Open to AI / ML & Quant internships

Matthew Helton
mathematics, in motion.

PhD candidate in Mathematics at NYU Courant, building machine-learning and dynamical-systems models for high-dimensional data — from cortical circuits to deep networks trained from scratch.

Scroll
§ 01Profile

I work where applied mathematics meets machine learning — designing models with a real theoretical spine, then proving they hold up against data.

My research applies deep learning and optimization to high-dimensional, neural data: crafting loss functions and minimization schemes that recover the structure underneath. The same toolkit — rigorous modeling, gradient-based optimization, and disciplined experimentation — is what I'm looking to bring to an AI/ML or quantitative research internship.

Research focus

  • Dynamical systems — limit cycles, bifurcations, averaging & Lyapunov analysis
  • Computational neuroscience — LIF & Hodgkin–Huxley network models
  • Deep learning — self-supervised & sequence models on real data
  • Optimization — custom losses, minimization methods, backprop from scratch
§ 02Research & Experience
Graduate Researcher 2025 — Present
Young Research Group · NYU Courant Institute
  • Built a 136,000-neuron, 36-million-synapse spiking network model of macaque V1 (excitatory, PV and SOM cells) to test how SOM interneurons shape surround suppression. Manuscript in preparation.
  • Directed AI coding agents to optimize and port the simulator to the GPU: about 68,000× faster per simulated second than the original code, with every change gated by bitwise or statistical checks against a validated reference.
  • Calibrated about 45 parameters with staged Bayesian optimization (a GPU implementation of TPE) on a validated reduced surrogate.
Research Scientist 2023 — 2024
Shea-Brown Lab · University of Washington
  • Engineered and benchmarked time-series models — including LSTMs — against baseline statistical methods for forecasting activity from the mouse brain.
  • Developed a time-delay autoencoder with linear (Koopman) latent dynamics for neural forecasting, tested on Lorenz-96 and widefield imaging (code).
  • Analyzed widefield and Neuropixels recordings with dynamic mode decomposition and its variants (notebooks).
§ 03Selected Projects
2025 · Self-Supervised Learning
DINO — Implemented & Trained from Scratch
  • In a three-person team, assembled the data for, implemented, and trained from scratch a DINO model with a ViT-S/8 backbone and MLP head on about 1.95 million images.
  • Placed 6th of 39 teams on held-out CUB-200 classification in a class-wide competition (code).
Matrix Calculus · NumPy
MLP From Scratch
  • Derived, via matrix calculus, all gradients needed for backpropagation through a two-layer MLP with MSE loss and ReLU activations.
  • Implemented the computation in NumPy and verified that every result matched PyTorch exactly.
2026 · Tooling · Local LLMs
plainclaude
  • Windows reading aid that reads Claude Desktop's accessibility tree and rewrites each finished reply in plain language with a local LLM (Ollama), rendered with markdown and LaTeX.
  • Open source under the MIT license (code).
§ 04Toolkit

Languages

PythonSQL

Libraries

PyTorchNumPySciPyNumbaPandasMatplotlibScikit-learn

Methods

Deep LearningSupervised & UnsupervisedRegression AnalysisPredictive ModelingBayesian OptimizationGPU SimulationStatistical Testing

Tools

GitGitHubColab
§ 05Education
Ph.D. in Mathematics
New York University · Courant Institute
Expected May 2029GPA 3.91 / 4.0
B.S. Applied Mathematics & Neuroscience
University of Washington · Mathematics Minor
June 2023GPA 3.97magna cum laude
§ 06Awards
2024
NSF Graduate Research Fellowship
National Science Foundation
2023
Boeing Excellence Award for Outstanding Research
UW Applied Mathematics
2023
Phi Beta Kappa
University of Washington

Let's build something provably good.

Currently seeking a Summer 2027 internship in AI/ML or quantitative research. If your team values mathematical rigor and models that actually hold up, I'd love to talk.