Adrian Molofsky
Adrian Molofsky

Adrian Molofsky

Artificial Intelligence, Systems

Biography

Hi! I am an M.S. and B.S. student in Computer Science at Stanford University advised by Dr. Mark Horowitz.

I recently spent a year as a research assistant at Stanford studying imitation learning for coordination modeling of immune cell behavior supervised by Dr. Barbara Engelhardt. Previously, I was a research intern twice at UC San Francisco reconstructing gene expression in the developing human embryo supervised by Dr. Tomasz Nowakowski and earlier examining stromal-immune cell interactions in the context of metabolic disorders supervised by Dr. Julia Sbierski-Kind.

I was born in Arcata and grew up in McKinleyville near the redwoods of Northern California. In my free time, I enjoy spending time outdoors backpacking, birdwatching, and swimming, and playing strategy games such as chess, bridge, and poker.

My goal is to make machine learning systems fair, robust, and interpretable. I am interested in how models learn from data, reason across multiple input modalities, and develop resilience to outliers. I am currently thinking about data-centric approaches to improve robustness, adaptive test-time reasoning for reliable inference, and methods for generalizing models to out-of-domain tasks.

Experience

Research Assistant
Stanford University, Supervisor: Dr. Barbara Engelhardt
2024 - 2025
Stanford, CA
  • Developed an end-to-end pipeline for training, evaluating, testing, and deploying deep learning models.
  • Improved data curation and inference across 165K+ patch embeddings representing immune cell behavior.
  • Distributed training across multiple GPUs, managing job scheduling, resource allocation, and data orchestration.
Data Science Intern
Gladstone Institutes, Supervisor: Dr. Barbara Engelhardt
Summer 2024
San Francisco, CA
  • Developed image reconstruction, analysis, and visualization pipelines for live-cell imaging datasets.
  • Implemented data preprocessing, feature extraction, and image registration across 4K+ image frames.
  • Integrated segmentation and tracking for predictive modeling of cell migration, proliferation, and morphology.
Research Intern
University of California, San Francisco, Supervisor: Dr. Tomasz Nowakowski
Summer 2023
San Francisco, CA
  • Developed multimodal pipelines integrating histology, spatial transcriptomics, and single cell RNA sequencing.
  • Trained, improved, and benchmarked optimization, Bayesian, and contrastive learning models across 38K+ cells.
  • Distributed training across multiple GPUs, configuring jobs, logging metrics, and tuning hyperparameters.
Research Intern
University of California, San Francisco, Supervisor: Dr. Julia Sbierski-Kind
Summer 2019
San Francisco, CA
  • Analyzed in vivo stromal–immune cell interactions in mouse models of obesity and liver fibrosis.
  • Applied confocal microscopy, flow cytometry, and histological techniques for data analysis.

Projects

Comparing Reinforcement Learning Methods for Sparse vs. Dense Rewards

Benchmarked PPO, DDPG, SAC, and TD3 policies on a 1,000 sample states from the Point Maze environment.

View Code →

Link Prediction on MIND Dataset with PyG

Built a graph neural network recommender system on the Microsoft News Dataset (MIND) to learn user-article click behavior.

View Report →

Price-Pure Prediction of Daily Price Changes in Binary Event Contracts

Forecasted daily price changes in binary event contracts backtesting on 10K+ time-series samples from Kalshi.

View Code →

Convolutional Neural Network Accelerator

Developed a ResNet-18 hardware accelerator with a systolic array, FIFO buffering, and banked memory hierarchy.

Micropolygon Rasterization Accelerator

Designed a rasterization hardware accelerator with micropolygon bounding, edge traversal, and backface culling.

Five-Stage Pipelined MIPS Processor

Developed a five-stage pipelined processor for the MIPS ISA with hazard detection, forwarding, and stall control.

Register Renaming in a RISC-V Processor

Implemented register renaming logic in a pipelined RISC-V processor to eliminate write after write and write after read hazards.

Performance Tradeoffs of Error-Correcting Codes within Network Routers

Benchmarked parity, checksum, and Hamming error-correcting codes in 8×8 2D Torus routers using BookSim.

Formalizing Intel’s Remote Action Request

Defined Intel’s Remote Action Request for remote TLB shootdowns through memory transiency models.

Mentorship

If you are a high school or college student from a low-income or underrepresented background seeking advice on college applications, career planning, or research opportunities, please shoot me an email.

I am committed to increasing diversity and inclusion across computer science and higher education.