Simon Frieder
ML Researcher. Mathematician. Scientist.
Researcher @ University of Oxford.
PhD/DPhil under T. Lukasiewicz.
Also affiliated to Tim Gowers' group at Cambridge University.
Publication list: Google Scholar.
I want to automate advanced mathematical problem-solving using deep learning.
This homepage is still a bit under construction! In time, it will replace my previous university website at
Oxford
that I don't maintain anymore.
Contact me at: simon.frieder (at) cs.ox.ac.uk
News:
- New paper published: Language Models as Science Tutors
- Reviewer for ICML 2024
- Two accepted papers (first-author) at ICLR 2024 (TP track)
- December 2023: Talk at Aizawa group from the University of Tokyo / National Institute of Informatics (NII) on math and language models.
- LLM vs ITP paper accepted for the MATH-AI workshop at NeurIPS 2023.
- Media reports on my research. My paper on Mathematical Capabilities ChatGPT has been covered by Ars Technica and Synced Review.
- Reviewer for the ICLR 2024 conference.
- Reviewer for NeurIPS MATH-AI and SoLaR workshops.
- Oral at the LOD 2023 conference on the mathematics of Predictive Coding Networks.
- Poster at the Neuromonster 2023 conference.
- March 2023: Invited Talk at Tim Gowers' group on Mathematical Capabilities of ChatGPT.
-
New article on ChatGPT published (accepted for publication at NeurIPS 2023 Datasets & Benchmarks).
Can everyone's favorite language model really do mathematics? Can it show that \(\int |fg| d \mu \leq \left\Vert g\right\Vert_1 \left\Vert f \right\Vert_\infty\)? How about proving that \( A+B \) is closed, if \(A\) is a compact subset of a topological space and \(B\) a closed one?
Read the arXiv preprint to find out. - Reviewer for NeurIPS 2023 conference (Main Track and Datasets and Benchmark Track).
- IJCAI '23. I'm serving as the IJCAI 2023 Assistant Workflow Chair.
Teaching:
- Supervisor for master thesis on language models: Starting winter term 2023, I'm supervising Philip Vonderlind's master thesis on language models and formal games, together with Thomas Lukasiewcz.
- Deep Learning in Healthcare course: I'm a demonstrator for the Deep Learning in Healthcare course, offered in Hilary term 2023 at my department at the University of Oxford.
- Various earlier courses in mathematics and machine learning with varying responsabilities (from 100% teaching to marking grades): Analysis/calculus, linear algebra for physicists, machine learning.
About:
My interests are divided between (in no particular order):- merging mathematics and machine learning;
- pure mathematics;
- rigorous evaluations of language models;
- interactive/automatic theorem proving in natural language;
- theoretical neuroscience and predictive coding networks;