Simplex Random Features
Reid, Isaac, Choromanski, Krzysztof, Likhosherstov, Valerii, Weller, Adrian
We present Simplex Random Features (SimRFs), a new random feature (RF) mechanism for unbiased approximation of the softmax and Gaussian kernels by geometrical correlation of random projection vectors. We prove that SimRFs provide the smallest possible mean square error (MSE) on unbiased estimates of these kernels among the class of weight-independent geometrically-coupled positive random feature (PRF) mechanisms, substantially outperforming the previously most accurate Orthogonal Random Features at no observable extra cost. We present a more computationally expensive SimRFs+ variant, which we prove is asymptotically optimal in the broader family of weight-dependent geometrical coupling schemes (which permit correlations between random vector directions and norms). In extensive empirical studies, we show consistent gains provided by SimRFs in settings including pointwise kernel estimation, nonparametric classification and scalable Transformers.
Oct-7-2023
- Country:
- Africa > Ethiopia (0.04)
- Asia > China
- Europe
- Austria (0.04)
- France (0.04)
- Germany > Hesse
- Darmstadt Region > Darmstadt (0.04)
- Serbia > Southern and Eastern Serbia
- Nišava District > Niš (0.04)
- Spain > Canary Islands (0.04)
- United Kingdom > England
- Cambridgeshire > Cambridge (0.04)
- Indian Ocean > Bass Strait (0.04)
- North America
- Canada
- British Columbia (0.04)
- Quebec > Montreal (0.04)
- Mexico > Gulf of Mexico (0.04)
- United States
- Arizona (0.04)
- California > Los Angeles County
- Long Beach (0.14)
- Florida > Miami-Dade County
- Miami (0.04)
- Georgia > Fulton County
- Atlanta (0.04)
- Hawaii > Honolulu County
- Honolulu (0.04)
- Massachusetts > Middlesex County
- Cambridge (0.04)
- Canada
- Oceania > Australia
- Tasmania (0.04)
- Genre:
- Research Report (0.64)
- Technology: