Akash Srivastava
I am a PI at the MIT-IBM AI Research Lab and the Chief Scientist at EBI within the Exploratory Sciences unit of IBM Research. I also serve as the technical lead for the synthetic data generation efforts at IBM. My lab is located on campus, at 314 Main St (new MIT Museum building) in Cambridge, where my group works on topics ranging from self-supervised learning, deep generative models, differential and statistical privacy, density ratio estimation, machine common-sense, model calibration and uncertainity quantification and foundational models. Before joining MIT-IBM, I obtained my PhD at the University of Edinburgh where I worked with Prof Charles Sutton and Prof Michael U. Gutmann on variational inference for generative models and deep learning.
Current Projects
- PI, MIT-IBM, 2023: Generative Modeling for Complex Mechanical Systems with Constraints. In collaboration with Prof. Faez Ahmed (MIT)
- PI, MIT-IBM, 2023: Synthetic data and randomness in business and societal decision-making. In collaboration with Prof. Dean Eckles (MIT)
- PI, MIT-IBM, 2023: Generative active learning of atomistic simulators for silica materials. In collaboration with Prof. Rafael Gomez-Bombarelli (MIT)
- PI, MIT-IBM, 2023: Rethinking the vehicle routing problem under the lens of modern machine learning techniques. In collaboration with Dr. Matthias Winkenbach (MIT)
- co-PI, MIT-IBM, 2023: Teaching Foundation Models 3D. Led by Prof. Vincent Sitzmann (MIT) and Dr. Leonid Karlinsky (MIT-IBM Research)
- PI, Climate Change AI, 2022: Towards greener last-mile operations: Supporting cargo-bike logistics through optimized routing of multi-modal urban delivery fleets.
- PI, MIT-IBM, 2021: Hybrid Generative Models. In collaboration with Prof. Faez Ahmed (MIT)
- Co-PI, MIT-IBM, 2021: Representation Learning as a Tool for Causal Discovery. Led by Prof. Caroline Uhler (MIT) and Dr. Kristjan H Greenewald (MIT-IBM Research)
- PI, MIT-IBM, 2020: Learning Priors for Transfer. In collaboration with Prof. Pulkit Agarwal (MIT)
- Co-PI, DARPA, 2019: Machine Common Sense. Led by Prof. Josh Tenenbaum (MIT) and Dr. Dan Gutfreund (MIT-IBM Research)
News:
- Read about how MIT-IBM lab and IBM Research are using synthetic data generation method to tackle real world problems in this blog post. It also features work from our work on generative models for engineering design problems.
- Our project on generative models for inverse linkage synthesis was recently featured in MIT’s spectrum magazine.
- Read about our on-going work on making the last mile deliveries greeener, in this blog post
- I am always looking for talented students to join my group as interns and/or collaborators. If you are a student at MIT, looking for an internship/work-experience/collaboration and are interested in any of the following topics, please get in touch as we have a rolling intake all year long: 1. Information obfuscation and synthetic data generation. 2. Time-series differential privacy. 3. User-level differential privacy. 4. Deep generative modeling for domains that require high precision and constraint satisfaction 5. Density ratio estimation in high-dimensional data 6. Understanding large language models using probabilistic graphical modeling 7. Uncertainty quantification and model calibration in self-supervised representation learning.