Transferability and Hardness of Supervised Classification Tasks

TL;DR

This paper introduces an information-theoretic method to estimate the difficulty and transferability of supervised classification tasks without relying on trained models, validated on large datasets with strong correlations to empirical results.

Contribution

It presents a novel, solution-agnostic approach using label statistics to predict task hardness and transferability, advancing understanding of task relationships in supervised learning.

Findings

Strong correlation between estimated and actual transferability.

Effective prediction of task difficulty without trained models.

Achieved state-of-the-art accuracy in transfer experiments.

Abstract

We propose a novel approach for estimating the difficulty and transferability of supervised classification tasks. Unlike previous work, our approach is solution agnostic and does not require or assume trained models. Instead, we estimate these values using an information theoretic approach: treating training labels as random variables and exploring their statistics. When transferring from a source to a target task, we consider the conditional entropy between two such variables (i.e., label assignments of the two tasks). We show analytically and empirically that this value is related to the loss of the transferred model. We further show how to use this value to estimate task hardness. We test our claims extensively on three large scale data sets -- CelebA (40 tasks), Animals with Attributes 2 (85 tasks), and Caltech-UCSD Birds 200 (312 tasks) -- together representing 437 classification tasks. We provide results showing that our hardness and transferability estimates are strongly correlated with empirical hardness and transferability. As a case study, we transfer a learned face recognition model to CelebA attribute classification tasks, showing state of the art accuracy for tasks estimated to be highly transferable.