I have started to organise the connections between my interests with a graph, which I have displayed on my webpage thomaswalker1.github.io. There is also an option for others to suggest further connections between ideas, so please do reach out if you have any!

PhD student at Rice University under the supervision of Professor Richard Baraniuk; exploring the geometry of machine learning models. thomaswalker1.github.io

I have just read a fascinating neural network paper from 1994 that explores exactly the concept I wanted to explore, albeit in a very simple setting. What are some of the earlier papers people have found remarkably relevant to modern day machine learning?

The subtle distinction between the geometry of machine learning and geometric machine learning. thomaswalker1.github.io/blog/geometry_…

What are the best ways to incorporate LLMs into your research workflow? I'm toying with using one as a guide as I read a paper. However, it still feels a bit clunky. A GitHub co-pilot for reading papers would be useful.

Are we Making Interpretability Hard for Ourselves? thomaswalker1.github.io/blog/are_we_ma…

Professor Randall Balestriero discussing some exciting research he has been working on recently, in particular around spline-based interpretability. • Why reconstruction learning can fail for perception • How deep nets partition space like a crystal • Spline theory new insights

Indeed, it is that simple! The wiggliness induced by each layer allows NNs to approximate non-linear functions. More layers -> more possible wiggle -> more non-linearity. A nice way of thinking about this is imagining NNs doing origami on an elastic piece of paper!

Models do not necessarily work in natural language. We should probably let them parse the task rather than prompting them directly. thomaswalker1.github.io/blog/you_shall…

Who got time to wait for delayed generalization (grokking)? We introduce GrokAlign, a provable solution to speed up the alignment between your model and your training data resulting in faster convergence + visual probing of your DN! Ofc it uses splines :) arxiv.org/abs/2506.12284

Very happy to have presented my poster on “GrokAlign: Geometric Characterisation and Acceleration of Grokking” at the High Dimensional Learning Workshop of ICML Conference thomaswalker1.github.io/blog/grokalign…