Siba Smarak Panigrahi

Siba Smarak Panigrahi
Email: Unscramble

More about me: Within My Mind?! - Classic Short Stories | Fun/Favorites | Lore Podcast Summaries

I am a second-year Ph.D. candidate at École Polytechnique Fédérale de Lausanne (EPFL) working with Prof. Maria Brbic in the MLBio Lab. My research interests span LLM agents, building AI Scientists or Co-scientists, AI for scientific discovery and material generation, and equivariant neural networks.

Previously, I was a visiting researcher at ServiceNow Research in the Multimodal Foundation Models team, building foundation model for structured document understanding. I was also a Research Intern at PROSE at Microsoft, working on designing algorithms and evaluation set-ups for email classification in real-world (online) settings. I was fortunate to work on topological data analysis at Adobe Research, India and on explainability in pre-trained language models at INK-Lab, University of Southern California under Prof. Xiang Ren.

I completed my M.Sc. in Computer Science from McGill University in a thesis-based program under the supervision of Prof. Siamak Ravanbakhsh. I was also affiliated with Mila during this time. I earned my B. Tech. in Computer Science and Engineering from Indian Institute of Technology Kharagpur. I was part of the CVIR Lab under the supervision of Prof. Abir Das and Dr. Rameswar Panda where I worked on contextual bias and multimodal learning problems.

I am open to collaborations / discussions. Please feel free to schedule a meeting on Calendly.

Other: Grad Applications | Singapore 🇸🇬 | Montreal 🇨🇦 | Banff 🇨🇦

Publications

	HeurekaBench: A Benchmarking Framework for AI Co-scientist Siba Smarak Panigrahi, Jovana Videnović, Maria Brbić International Conference on Learning Representations (ICLR*) 2026 How to create benchmarks for evaluating AI co-scientists on open-ended scientific discovery tasks? website / paper / code
	BigDocs: An Open and Permissively-Licensed Dataset for Training Multimodal Models on Document and Code Tasks Juan Rodriguez, Xiangru Jian, Siba Smarak Panigrahi, et al. International Conference on Learning Representations (ICLR) 2025; RBFM workshop (NeurIPS*) 2024 Can a open and permissively-licensed dataset for continual pre-training improve document understanding capabilities of multimodal models? website / paper / dataset
	SymmCD: Symmetry-Preserving Crystal Generation with Diffusion Models Daniel Levy, Siba Smarak Panigrahi*, Sékou-Oumar Kaba, Qiang Zhu, Mikhail Galkin, Santiago Miret, Siamak Ravanbakhsh International Conference on Learning Representations (ICLR) 2025; AI4Mat workshop (NeurIPS) 2024 (spotlight) How to design symmetric crystals with diffusion models? pdf / code
	Improved Canonicalization for Model Agnostic Equivariance Siba Smarak Panigrahi, Arnab Kumar Mondal Equivariant Vision (EquiVision) workshop (CVPR) 2024 How to convert arbitrary neural network to an equivariant neural network, without equivariant canonicalization? pdf / EquiAdapt package (code)
	Efficient Dynamics Modeling in Interactive Environments with Koopman Theory Arnab Kumar Mondal, Siba Smarak Panigrahi, Sai Rajeswar, Kaleem Siddiqi, Siamak Ravanbakhsh International Conference on Learning Representations (ICLR) 2024; European Workshop on Reinforcement Learning (EWRL) 2023 How to model the dynamics of RL environments with Koopman theory and imitate linear dynamics? pdf
	Equivariant Adaptation of Large Pretrained Models Arnab Kumar Mondal, Siba Smarak Panigrahi, Sékou-Oumar Kaba, Sai Rajeswar, Siamak Ravanbakhsh Conference on Neural Information Processing Systems (NeurIPS**) 2023 How to convert arbitrary neural network to an equivariant neural network? pdf / ICML blog / Mila blog / EquiAdapt package (code)
	[Re]: Value Alignment Verification Siba Smarak Panigrahi, Sohan Patnaik The ML Reproducibility Challenge (MLRC), 2021; NeurIPS Journal Showcase Track, 2022; NeurIPS Spotlight, 2022 pdf / code
	Leveraging Pretrained Language Models for Key Point Matching Manav Nitin Kapadnis, Sohan Patnaik, Siba Smarak Panigrahi, Varun Madhavan, and Abhilash Nandy 8th workshop on ArgumentMining at Empirical Methods in Natural Language Processing (EMNLP), 2021 pdf / code
	Multi-class Emotion Classification Using EEG Signals Divya Acharya, Riddhi Jain, Siba Smarak Panigrahi, Rahul Sahni, Siddhi Jain, Sanika Prashant Deshmukh, and Arpit Bhardwaj 10th International Advance Computing Conference (IACC), 2020 pdf / code

Packages

EquiAdapt
Library to make any existing neural network architecture equivariant.

GitHub repo / PyPI package / Documentation

Template: this