About Me

I am a PhD student at Cambridge University supervised by Hong Ge in the Machine Learning Group, supported by the Harding Distinguished Postgraduate Scholars Programme. I am interested in neural network theory, specifically in understanding the inductive biases and using them to explain generalisation and other empirical phenomena that we observe. More recently I have been working on in-context learning and its relationship to implicit parameter updates. I am also interested in mechanistic interpretability and in approaches that improve the robustness and safety of deep learning methods and AI in general.

My honours thesis was supervised by Michael Thielscher and Alan Blair and before that I completed research placements with Andreas Schutt and Haris Aziz.

Publications

Equivalence of Context and Parameter Updates in Modern Transformer Blocks
Adrian Goldwaser, Michael Munn, Javier Gonzalvo and Benoit Dherin ICML 2026 (oral presentation)
[arxiv]
Transmuting prompts into weights
Hanna Mazzawi, Benoit Dherin, Michael Munn, Adrian Goldwaser, Michael Wunder and Javier Gonzalvo arXiv preprint, 2026
[arxiv]
Simulating the Implicit Effect of Learning Rates in Gradient Descent
Adrian Goldwaser, Bruno Mlodozeniec and Hong Ge Bridging the Gap between Practice and Theory (BGPT) at ICLR 2024
[pdf]
Understanding Sparse Feature Updates in Deep Networks using Iterative Linearisation
Adrian Goldwaser and Hong Ge arXiv preprint. Earlier version in Optimization for Machine Learning (OPT-ML) at NeurIPS 2022
[arxiv] [doi] [workshop pdf] [bib]
Deep Reinforcement Learning for General Game Playing
Adrian Goldwaser and Michael Thielscher AAAI 2020 (oral presentation)
[pdf] [bib] [doi]
Optimal Torpedo Scheduling
Adrian Goldwaser and Andreas Schutt JAIR 2018
[pdf] [bib] [doi]
Optimal Torpedo Scheduling
Adrian Goldwaser and Andreas Schutt CP 2017 - named Best Student Paper
[pdf] [bib] [doi]
Coalitional Exchange Stable Matchings in Marriage and Roommate Markets
Haris Aziz and Adrian Goldwaser AAMAS 2017 (extended abstract)
[pdf] [bib] [doi]
Priority Search with MiniZinc
Thibaut Feydy, Adrian Goldwaser, Andreas Schutt, Peter J. Stuckey and Kenneth D. Young ModRef 2017
[pdf] [bib]

Teaching

Service

  • Reviewer for ICML 2026 — awarded Gold Reviewer

Personal

I have been juggling since 2006, you can see some projects related to that here.