Smooth Loss Functions for Deep Top-k Classification

TL;DR

This paper introduces smooth, non-sparse gradient loss functions tailored for deep top-k classification, demonstrating improved robustness over cross-entropy in noisy and limited data scenarios.

Contribution

The authors propose a family of smooth loss functions for top-k deep learning, with an efficient algorithm leveraging polynomial algebra and GPU approximations.

Findings

Our loss functions outperform cross-entropy in noisy data conditions.

The proposed algorithm reduces computational complexity to O(k n).

Smooth loss functions show increased robustness to overfitting.

Abstract

The top-k error is a common measure of performance in machine learning and computer vision. In practice, top-k classification is typically performed with deep neural networks trained with the cross-entropy loss. Theoretical results indeed suggest that cross-entropy is an optimal learning objective for such a task in the limit of infinite data. In the context of limited and noisy data however, the use of a loss function that is specifically designed for top-k classification can bring significant improvements. Our empirical evidence suggests that the loss function must be smooth and have non-sparse gradients in order to work well with deep neural networks. Consequently, we introduce a family of smoothed loss functions that are suited to top-k optimization via deep learning. The widely used cross-entropy is a special case of our family. Evaluating our smooth loss functions is computationally challenging: a na\"ive algorithm would require $\mathcal{O}(\binom{n}{k})$ operations, where n is the number of classes. Thanks to a connection to polynomial algebra and a divide-and-conquer approach, we provide an algorithm with a time complexity of $\mathcal{O}(k n)$. Furthermore, we present a novel approximation to obtain fast and stable algorithms on GPUs with single floating point precision. We compare the performance of the cross-entropy loss and our margin-based losses in various regimes of noise and data size, for the predominant use case of k=5. Our investigation reveals that our loss is more robust to noise and overfitting than cross-entropy.