Publications

Oral Presentation AAAI-26 | Singapore

"Tractable Sharpness-Aware Learning of Probabilistic Circuits"

Hrithik Suresh, Sahil Sidheekh, Vishnu Shreeram M. P., Sriraam Natarajan, Narayanan C. Krishnan

This work presents a hybrid training method for Probabilistic Circuits that combines the generalizability of EM with the speed of SGD. We implement a Hessian-based regularizer to guide models toward flatter optima, reducing overfitting. We prove that the Hessian trace (sharpness proxy) is efficiently computable for PCs, enabling closed-form updates.

Patents

Patent Filed Application No: 202541068350 | Filed: July 17, 2025

"Method, Apparatus, and System for Increasing Throughput in a Surface Mount Technology Assembly Line"

This invention presents a novel method for optimizing feeder arrangements in SMT pick-and-place machines. The algorithm increases components-per-hour (CPH) while meeting all manufacturing constraints. The approach solves combinatorial optimisation problem using Genetic Algorithms, Mixed-Integer Linear Programming (MILP), and Nonlinear Programming techniques.

Honors & Awards

GATE (DA)

AIR 73 (2025) & AIR 155 (2026)

Data Science and Artificial Intelligence

Academic Excellence

Highest CGPA

Among Data Science students for 1st Year (2022-'23), 2nd Year (2023-'24), and 3rd Year (2024-'25) at IIT Palakkad

Inter IIT Tech Meet 13.0

12th Place (out of 23 IITs)

Representing IIT Palakkad - Dynamic Agentic RAG for Information Retrieval