Ishank Goel

Title: Ishank Goel: Inventor in Model Optimization

Introduction

Ishank Goel is an inventor based in Pune, India. His work includes a patent for a two-step approach to optimizing deep learning models.

Latest Patents

Goel is associated with 1 patent, titled “Method and System for Two-Step Hierarchical Model Optimization.” The invention describes a method that combines an automated pruning, weight clustering, and quantization technique (AutoPCQ) with conditional use of automated knowledge distillation (AKD).

AutoPCQ frames the selection of model configurations as an optimization problem, using Bayesian optimization and reinforcement learning iteratively. When needed, AKD searches for a student model to learn from a teacher model, within a restricted neural network architecture search space. The approach aims to automate model optimization efficiently while preserving accuracy.

Career Highlights

Goel is associated with Tata Consultancy Services Limited. His patented work addresses the challenge of reducing manual intervention in deep learning model optimization.

Collaborations

The provided information lists Amit Kalele and Ravindran Subbiah as Goel’s coworkers.

Conclusion

Ishank Goel’s patent presents an automated, hierarchical approach to deep learning model optimization, combining configuration search with conditional knowledge distillation.