wtf are we even talking about at this point?

mice are not monkeys are not humans, none of them are RNNs. There are 1000 things each can/not do bc they have different bodies and live in different environments.

nvm how wild it is that the article manages to go from NHP neuro to AI & anti-China lol??

i also can't read a single email on horde because they all give me this error

sounds about right

Manuel Brenner and I managed to organize this workshop not knowing we were gonna end up being new neighbors this month. The lineup speaks for itself.

There were some last minute cancellations so we jumped in to fill their big shoes and will talk about the future directions of the new labs too!

Luckily goodbye is temporary and many are coming to Bernstein in a week.

If you’re also at Bernstein and are interested, or just want to chat about science, life, or whatever, send me a message!

p.s. Frankfurt is probably the easiest place to leave whether you need to fly or take the train. FYI.

Lastly, on that note about people:

4 years in Tübingen flew by, and I truly had to adjust to "the village" when I first came.

But now I really, really cannot imagine leaving.

I’m thankful for so many people’s support, guidance, love, and sheer buffoonery.

For the next little while, a big focus will be on developing and applying ML methods,

both for assisting “white-box” mechanistic & biophysical models (e.g., simulation-based inference)

and to serve as “blackbox” generative models that can subsequently be dissected and interpreted.

Coming from engineering, I wanted to translate what we know from animal and theory work to benefit human lives.

Now, we will use multimodal & multi-species brain data, mechanistic models, and machine learning to link data and theory,

while driving method development in all the above.

The research group will tackle a fairly broad but well-defined technical question:

What can we infer about the brain that we can’t easily measure, from signals that *can* be easily but (often) poorly measured?

This question has been close to my heart from even before my own PhD.

LOL alright we're done here. I can now be replaced by AI.

asked Claude to brainstorm for a new blogpost given two previous ones and it immediately starts talking like me.

let's just say it's very...disconcerting.

also TIL I'm a snarky asshole lol.

sending everybody to this @patrickmineault.bsky.social blogpost because my rant on the apparent hubris in neuroscience is ... not as nicely worded.

www.neuroai.science/p/what-are-f...

I've been pondering a lot on what is the most efficient entry route into the "fancy" ML stuff today for a dummy*.

this is the roadmap I ended up with, and it's disappointing how little prob/stats you need—just Gaussians 🐢.

*🙋‍♂️ me. I was the dummy, as I was getting ready to start a postdoc 4yrs ago.

arxiv.org/abs/2303.06349 (LOVE this paper btw)

the specific instance I have in mind is changing the standard recurrent matrix to complex diagonal in a vanilla RNN, so it's rather adding something. But it's subtractive if you think from the other direction so it's an inconsequential semantic debate?

yes!!! finally a fitting definition grounded in history that I hope everyone can adopt.

(plus shameless self-plug just to tell the communism joke)