Vedant Nanda
I am an AI Scientist at Mistral. My research focuses on efficient language models and how we can pretrain them to be more inference-friendly.
Previously, I worked as a researcher at Aleph Alpha in Germany where I worked on inference optimizations for LLMs (see this talk) and pretraining optimizations of our tokenizer-free models (HuggingFace collection).
I got my PhD in Computer Science from University of Maryland, College Park, as part of the Maryland-Max Planck joint program, where I was advised by Krishna P. Gummadi and John P. Dickerson, and collaborated with Adrian Weller at Cambridge. During my PhD, I studied trustworthy deep learning, working on fairness (ICML2019 AAAI2020 AAAI2023), robustness (ICML2022 & AAAI2023), and connections between them (FAccT2021). Driven by important role of scale in deep learning, I also explored efficient representation learning (NeurIPS2023) and taught a seminar on Systems for LLMs. I also interned twice at AWS, working on counterfactual explanations with AWS Clarify and fairness in generative AI with AWS Bedrock.
I am fortunate to have worked with some amazing mentors (in alphabetical order): Soheil Feizi (UMD), Tom Goldstein (UMD), Hoda Heidari (ETH Zürich), Ponnurangam Kumaraguru (IIIT Delhi), Bradley C. Love (UCL), Rijurekha Sen (IIT Delhi), Pushpendra Singh (IIIT Delhi), Mariya Toneva (MPI-SWS), and Muhammad Bilal Zafar (Amazon).
For more, check out my CV.
Email: vnanda [at] mpi-sws [dot] org
Publications
Conference Papers
- The Impact of Inference Acceleration Strategies on Bias of Large Language Models
- Elisabeth Kirsten, Ivan Habernal, Vedant Nanda, Muhammad Bilal Zafar
- Paper | Code
- Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics (NAACL) 2025
- Lawma: The Power of Specialization for Legal Tasks
- Ricardo Dominguez-Olmedo, Vedant Nanda, Rediet Abebe, Stefan Bechtold, Christoph Engel, Jens Frankenreiter, Krishna Gummadi, Moritz Hardt, Michael Livermore
- Paper | Code
- International Conference on Learning Representations (ICLR) 2025
- Towards Reliable Latent Knowledge Estimation in LLMs: Zero-Prompt Many-Shot Based Factual Knowledge Extraction
- Qinyuan Wu, Mohammad Aflah Khan, Soumi Das, Vedant Nanda, Bishwamittra Ghosh, Camila Kolling, Till Speicher, Laurent Bindschaedler, Krishna P. Gummadi, Evimaria Terzi
- Paper | Code | Dataset
- Conference on Web Search and Data Mining (WSDM) 2025
- Investigating the Effects of Fairness Interventions Using Pointwise Representational Similarity
- Camila Kolling, Till Speicher, Vedant Nanda, Mariya Toneva, Krishna P. Gummadi
- Paper
- Transactions on Machine Learning Research (TMLR) 2025
- Understanding the Role of Invariance in Transfer Learning
- Till Speicher, Vedant Nanda, Krishna P. Gummadi
- Paper | Code
- Transactions on Machine Learning Research (TMLR) 2024
- Foundations of Trustworthy Deep Learning: Fairness, Robustness, and Explainability
- Vedant Nanda
- Slides
- PhD Thesis, University of Maryland 2024
- Diffused Redundancy in Pre-trained Representations
- Vedant Nanda, Till Speicher, John P. Dickerson, Soheil Feizi, Krishna P. Gummadi, Adrian Weller
- ArXiv | Code | Slides | Poster
- Conference on Neural Information Processing Systems (NeurIPS) 2023
- Acceptance Rate: 26.1%
- What Happens During Finetuning of Vision Transformers: An Invariance Based Investigation
- ArXiv
- Gabriele Merlin, Vedant Nanda, Ruchit Rawal, Mariya Toneva
- Conference on Lifelong Learning Agents (CoLLAs) 2023
- Do Invariances in Deep Neural Networks Align with Human Perception?
- ArXiv | Code | Slides | Poster
- Vedant Nanda, Ayan Majumdar, Camila Kolling, John P. Dickerson, Krishna P. Gummadi, Bradley C. Love, Adrian Weller
- AAAI Conference on Artifical Intelligence, (AAAI) 2023, selected for oral presentation
- Acceptance Rate: 19.6%
- Rawlsian Fairness in Online Bipartite Matching: Two-sided, Group, and Individual
- ArXiv | Code coming soon!
- Seyed A. Esmaeili, Sharmila Duppala, Davidson Cheng, Vedant Nanda, Aravind Srinivasan, John P. Dickerson
- AAAI Conference on Artifical Intelligence, (AAAI) 2023
- Acceptance Rate: 19.6%
- An earlier version appeared as an extended abstract in AAMAS 2022
- Measuring Representational Robustness of Neural Networks Through Shared Invariances
- ArXiv | Code | Slides | Poster | Talk
- Vedant Nanda, Till Speicher, Camila Kolling, John P. Dickerson, Krishna P. Gummadi, Adrian Weller
- International Conference on Machine Learning (ICML) 2022, selected for oral presentation Acceptance Rate: 2% (oral), 19.84% (overall)
- Fairness Through Robustness: Investigating Robustness Disparity in Deep Learning
- ArXiv | Paper | Code | Slides | Live QnA @ FAccT
- Vedant Nanda*, Samuel Dooley*, Sahil Singla, Soheil Feizi, John P. Dickerson
- ACM Conference on Fairness, Accountability, and Transparency (FAccT, formerly FAT*) 2021 Acceptance Rate: 28%
- * joint first author
- Balancing the Tradeoff between Profit and Fairness in Rideshare Platforms during High-Demand Hours
- ArXiv | Paper | Code | Slides | Poster
- Vedant Nanda, Pan Xu, Karthik A. Sankararaman, John P. Dickerson, Aravind Srinivasan
- AAAI Conference on Artifical Intelligence (AAAI) 2020 Acceptance Rate: 20.6%
- Also accepted as oral (acceptance rate: 14.7%) at AIES 2020
- On the Long-term Impact of Algorithmic Decision Policies: Effort Unfairness and Feature Segregation through Social Learning
- Paper | Supplementary | Code | Slides
- Hoda Heidari*, Vedant Nanda*, Krishna P. Gummadi
- International Conference on Machine Learning (ICML) 2019 Acceptance Rate: 22.6%
- * joint first author
- Leveraging Facebook's Free Basics Engine for Web Service Deployment in Developing Regions
- S. Singh *, Vedant Nanda *, R. Sen, S. Sengupta, P. Kumaraguru, K.P. Gummadi
- International Conference on Information and Communication Technologies and Development (ICTD) 2017 Acceptance Rate: 36%
- * joint first author
Pre-prints
- Technical Challenges for Training Fair Neural Networks
- Valeriia Cherepanova*, Vedant Nanda*, Micah Goldblum, John P. Dickerson, Tom Goldstein
- * joint first author
- Enhancing Model Robustness via Machine-Checkable Concepts
- Vedant Nanda, Till Speicher, John P. Dickerson, Krishna P. Gummadi, Muhammad Bilal Zafar
Workshop Papers/Posters
- The Impact of Inference Acceleration Strategies on Bias of Large Language Models
- Paper
- Elisabeth Kirsten, Ivan Habernal, Vedant Nanda, Muhammad Bilal Zafar
- Safe Generative AI Workshop @ NeurIPS 2024
- Learning to Explain Machine Learning
- Paper | Talk
- Vedant Nanda*, Duncan McElfresh*, John P. Dickerson
- CHI Workshop on Human-Centered Perspectives in Explainable AI (HCXAI) 2021
- * joint first author
- Technical Challenges for Training Fair Neural Networks
- Valeriia Cherepanova*, Vedant Nanda*, Micah Goldblum, John P. Dickerson, Tom Goldstein
- ICLR Workshop on Responsible AI (RAI) 2021
- * joint first author
- Unifying Model Explainability and Robustness via Reasoning Labels
- Code coming soon!
- Vedant Nanda, Junaid Ali, Krishna P. Gummadi, Muhammad Bilal Zafar
- Safety and Robustness in Decision Making (SRDM) workshop at NeurIPS 2019
- Stop the KillFies! Using Deep Learning Models to Identify Dangerous Selfies
- Vedant Nanda, H. Lamba, D. Agarwal, M. Arora, N. Sachdeva, P. Kumaraguru
- MSM workshop, WWW'18. Companion of the The Web Conference 2018
- An Empirical Analysis of Facebook's Free Basics
- S. Singh *, Vedant Nanda *, R. Sen, S. Ahmad, S. Sengupta, A. Phokeer, Z.A. Farooq, T.A. Khan, P. Kumaragaguru, I.A. Qazi, D. Choffnes, K.P. Gummadi
- ACM SIGMETRICS Performance Evaluation Review 2017 (Poster)
- * joint first author