cv

PhD in Computer Science

University of Texas at Dallas, Texas, USA

• Developing neurosymbolic systems for enhancing interpretability of deep learning models.
• Combining LLMs and symbolic reasoing using logic programming for reliable natural language generation.

MS in Computer Science

University of Texas at Dallas, Texas, USA

B.Tech. in Instrumentation and Control Engineering

National Institute of Technology Jalandhar, Punjab, India

Computer Vision Intern

Tech For Good Inc., Boston, MA, USA

• Coordinated a team to annotate a 5,000-image dataset of firearms in active shooter scenarios, achieving 90% accuracy after experimenting with various object detection models such as YOLO, FastRCNN and FasterRCNN.

Research Analyst

Schizophrenia and Social Cognition lab, The University of Texas at Dallas, TX, USA

• Worked on analyzing schizophrenic patient data and develpoing a ML model to predict the occurance of the disease in subjects with an 89% accuracy.

Research Intern

IIM Amritsar, India

• Created a software by integrating DEMATEL, MMDE, and ISM decision-making techniques to find the degree of impact of the enablers and barriers to sustainable manufacturing.

NeSyFOLD: A Framework for Interpretable Image Classification

Parth Padalkar, Huaduo Wang, Gopal Gupta @AAAI 2024, oral presentation (<4\% selection rate)

• Introduced a neurosymbolic framework, NeSyFOLD, aimed at creating interpretable predictions for image classification tasks, using Convolutional Neural Networks (CNNs).
• A rule-set generated from the CNN, along with the CNN serves as the interpretable model for making predictions. Showed an average increase of 8% in accuracy and an 83% reduction in rule-set size than previous SOTA.

Using Logic Programming and Kernel-Grouping for Improving Interpretability of Convolutional Neural Networks

Parth Padalkar, Huaduo Wang, Gopal Gupta, @Practical Aspects of Declarative Languages (PADL) 2024

• Improved on NeSyFOLD. Developed a novel algorithm for grouping the outputs of similar kernels in the CNN.
• Showed that using groups of kernels for generating a rule-set leads to comparable performance and a 14% drop in the rule-set size on average, than using individual kernels.

Automated interactive domain-specific conversational agents that understand human dialogs

Yankai Zeng, Abhiramon Rajasekharan, Parth Padalkar, et a., @Practical Aspects of Declarative Languages (PADL) 2024

• Developed a chat-bot using LLMs and logic programming that is more reliable than using an LLM-based chatbot.
• Showed application as a hotel concierge that can recommend restaurants with more reliability than Bing AI.

Reliable Natural Language Understanding with Large Language Models and Answer Set Programming

Abhiramon Rajasekharan, Yankai Zeng, Parth Padalkar, Gopal Gupta, @International Conference on Logic Programing (ICLP) 2023

• Proposed STAR, a framework that combines LLMs with Answer Set Programming (ASP) to improve reasoning in natural language understanding tasks.
• Applied the STAR framework to tasks involving qualitative reasoning, mathematical reasoning, and goal-directed conversations and demonstrated its superior performance to vanilla LLMs.