TAPAS: Two-pass Approximate Adaptive Sampling for Softmax

TL;DR

TAPAS introduces a two-pass adaptive sampling method for softmax models that efficiently approximates gradients, improving multi-class classification with large label spaces.

Contribution

It proposes a novel two-pass sampling strategy for softmax, with an efficient distributed implementation, enhancing performance on large-scale classification tasks.

Findings

Low computational overhead demonstrated on synthetic and real data.

Effective in minimizing rank loss for large label spaces.

Works well for multi-class classification problems.

Abstract

TAPAS is a novel adaptive sampling method for the softmax model. It uses a two pass sampling strategy where the examples used to approximate the gradient of the partition function are first sampled according to a squashed population distribution and then resampled adaptively using the context and current model. We describe an efficient distributed implementation of TAPAS. We show, on both synthetic data and a large real dataset, that TAPAS has low computational overhead and works well for minimizing the rank loss for multi-class classification problems with a very large label space.