Industry

2025 – Present

Machine Learning Engineer

Nirva Labs · Contract via Gradient Works LLC

Building AI-powered audio understanding systems: end-to-end ML pipelines for speaker diarization, voiceprint matching, and LLM-driven content extraction. Full-stack contributions spanning Python backend (FastAPI), iOS app (SwiftUI), and cloud infrastructure (AWS).

Publications

Two-Level Actor-Critic Using Multiple Teachers
Su Zhang, Srijita Das, Sriram Ganapathi Subramanian, Matthew E. Taylor
Transactions on Machine Learning Research (TMLR) & AAMAS 2023
Enhanced Learning from Multiple Demonstrations with a Two-level Structured Approach
Su Zhang, Matthew E. Taylor
ALA Workshop at AAMAS 2018
Maintenance for Case Streams: A Streaming Approach to Competence-Based Deletion
Yang Zhang, Su Zhang, David Leake
ICCBR 2017
Case-Base Maintenance: A Streaming Approach
Yang Zhang, Su Zhang, David Leake
ICCBR Workshops 2016

Selected Research

Multiagent Coordination via Option Value Decomposition

2023 – 2024

Multi-agent option-critic framework (CTDE); addressed sparse reward with optimistic heuristic integration.

Two-Level Actor-Critic Using Multiple Teachers

2021 – 2023

Hierarchical RL framework for learning from heterogeneous-quality demonstrations via teacher selection and low-level policy optimization. Published in TMLR and AAMAS.

Efficient Exploration with Probability Map

2019 – 2021

Leveraged prior probability distributions as exploration heuristics to accelerate RL agent training in robotic search tasks.