Discovery of Optimal Settings for State-Of-Art Deep Learning Models

Date: 01/07/2021
Client: IIT Madras in collab. with Intel AI Labs

Heuristics for Auto-Tuning CNNs:

Developed heuristics for auto-tuning convolutional neural nets (CNNs) to improve state-of-the-art model performance on image segmentation tasks.

Created statistical framework (design-of-experiments with response surface ANOVA) to find optimise hyperparameter settings for models on the Brain Tumor Segmentation (BraTS2020) dataset.

Performance Improvements:

Enhanced the state-of-the-art model accuracy on the BraTS 2020 dataset by 5%, achieved with just one NVidia V100 32 GB GPU, with single execution over 50 hours.

Research Impact:

The developed framework is actively used in a collaboration between the medical imaging lab at IIT Madras and the Intel AI Lab Chennai for ongoing experiments on the BraTS datasets.