Maxwell Adam
I'm an undergraduate at The University of Melbourne studying neuroscience and math
Melbourne, Australia
max@v3rv.com
github.com/bfpill
Click here for a PDF Version
Experience
Timaeus - Research Scientist
July 2025 - present
Timaeus - Research Intern
Jun 2025 - July 2025
Zendesk, Melbourne Aus — Engineering Intern
September 2022 - March 2023
Research (papers)
The Loss Kernel: A Geometric Probe for Deep Learning Interpretability
Maxwell Adam, Zach Furman, Jesse Hoogland
We introduce the loss kernel, an interpretability method for measuring similarity between data points according to a trained neural network.
Bayesian Influence Functions for Hessian-Free Data Attribution
Philipp Alexander Kreer, Wilson Wu, Maxwell Adam, Zach Furman, Jesse Hoogland
We propose the local Bayesian influence function (BIF), an extension of classical influence functions that replaces Hessian inversion with loss landscape statistics that can be estimated via stochastic-gradient MCMC sampling.
Research (blogposts)
Interpreting Complexity - Singular Learning Theory for Interpretability - March 2025
Independent research into using Singular Learning Theory (SLT) ideas for interpretability. Introduce novel interpretability and SLT tools. Published findings on LessWrong. Read the post, or the twitter thread
A Contrastive Analysis of Features in Twin Toy Transformers that (play) Chess - November 2024
Trained SAE's on models trained to predict legal chess moves. Found nice features! Features interactive neuron browsers for exploring the models and interactive graphs, all written in typescript. Read the post
Education
The University of Melbourne - mix of neuro, math, computer science - February 2024 - Present
RMIT University, Melbourne Aus — Bachelor of CS February 2023 - November 2023
Acton Boxborough High School, Boston Ma — HS Diploma 2020 - 2022
Graduated High School in America. Took Honors Physics, Honors Calculus, College Level Java and Unix/Linux Shell Scripting.