Projects / Work Experience
Projects
Description: Worked on a real-world dataset designed to model the intervention effects of health workers across India, developing efficient algorithms for sequential decision-making under scarce resources. Contributed to the modeling of a general Restless Bandit framework suitable for the intervention dataset at Google Research, India, and characterized the assumptions necessary for the asymptotic optimality of various computationally efficient policies.
It has been accepted for publication as a full paper at International Conference on Autonomous Agents and Multiagent Systems (AAMAS) 2025.
Description: Theoretical insights for a few widely known phenomena of Knowledge Distillation including Model Compression (using traditional labels vs. soft labels for training in Supervised Classification task) using the tools of Neural Tangent Kernel.
It has been accepted for publication at Learning for Dynamics and Control (L4DC) 2025.
Presented the work at CSL Student Conference 2025.
Description: Extended the Dawid-Skene (DS) model to multi-type cases and analyzed the error bounds of existing algorithms. Focusing on the case where there are two types of tasks, we propose a spectral method to partition tasks into two groups that cluster tasks by type. Our analysis reveals that task types can be perfectly recovered if the number of workers n scales logarithmically with the number of tasks d.
Accepted for a talk at INFORMS Applied Probability Society Conference 2025 to be held in July.
Under review at TMLR.
Description: Proposed a polynomial-time, delay-reliable scheduling strategy for multi-resource allocation in wired media, optimal in the presence of adversarial packet arrival. Developed an approximate solution for handling multiple packet types under the same framework.
Under preparation for GLOBECOM 2025.
Description: Calibrated the parameters of two widely-used Effective SNR mappings (link quality metrics) - EESM (Exponential Effective SNR Mapping) and TIESNR (Time-Invariant Effective SNR) for 5G radio through link-level simulations in MATLAB's 5G Toolbox.
Work Experience
Description: Automated the fault aggregation component of error collation from the IP metadata of the upcoming Orin SoC architecture. This involved accessing the Google Drive API using a Perl script and efficiently parsing IP metadata. Developed a comprehensive latency calculator using the maps between IPs within the same SoC architecture. Achieved exponential reductions in latency calculation time by applying shortest path algorithms and recursion in Python.