Boosting Image Recognition with Non-differentiable Constraints

TL;DR

This paper introduces a reinforcement learning method to incorporate non-differentiable, rule-based constraints into image recognition models, significantly improving accuracy and robustness in digit sequence recognition tasks.

Contribution

It proposes a novel policy gradient reinforcement learning approach to effectively integrate non-differentiable rules into image recognition models, enhancing performance with limited data.

Findings

Increased accuracy by up to 23.6% on MNIST-based datasets.

Improved robustness against synthesized adversarial examples.

Effective rule-based inductive bias with limited training data.

Abstract

In this paper, we study the problem of image recognition with non-differentiable constraints. A lot of real-life recognition applications require a rich output structure with deterministic constraints that are discrete or modeled by a non-differentiable function. A prime example is recognizing digit sequences, which are restricted by such rules (e.g., \textit{container code detection}, \textit{social insurance number recognition}, etc.). We investigate the usefulness of adding non-differentiable constraints in learning for the task of digit sequence recognition. Toward this goal, we synthesize six different datasets from MNIST and Cropped SVHN, with three discrete rules inspired by real-life protocols. To deal with the non-differentiability of these rules, we propose a reinforcement learning approach based on the policy gradient method. We find that incorporating this rule-based reinforcement can effectively increase the accuracy for all datasets and provide a good inductive bias which improves the model even with limited data. On one of the datasets, MNIST\_Rule2, models trained with rule-based reinforcement increase the accuracy by 4.7\% for 2000 samples and 23.6\% for 500 samples. We further test our model against synthesized adversarial examples, e.g., blocking out digits, and observe that adding our rule-based reinforcement increases the model robustness with a relatively smaller performance drop.