Parthe Pandit

Implicit Bias of Linear RNNs

2021 · arXiv (Cornell University)

Contemporary wisdom based on empirical studies suggests that standard recurrent neural networks (RNNs) do not perform well on tasks requiring long-term memory. However, precise reasoning for this behavior is still unknown. This paper provides a …

On the Nystrom Approximation for Preconditioning in Kernel Machines

2023 · arXiv (Cornell University)

Kernel methods are a popular class of nonlinear predictive models in machine learning. Scalable algorithms for learning kernel models need to be iterative in nature, but convergence can be slow due to poor conditioning. Spectral …