Hybrid Random Features

Krzysztof Marcin Choromanski, Han Lin, Haoxian Chen, Arijit Sehanobish, Yuanzhe Ma, Deepali Jain, Jake Varley, Andy Zeng, Michael S Ryoo, Valerii Likhosherstov, Dmitry Kalashnikov, Vikas Sindhwani, Adrian Weller

Keywords: random features, softmax kernel, attention mechanism, compositional kernels

Abstract: We propose a new class of random feature methods for linearizing softmax and Gaussian kernels called hybrid random features (HRFs) that automatically adapt the quality of kernel estimation to provide most accurate approximation in the defined regions of interest. Special instantiations of HRFs lead to well-known methods such as trigonometric (Rahimi & Recht, 2007) or (recently introduced in the context of linear-attention Transformers) positive random features (Choromanski et al., 2021). By generalizing Bochner’s Theorem for softmax/Gaussian kernels and leveraging random features for compositional kernels, the HRF-mechanism provides strong theoretical guarantees - unbiased approximation and strictly smaller worst-case relative errors than its counterparts. We conduct exhaustive empirical evaluation of HRF ranging from pointwise kernel estimation experiments, through tests on data admitting clustering structure to benchmarking implicit-attention Transformers (also for downstream Robotics applications), demonstrating its quality in a wide spectrum of machine learning problems.

One-sentence Summary: We propose a new class of random feature methods for softmax and Gaussian kernel estimation that are adaptable to provide particularly accurate approximation in the desired regions of interest.