Shetal Shah
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Shetal Shah has a total of 13 co-authors. Here are the top co-authors by number of shared publications:
- S. Akshay
Indian Institute of Technology Bombay
4 shared publications - Supratik Chakraborty
Indian Institute of Technology Bombay
4 shared publications - Aniruddha Joshi
University of California at Berkeley
2 shared publications - Bhupesh Chawda
IIT Bombay
2 shared publications - Bikash Chandra
IIT Bombay
2 shared publications - Biplab Kar
IIT Bombay
2 shared publications
Latest Publications
- 26 October 2025
- Published by Springer Nature
- p. 321-341
Locally Pareto-Optimal Interpretations for Black-Box Machine Learning Models
- 26 October 2025
- Published by Springer Nature
- p. 321-341
Locally Pareto-Optimal Interpretations for Black-Box Machine Learning Models
- Open Access
Locally Pareto-Optimal Interpretations for Black-Box Machine Learning Models
Creating meaningful interpretations for black-box machine learning models involves balancing two often conflicting objectives: accuracy and explainability. Exploring the trade-off between these objectives is essential for developing trustworthy interpretations. While many techniques for multi-objective interpretation synthesis have been developed, they typically lack formal guarantees on the Pareto-optimality of the results. Methods that do provide such guarantees, on the other hand, often face severe scalability limitations when exploring the Pareto-optimal space. To address this, we develop a framework based on local optimality guarantees that enables more scalable synthesis of interpretations. Specifically, we consider the problem of synthesizing a set of Pareto-optimal interpretations with local optimality guarantees, within the immediate neighborhood of each solution. Our approach begins with a multi-objective learning or search technique, such as Multi-Objective Monte Carlo Tree Search, to generate a best-effort set of Pareto-optimal candidates with respect to accuracy and explainability. We then verify local optimality for each candidate as a Boolean satisfiability problem, which we solve using a SAT solver. We demonstrate the efficacy of our approach on a set of benchmarks, comparing it against previous methods for exploring the Pareto-optimal front of interpretations. In particular, we show that our approach yields interpretations that closely match those synthesized by methods offering global guarantees.
Open AccessLocally Pareto-Optimal Interpretations for Black-Box Machine Learning Models
Creating meaningful interpretations for black-box machine learning models involves balancing two often conflicting objectives: accuracy and explainability. Exploring the trade-off between these objectives is essential for developing trustworthy interpretations. While many techniques for multi-objective interpretation synthesis have been developed, they typically lack formal guarantees on the Pareto-optimality of the results. Methods that do provide such guarantees, on the other hand, often face severe scalability limitations when exploring the Pareto-optimal space. To address this, we develop a framework based on local optimality guarantees that enables more scalable synthesis of interpretations. Specifically, we consider the problem of synthesizing a set of Pareto-optimal interpretations with local optimality guarantees, within the immediate neighborhood of each solution. Our approach begins with a multi-objective learning or search technique, such as Multi-Objective Monte Carlo Tree Search, to generate a best-effort set of Pareto-optimal candidates with respect to accuracy and explainability. We then verify local optimality for each candidate as a Boolean satisfiability problem, which we solve using a SAT solver. We demonstrate the efficacy of our approach on a set of benchmarks, comparing it against previous methods for exploring the Pareto-optimal front of interpretations. In particular, we show that our approach yields interpretations that closely match those synthesized by methods offering global guarantees.
- 5 December 2023
- Published by Springer Nature in Annals of Mathematics and Artificial Intelligence
- Vol. 92 (5) , 1051-1096
Tractable representations for Boolean functional synthesis
- 5 December 2023
- Published by Springer Nature in Annals of Mathematics and Artificial Intelligence
- Vol. 92 (5) , 1051-1096
Tractable representations for Boolean functional synthesis
- Open Access
Data Generation for Testing and Grading SQL Queries
Correctness of SQL queries is usually tested by executing the queries on one or more datasets. Erroneous queries are often the results of small changes, or mutations of the correct query. A mutation Q' of a query Q is killed by a dataset D if Q(D) $\neq$ Q'(D). Earlier work on the XData system showed how to generate datasets that kill all mutations in a class of mutations that included join type and comparison operation mutations. In this paper, we extend the XData data generation techniques to handle a wider variety of SQL queries and a much larger class of mutations. We have also built a system for grading SQL queries using the datasets generated by XData. We present a study of the effectiveness of the datasets generated by the extended XData approach, using a variety of queries including queries submitted by students as part of a database course. We show that the XData datasets outperform predefined datasets as well as manual grading done earlier by teaching assistants, while also avoiding the drudgery of manual correction. Thus, we believe that our techniques will be of great value to database course instructors and TAs, particularly to those of MOOCs. It will also be valuable to database application developers and testers for testing SQL queries.Open AccessData Generation for Testing and Grading SQL Queries
Correctness of SQL queries is usually tested by executing the queries on one or more datasets. Erroneous queries are often the results of small changes, or mutations of the correct query. A mutation Q' of a query Q is killed by a dataset D if Q(D) $\neq$ Q'(D). Earlier work on the XData system showed how to generate datasets that kill all mutations in a class of mutations that included join type and comparison operation mutations. In this paper, we extend the XData data generation techniques to handle a wider variety of SQL queries and a much larger class of mutations. We have also built a system for grading SQL queries using the datasets generated by XData. We present a study of the effectiveness of the datasets generated by the extended XData approach, using a variety of queries including queries submitted by students as part of a database course. We show that the XData datasets outperform predefined datasets as well as manual grading done earlier by teaching assistants, while also avoiding the drudgery of manual correction. Thus, we believe that our techniques will be of great value to database course instructors and TAs, particularly to those of MOOCs. It will also be valuable to database application developers and testers for testing SQL queries. - 2 August 2015
- Published by Springer Nature in The VLDB Journal
- Vol. 24 (6) , 731-755
Data generation for testing and grading SQL queries
- 2 August 2015
- Published by Springer Nature in The VLDB Journal
- Vol. 24 (6) , 731-755
Data generation for testing and grading SQL queries
- 1 April 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1043-1052
Handling Non-linear Polynomial Queries over Dynamic Data
Applications that monitor functions over rapidly and unpredictably changing data, express their needs as continuous queries. Our focus is on a rich class of queries, expressed as polynomials over multiple data items. Given a set of polynomial queries at a coordinator C, and a user-specified accuracy bound (tolerable imprecision) for each query, we address the problem of assigning data accuracy bounds or filters to the source of each data item. Assigning data accuracy bounds for non-linear queries poses special challenges. Unlike linear queries, data accuracy bounds for non-linear queries depend on the current values of data items and hence need to be recomputed frequently. So, we seek an assignment such that a) if the value of each data item at C is within its data accuracy bound then the value of each query is also within its accuracy bound, b) the number of data refreshes sent by sources to C to meet the query accuracy bounds, is as low as possible, and c) the number of times the data accuracy bounds need to be recomputed is as low as possible. In this paper, we couple novel ideas with existing optimization techniques to derive such an assignment. Specifically, we make the following contributions: (i) Propose a novel technique that significantly reduces the number of times data accuracy bounds must be recomputed; (ii) Show that a small increase in the number of data refreshes can lead to a large reduction in the number of recomputations; we introduce this as a tradeoff in our approach; (iii) Give principled heuristics for addressing negative coefficient polynomial queries where no known optimization techniques can be used; we also prove that under many practically encountered conditions our heuristics can be close to optimal; and (iv) Present a detailed experimental evaluation demonstrating the efficacy of our techniques in handling large number of polynomial queries.
- 1 April 2008
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1043-1052
Handling Non-linear Polynomial Queries over Dynamic Data
Applications that monitor functions over rapidly and unpredictably changing data, express their needs as continuous queries. Our focus is on a rich class of queries, expressed as polynomials over multiple data items. Given a set of polynomial queries at a coordinator C, and a user-specified accuracy bound (tolerable imprecision) for each query, we address the problem of assigning data accuracy bounds or filters to the source of each data item. Assigning data accuracy bounds for non-linear queries poses special challenges. Unlike linear queries, data accuracy bounds for non-linear queries depend on the current values of data items and hence need to be recomputed frequently. So, we seek an assignment such that a) if the value of each data item at C is within its data accuracy bound then the value of each query is also within its accuracy bound, b) the number of data refreshes sent by sources to C to meet the query accuracy bounds, is as low as possible, and c) the number of times the data accuracy bounds need to be recomputed is as low as possible. In this paper, we couple novel ideas with existing optimization techniques to derive such an assignment. Specifically, we make the following contributions: (i) Propose a novel technique that significantly reduces the number of times data accuracy bounds must be recomputed; (ii) Show that a small increase in the number of data refreshes can lead to a large reduction in the number of recomputations; we introduce this as a tradeoff in our approach; (iii) Give principled heuristics for addressing negative coefficient polynomial queries where no known optimization techniques can be used; we also prove that under many practically encountered conditions our heuristics can be close to optimal; and (iv) Present a detailed experimental evaluation demonstrating the efficacy of our techniques in handling large number of polynomial queries.