Don't miss this opportunity!🌱The application period for the in-person program at Mila closes on April 14.

Join us at the 2024 Annual Meeting of the Cognitive Neuroscience Society in Toronto from April 13-16 at Booth 101. 🎉

Our workshop is on April 13 from 11:30 am to 12:30 pm, exploring embracing high-dimensional data with XR-based experiments.

#CNS2024

Check out the lineup for #TAMeG2024 !

✅26 talks/data blitzes
✅Blake Richards keynote
✅Panel: episodic memory in AI Blake Richards sara aronowitz Nick Frosst Morgan Barense
✅ Open bar, awards, powerpoint karaoke

Registration is open!
TAMeG.ca

Check this out! Our very own Tony Zador is one of the organizers for the upcoming NAISYS Meeting in September! He is joined by Blake Richards and Doris Tsao to bring you a great line up of Speakers.
Click the link to find out more!

meetings.cshl.edu/meetings.aspx?…

Standard transformer-based language models use the same amount of compute for each token. Our new method, which we call Mixture-of-Depths, allows transformers to instead learn to dynamically allocate compute to specific positions in a sequence.

arxiv.org/abs/2404.02258

*Mixture-of-Depths: Dynamically allocating compute in transformer-based LMs*
by sam ritter Blake Richards Adam Santoro

A variant of MoEs having only a single expert per block which can be either skipped or applied per-token up to some given capacity.

arxiv.org/abs/2404.02258

Students, there is still a bit of time to sign up for our NeuroAI course! We've got an awesome curriculum for you, courtesy of folks like xaq , Eva Dyer , Nikolaus Kriegeskorte , Surya Ganguli , megan peters 🧠 on bsky & 🐘neuromatch.social , Blake Richards , Grace Lindsay , Guillaume Dumas . Join us!

Universal Differential Equations (UDE) as a Common Modeling Language for Neuroscience arxiv.org/abs/2403.14510 — Great review & perspective, Artificial Cognitive Systems! Ping Blake Richards Guillaume Lajoie Irina Rish

Substantial preprint: we explore how adaptive decoders for brain-computer interfaces influence learning in the brain. We mix theory and experiments to uncover how closed-loop error manipulation shapes neural representations in the brain. Useful to probe learning and improve BCI!

Our systems neuroscience lab University of Toronto University of Toronto Scarborough is recruiting postdocs! We use in vivo recording and optogenetic manipulations during innovative behavioural tasks to investigate the neural dynamics of emotional memory. Please RT! See details below

Recording of our
CosyneMeeting 2024 workshop

'what are the learning rules of the brain?'

online here: youtu.be/ivVBrvcTxlI

I really enjoyed it, and learned a lot.

The Prisma multi modal mech interp discord is shaping up to be the most fascinating blend of AI alignment people and artsy designer stable diffusion esque people.

The subculture I never knew I needed.
discord.com/invite/2U2N8Qm…

Teaser...

#brain #asymmetry
#human #defining cognition
#machine #learning

It is astounding how many people know The Bitter Lesson but haven't internalized it

If you haven't heard of it no biggie 👀📕👇 (very short I promise 🤓)

But once you read it, actually follow through by designing or using models that implement it!! 🙏😅

incompleteideas.net/IncIdeas/Bitte…

LLMs are amazing but they’re not grounded in external, embodied environments. That’s why I’m excited to finally be able to talk about the project I’ve been working on for over a year: SIMA, an agent that can follow natural language in video games! dpmd.ai/3TiYV7d

I'm excited to release Prisma, a mechanistic interpretability library for multimodal models like CLIP and ViTs. Incubated at Blake Richards's lab & in collab with Neel Nanda.

Recent mech interp work has focused on language, but many techniques transfer. Behold, the dogit lens:

This is a fantastic new (and super fun to play with) resource for mechanistic interpretability work. 😀

Well done Sonia Joseph and team!!!!

If you are interested to learn more about how the RL_Conference came into existence listen to this podcast. Tomorrow I will share more information on another important point that was missed during the interview. #RL4Life

“our results suggest that the ventral visual cortex, like DNN models, provides a texture-like basis set of features, and that further neural computations, perhaps downstream of IT, are necessary to account for the shape selectivity of visual perception.”

openreview.net/forum?id=D6Yvy…

Excited to share our findings probing how engram ensembles are formed endogenously! tldr? Both neuronal excitability in min-h before an event & pre-existing functional connectivity help determine which neurons are allocated to an engram ensemble. tinyurl.com/y67mbvuv 1/6