Outlier Gradient Analysis: Efficiently Identifying Detrimental Training Samples for Deep Learning Models

Anshuman Chhabra, Bo Li, Jian Chen, Prasant Mohapatra, Hongfu Liu

Abstract: A core data-centric learning challenge is the identification of training samples that are detrimental to model performance. Influence functions serve as a prominent tool for this task and offer a robust framework for assessing training data influence on model predictions. Despite their widespread use, their high computational cost associated with calculating the inverse of the Hessian matrix pose constraints, particularly when analyzing large-sized deep models. In this paper, we establish a bridge between identifying detrimental training samples via influence functions and outlier gradient detection. This transformation not only presents a straightforward and Hessian-free formulation but also provides insights into the role of the gradient in sample impact. Through systematic empirical evaluations, we first validate the hypothesis of our proposed outlier gradient analysis approach on synthetic datasets. We then demonstrate its effectiveness in detecting mislabeled samples in vision models and selecting data samples for improving performance of natural language processing transformer models. We also extend its use to influential sample identification for fine-tuning Large Language Models.

Lay Summary:

When training AI models, not all examples are helpful — some can confuse the model or even make it worse. These “bad” training examples can be challenging to identify, especially in massive datasets used for modern deep learning models. Existing tools for finding them are either not very accurate or too slow to use with today’s large models.

In this work, we introduce a faster, simpler method called Outlier Gradient Analysis. Instead of relying on heavy computations, we look at how each training example influences the model’s learning process through its gradient — a kind of learning signal. By identifying outliers in these gradients, we can spot training examples that are likely to hurt performance.

We test our method on a range of tasks, like identifying mislabeled images and finding beneficial data for large language models. Outlier Gradient Analysis is highly effective across different domains, and offers a practical way of identifying errors in training datasets and building more reliable AI systems.

Primary Area: Deep Learning->Everything Else

Keywords: Data-centric learning, Detrimental sample detection

Submission Number: 7375