Jin Schofield

🔬 Created novel uniformity-only contrastive representation learner that rivals variational auto-encoder accuracy with 11% fewer parameters. Co-authored BuilderBench paper (under review for ICLR 2026). Researching exploration for contrastive reinforcement learning using temporal representations. 🛠️ Tech: JAX, NumPy

🚁 Re-architected and trained large core ML object detection model, allowing for detection of drones in additional ~50% of camera FOV that was previously too noisy, for autonomous precision robotics system for drone interception using recently published research papers while decreasing latency by ~75% compared to research paper implementation. Constructed a pipeline for fine-grained control of hyper-realistic synthetic training data generation using recently published research papers, increasing accuracy by additional 5%. 🛠️ Tech: CUDA, PyTorch, AWS, TensorRT

👨‍🏫 Taught and aided students with data structures, algorithms, and programming systems material and assignments.

🤖 Developed novel conditional diffusion methods with Vikram Ramaswamy for cellular automata synthesis in Conway's Game of Life; researched diffusion for text language data with Diffusion-of-Thoughts.

📈 Built RNN, XGBoost, and Random Forest models to drive a trading strategy that achieved 305% cumulative returns over 12 months; constructed datasets using SQL and Python under mentorship from a BlackRock quantitative researcher and previous Meta data scientist.

🚀 Implemented novel in-context learning techniques within latent diffusion models, improving predictive accuracy by 26%.

💡 Designed features for an AI-driven educational tool and authored a weekly newsletter reaching 45,000+ subscribers.

📊 Led due diligence and investment decisions for early-stage Princeton-affiliated startups. Evaluated pitch decks, financial models, and market opportunities. Contributed to the growth of student entrepreneurship at Princeton through mentorship and funding decisions.

🔍 Analyzed large-scale patient datasets in R and built outcome prediction models using machine learning techniques.

🔧 Programmed back-end C++ code of SeekThermal infrared cameras for immunoassay bodily fluid THC detection device. Programmed GUI in Python.