No “Zero-Shot” Without Exponential Data: Pretraining Concept Frequency Determines Multimodal Model Performance

repo: github.com/bethgelab/freq…
hf: huggingface.co/datasets/bethg…
abs: arxiv.org/abs/2404.04125

Biological Psychiatry

Psychiatric Illnesses as Disorders of Network Dynamics biologicalpsychiatrycnni.org/article/S2451-…

Yes, confusion, and also some subtleties ... and it's interesting to see where and why current SOTA algos fail.

More bad news for free-lunch aficionados

Thanks Hamidreza!

Wow, jelly implementation of the Lorenz attractor

'Maths underpins most of the challenges we face today.'

Sophie Maclean, PGR in Mathematics, shared her insights on Channel 5 about the crucial role of maths in our lives 🧮✨ The UK's first ever Maths Summit was a critical leap in demonstrating that to the world, she said.

Anybody in or traveling to Ethiopia in my network? I need to help a Sudanese student stuck in Addis Ababa. There's no easy way to send money it seems. 🙏

Oops. The paper itself is here: arxiv.org/abs/2403.02579

Started the Day w/ Reconstructing Dynamical Systems from Neural Data organized by Daniel Durstewitz Il Memming Park

1st talk by the wonderful Bing Wen Brunton on, fields, flows & time-delay embedding

Agreed upon limitation of Dynamical Systems Modeling: out-of-domain generalization! 👀👀

'Transformers_Are_What_You_Dont_Need' stands out as the premier guide, illustrating the limitations of transformers in time series forecasting.

It not only sheds light on why transformers may fall short but also introduces top-tier non-transformer models that surpass them.

A…

Is chaos an analogue sampling scheme? I wrote a Perspective for Nature Reviews Physics discussing connections between classical work on information flow in nonlinear systems, and modern generative ML models

nature.com/articles/s4225…

sites.google.com/research.fcham…
We have a schedule now. Thanks to Ayesha Vermani for making the website! #COSYNE24 CosyneMeeting