Research Publications

Contributions to Machine Learning, AI Alignment, and Network Science

Published Work

Peer-reviewed publications in top-tier conferences

Metatuning: A Novel Lightweight Adaptation Framework for Aligning Large Language Models

Aniruddha Chattopadhyay, Kaushik Roy (Asst. prof University of Alabama)

Neurosymbolic Learning and Reasoning Conference (NeSy 2025)May 2025Accepted at NeSy 2025; proceedings to appear in the Journal of Machine Learning Research (JMLR)

First-author long paper on Metatuning, a novel lightweight adaptation framework for aligning large language models (LLMs) with symbolic reasoning objectives. Introduces metatuning as a middle ground between few-shot prompting and full fine-tuning, enabling efficient structural alignment on consumer hardware.

EduTree: An Academic Genealogy Graph (AGG) Modeling Mentorship Lineages

Aniruddha Chattopadhyay, et al.

ACM/IEEE Joint Conference on Digital Libraries (JCDL 2020)Aug. 2020Best M.Tech Thesis Project for contributions to network science and machine learning

Presented EduTree, an academic genealogy graph (AGG) modeling mentorship lineages and institutional influence within the field of education. Applied graph-theoretic centrality measures and topic modeling to quantify researcher impact and trace the evolution of research clusters. Identified high-centrality mentors, pioneering institutions, and thematic trajectories shaping the discipline's academic network.

Research Interests

Current Focus

  • Multimodal AI systems combining vision, language, and audio modalities
  • Efficient adaptation methods for large language models
  • AI safety and alignment in generative systems
  • Real-time inference optimization for edge deployment

Future Directions

I'm particularly interested in pursuing graduate research that bridges the gap between theoretical advances in AI and practical deployment challenges. My goal is to contribute to making AI systems more efficient, reliable, and accessible for real-world applications.