GRADSTOP: Early Stopping of Gradient Descent via Posterior Sampling presented at ECAI 2025

The scope of the conference covered core AI topics including machine learning, computer vision, natural language processing, knowledge representation and reasoning, planning and search, as well as multidisciplinary work. Katsiaryna and Arash are delighted that their work was accepted for an oral presentation, and they got a chance to receive feedback from AI community. 

When discussing their research, Katsiaryna stated “we provided a theoretically justified method for early stopping that does not need a validation set, allows using all available data for training, and relies only on the gradients provided by gradient descent algorithm. The method is beneficial in data scarce setting, for example, in processing of medical data."

From the abstract "Machine learning models are often learned by minimising a loss function on the training data using a gradient descent algorithm. These models often suffer from overfitting, leading to a decline in predictive performance on unseen data. A standard solution is early stopping using a hold-out validation set, which halts the minimisation when the validation loss stops decreasing. However, this hold-out set reduces the data available for training. This paper presents GRADSTOP, a novel stochastic early stopping method that only uses information in the gradients, which are produced by the gradient descent algorithm “for free.” Our main contributions are that we estimate the Bayesian posterior by the gradient information, define the early stopping problem as drawing sample from this posterior, and use the approximated posterior to obtain a stopping criterion. Our empirical evaluation shows that GRADSTOP achieves a small loss on test data and compares favourably to a validation-set-based stopping criterion. By leveraging the entire dataset for training, our method is particularly advantageous in data-limited settings, such as transfer learning. It can be incorporated as an optional feature in gradient descent libraries with only a small computational overhead. The source code is available at https://github.com/edahelsinki/gradstop."

Link to the paper: https://ebooks.iospress.nl/doi/10.3233/FAIA251043

More information about ECAI 2025 can be found https://ecai2025.org/