aalok thakkar: research

formal methods and logic: My doctoral research focussed on automated program synthesis. A good starting point to learn about this work is this survey. In program synthesis, I am looking at the following two questions:

1. Given a set of specifications which cannot be satisfied by any program, how can we find a program that satisfies a maximal subset of the specifications? Shobhit Singh and Rajdeep Das work on this project.
2. Given a database transformation (which includes a schema transformation and a value transformation), how can we systematically generate a query on the transformed database that is semantically equivalent to a specified query on the original database? This project is in collaboration with TCS Research, and Shobhit Singh and Shrey Arora work on it.

Beyond program synthesis, I am interested in automata theory. Three questions I am working on are:

1. When does a transformation of a formal language preserve its regularity?
2. When is it possible to represent probability distributions over strings using finite-state models? Smayan Agarwal works on this project.
3. How can we test if two probability distributions over strings are close to each other? Smayan Agarwal and Shobhit Singh work on this project.

I am also interested in logic programming and probablistic logic programming.

trustworthy AI: I am exploring several key aspects of trustworthy AI:

1. lipschitz continuity: Investigating ways to make AI models more robust by ensuring controlled sensitivity to input changes. This raises many interesting theoretical and practical questions. Prabhpreet Setia and Pranav Jayanandan are working on this.
2. testing AI: In collaboration with Moolya, we are developing methods to rigorously evaluate AI systems, identify failure modes, and ensure reliable behavior under diverse scenarios.
3. AI safety benchmarking for India: This research is funded by EkStep Foundation's AI and Data Adoption fellowship. This is in collaboration with Debayan Gupta and Anirban Sen. Several students and research assistants are working on this project.

CS pedagogy: I am interested in the impact of AI on computer science education and in developing curricula and tools that are relevant to today's AI-enhanced learning environment. This includes:

1. Designing courses that emphasize conceptual understanding, formal methods, and systems thinking to go beyond what AI tools can solve automatically. As a part of this project, Vaani Goenka has built the BOOP framework.
2. Developing intelligent assistants and teaching tools that guide learning while preserving active engagement and critical thinking.
3. Creating educational resources and platforms that are accessible to all students, including those with disabilities, with a focus on the Indian educational context. As a part of this, Vaanee Tripathi has worked on redesigning CS curricula.

The above projects are supported by Mphasis Limited as a part of the Mphasis AI & Applied Tech Lab at Ashoka.

computational sustainability: I am working on two projects that bring together computation and the environment:

1. darjeeling revival project: We are developing statistical models to estimate the carbon captured by Enhanced Rock Weathering (ERW) in Northern Bengal. The project is funded by the Alt Carbon, and a number of students and research assistants are working on it. The project includes field work and data collection, software development, as well as statistical modeling and analysis.
2. OpenGIS India: This is a project in collaboration with Meghna Agarwala , where we use GIS and remote sensing to study forest dynamics, changing land usage and forest fire patterns, floods and river dynamics, and animal-human conflicts across India.