Aravindhan Saravanan

Title: Aravindhan Saravanan and Autonomous Vision Pipeline Composition

Introduction

Aravindhan Saravanan is an inventor based in Chennai, India, whose patent work addresses the automated design of computer-vision workflows. His disclosed invention combines symbolic planning, transformer architecture, and deep reinforcement learning to help compose vision pipelines for specific tasks.

Latest Patents

Saravanan is associated with one patent, titled “Methods and Systems for Autonomous Task Composition of Vision Pipelines Using an Algorithm Selection Framework.” The disclosure describes a two-stage process. First, a symbolic planner generates a high-level sequence of tasks to form a vision workflow. Then, a graph search using a transformer architecture selects suitable algorithms for each task from an algorithmic space.

To improve the system’s robustness, the approach updates the visual transformer’s embedding, key, and query network weights through deep reinforcement learning. It uses Proximal Policy Optimization (PPO) as the underlying algorithm. The framework is designed to search for solution templates that were not previously encountered.

Career Highlights

Saravanan works at Tata Consultancy Services Limited. His patent reflects work at the intersection of computer vision, algorithm selection, transformer models, and reinforcement learning. The disclosed framework aims to automate the composition of vision pipelines rather than require each workflow to be assembled entirely by hand.

Collaborations

The listed coworkers associated with Saravanan include Abhishek Roy Choudhury and Vighnesh Vatsal. Their inclusion provides context for the professional environment surrounding his work.

Conclusion

Aravindhan Saravanan’s patent presents a structured approach to building vision workflows by combining symbolic planning with transformer-based algorithm search and reinforcement learning. His work at Tata Consultancy Services Limited highlights an effort to make vision pipeline composition more autonomous and adaptable.