HYPPO: A Surrogate-Based Multi-Level Parallelism Tool for Hyperparameter Optimization

TL;DR

HYPPO is a software tool that uses surrogate models and parallelism to efficiently optimize hyperparameters in deep learning, reducing computational costs and improving robustness.

Contribution

HYPPO introduces an adaptive surrogate-based approach with asynchronous nested parallelism for hyperparameter optimization in deep learning models.

Findings

Reduces hyperparameter evaluation count by an order of magnitude.

Decreases HPO process throughput time by two orders of magnitude.

Demonstrates effectiveness on time-series, image classification, and CT reconstruction tasks.

Abstract

We present a new software, HYPPO, that enables the automatic tuning of hyperparameters of various deep learning (DL) models. Unlike other hyperparameter optimization (HPO) methods, HYPPO uses adaptive surrogate models and directly accounts for uncertainty in model predictions to find accurate and reliable models that make robust predictions. Using asynchronous nested parallelism, we are able to significantly alleviate the computational burden of training complex architectures and quantifying the uncertainty. HYPPO is implemented in Python and can be used with both TensorFlow and PyTorch libraries. We demonstrate various software features on time-series prediction and image classification problems as well as a scientific application in computed tomography image reconstruction. Finally, we show that (1) we can reduce by an order of magnitude the number of evaluations necessary to find the most optimal region in the hyperparameter space and (2) we can reduce by two orders of magnitude the throughput for such HPO process to complete.