We've been running an zoom reading group for (ex-)physicists on machine learning theory. We started in late 2024 / early 2025, inspired by the physics nobel prize 2024.

We meet via zoom, so the group is not restricted to any particular institution or country. However, the timing (evening in Europe) is inconvenient for people located in Asia (barring insomnia).

Currently it takes place Tuesdays, evenings (20:00-21:30 Germany/Italy, 7pm-8:30pm Ireland/UK).
The reason for the evening scheduling is that most of us have day jobs (either in academia or in industry) unrelated to machine learning theory.
Some people join from the States: most of the year, it is starting 2pm Boston, 11am California.

In each session, one person presents a topic or a group of papers.
No background is really needed to participate, other than a physicists' attitude.
The usual expectation from participants is that they should be willing to present something once every few months; the load is quite light.

If you are interested, please contact me by email:
haque ATT pks.mpg.de   or   masudul.haque ATT tu-dresden.de   or   masud.haque ATT gmx.de
If we don't know each other already, please introduce yourself briefly.

Topics that we've been discussing, or are likely to be discussed in coming sessions:
scaling laws and information dynamics in neural networks, Hopfield networks, transformer/attention (algorithm behind ChatGPT and alphafold), connections to random matrix and spin glass physics, diffusion models (algorithm behind dall-e and midjourney), the nature of the optimization landscape and dynamics of optimization, Boltzmann machines, etc.
Sometimes we might pick a machine learning paper (e.g., from arXiV, PNAS or Phys.Rev.E) for discussion.

The participants are physics-trained, which biases the choice of topics and the style of discussion. However, the theme of the journal club series is really machine learning (theory).
(If you are primarily interested in the application of ML to a particular branch of physics, there are probably other more suitable reading groups.)

To give an idea of the topics covered, below is a list of some topics discussed in summer 2025 (May-August):

• Symmetries of neural networks

• Scaling laws for neural networks

• Gaussian processes

• Reinforcement learning intro

• Convolutional neural networks

• Dimensionality reduction

• Alphageometry

• Information bottleneck

• Diffusion models

• Transformers for vision