A Unified Framework of Surrogate Loss by Refactoring and Interpolation

TL;DR

UniLoss is a unified framework that simplifies the creation of surrogate losses for deep learning by refactoring performance evaluation into differentiable steps, enabling versatile optimization across tasks.

Contribution

The paper introduces UniLoss, a novel framework that reduces manual design of surrogate losses by refactoring metrics into differentiable components via interpolation.

Findings

Achieves comparable performance to task-specific losses.

Validates effectiveness on three tasks and four datasets.

Provides a general approach for surrogate loss generation.

Abstract

We introduce UniLoss, a unified framework to generate surrogate losses for training deep networks with gradient descent, reducing the amount of manual design of task-specific surrogate losses. Our key observation is that in many cases, evaluating a model with a performance metric on a batch of examples can be refactored into four steps: from input to real-valued scores, from scores to comparisons of pairs of scores, from comparisons to binary variables, and from binary variables to the final performance metric. Using this refactoring we generate differentiable approximations for each non-differentiable step through interpolation. Using UniLoss, we can optimize for different tasks and metrics using one unified framework, achieving comparable performance compared with task-specific losses. We validate the effectiveness of UniLoss on three tasks and four datasets. Code is available at https://github.com/princeton-vl/uniloss.