and efficiently) across LLMs with a vision-inspired architecture.

AsymDSD is a self-supervised learning framework for 3D point clouds that combines masked modeling and invariance learning through latent space prediction.