Amey Pawar

Intel

May 2020 – Present

Santa Clara, CA

Applied Machine Learning Intern

• Worked with Applied ML team to perform hyperparameter optimization, real-time metrics visualization, and distributed training with Intel MPI on CPU clusters to improve the accuracy and time to converge for existing interactive image segmentation model. • Implemented the CVPR - 2019 paper which aims to improve the accuracy of an interactive segmentation model during inference time by backpropagating through the network in Pytorch and integrated this model agnostic approach with existing segmentation model to improve the mean and boundary IOU of segmentation mask. • Designed a novel NAS Enhanced Encoder-Decoder (NEED) based Image Segmentation model as a part of Applied ML team, which achieved results comparable to the State of the Art (SOTA) and submitted the paper to ICCV - 2021. • Developed Learning Curve Extrapolation technique in combination with various other pruning methods such as Successive Halving and Hyperband along with Bayesian Optimization in Optuna framework, in order to come up with better ensemble pruning strategy. • Currently focusing on application of Proxyless Neural Architecture Search (NAS) Technique as a part of hardware-aware component of the deep learning workflow

Skills