Abhineet Agarwal
I am a fourth-year statistics PhD student at UC Berkeley, where I am fortunate to be advised by Prof. Bin Yu. My main research interests are in:
Interpretable Machine Learning. My research has focused on tree-based methods, where I have worked on both developing inherently interpretable algorithms (e.g., shallow decision trees) and post-hoc explanation methods for tree ensembles such as Random Forests and XGBoost. Recently, I have been thinking about how to apply these tools to interpret deep learning and large language models.
Causal Machine Learning. I have worked on developing data-driven causal frameworks for estimating personalized outcomes under combinations of treatments (e.g., factorial design experiments) or rankings of items.
Papers
Liwen Sun, A. Agarwal, A. Kornblith, B. Yu, C.Xiong. "ED-Copilot: Reduce Emergency Department Wait Time with Language Model Diagnostic Assistance."
Q.Wang*, T.Tang*, ..., A.Agarwal, ... E.Ashley. "Epistasis Regulates Genetic Control of Cardiac Hypertrophy."
Journal: In submission to Nature Cardiovascular Research, 2023
A. Agarwal, A. Agarwal, S. Vijaykumar. "Synthetic Combinations: A Causal Inference Framework For Combinatorial Interventions."
Conference: Neural Information Processing Systems (NeurIPS), 2023.
A. Agarwal*, A. Kenney*, T. Tang*, YS. Tan*, B. Yu. "MDI+: A Flexible Random Forest-Based Feature Importance Framework."
Journal: In submission to Journal of the American Statistical Association (JASA).
YS. Tan*, C. Singh*, K. Nasseri*, A. Agarwal*, J. Duncan, O. Ronen, M. Epland, A. Kornblith, B. Yu. "Fast Interpretable Greedy Tree Sums"
Journal: In submission to Proceedings of the National Academy of Sciences (PNAS)
A. Agarwal*, YS. Tan*, O. Ronen, C. Singh, B. Yu. "Hierarchical Shrinkage: Improving the Interpretability and Accuracy of Tree-Based Methods"
Conference: International Conference on Machine Learning (ICML), 2022. Oral Presentation
YS. Tan, A. Agarwal, B. Yu. "A Cautionary Tale on Fitting Decision Trees to Data From Additive Models: Generalization Lower Bounds"
Conference: International Conference on Artificial Intelligence and Statistics (AISTATS), 2022.
J. Duncan*, R. Kapoor*, A. Agarwal*, C. Singh*, B. Yu. "VFlow: Python package for building trustworthy data science pipelines with PCS".
Journal: Journal of Open Source Software (JOSS), 2022.
*Denotes equal contribution.
News
Oct '23: I am presenting Synthetic Combinations at the 2023 INFORMS Workshop on Adaptive Learning for Platform Operations
Sep '23: Synthetic Combinations was selected for a long presentation at the MIT Conference on Digital Experimentation
Aug '23: Synthetic Combinations was selected for a poster presentation on at the Cornell ORIE Young Researchers Workshop
Aug '23: I gave a talk at the Joint Statistical Meetings in the session on "Recent advances on decision tree and random forest theory"
Nov '22: I gave a talk on our paper MDI+ as the student speaker at the Berkeley-Stanford Statistics Colloquium
Oct '22: I gave a talk on our papers Fast Interpretable Greedy Tree Sums and MDI+ at Blue Cross Blue Shield
Sep '22: I gave a talk on our paper Fast Interpretable Greedy Tree Sums at the Simons Institute Summer Cluster on Interpretable ML
Jul '22: Our paper Hierarchical Shrinkage was accepted for an oral presentation at ICML'22