Hadi Vafaii
@hadivafaii.bsky.social
860 followers 180 following 120 posts
Postdoc at UC Berkeley, Redwood Center | 🧠🤖 | 🎹 | 🎾 | https://mysterioustune.com/
Posts Media Videos Starter Packs
Pinned
hadivafaii.bsky.social
Elegant theoretical derivations are exclusive to physics. Right?? Wrong!

In a new preprint, we:
✅ "Derive" a spiking recurrent network from variational principles
✅ Show it does amazing things like out-of-distribution generalization
👉[1/n]🧵

w/ co-lead Dekel Galor & PI @jcbyts.bsky.social

🧠🤖🧠📈
Reposted by Hadi Vafaii
nicolecrust.bsky.social
Wonderful to see this high-level conversation about reinforcement learning happen not just “at the top” but across many professional levels (including trainees). 👏👏👏 to everyone involved. Such a thoughtful way to expedite progress. I look forward to following along.
hadivafaii.bsky.social
I'm excited to announce that we've added two more contenders to the RL Debate Series!

- Anne Collins (Professor @ucberkeleyofficial.bsky.social)
- Niels Leadholm (Research Manager @thousandbrains.org)

Things are heating up!!

See the flyer for details👇

🌎 sensorimotorai.github.io/debates/

🧠🤖🧠📈
hadivafaii.bsky.social
The first meeting of the 'RL Debate Series' is now on YouTube!

Watch @elisennesh.bsky.social abolish the value function:

youtube.com/watch?v=E0A0...

🧠🤖🧠📈
Reposted by Hadi Vafaii
hadivafaii.bsky.social
Now accepted to #NeurIPS2025 - see y'all in sunny San Diego ☀️

Also, big thanks to our reviewers. Their deep reading of the paper and thoughtful feedback really helped sharpen our contributions.

Stay tuned for a much-enhanced final version in a couple weeks!
hadivafaii.bsky.social
Elegant theoretical derivations are exclusive to physics. Right?? Wrong!

In a new preprint, we:
✅ "Derive" a spiking recurrent network from variational principles
✅ Show it does amazing things like out-of-distribution generalization
👉[1/n]🧵

w/ co-lead Dekel Galor & PI @jcbyts.bsky.social

🧠🤖🧠📈
Reposted by Hadi Vafaii
gracekind.net
Very good juxtaposition
hadivafaii.bsky.social
Join 170+ interdisciplinary researchers from around the world who are pursuing a deeper, first-principles understanding of AI and active learning 🧠🤖🧠📈

💬 Join the conversation on Slack: join.slack.com/t/sensorimot...
🌍 Our website: sensorimotorai.github.io

See you on Slack!

🧵[5/5]
Join Sensorimotor AI on Slack
Slack is a new way to communicate with your team. It’s faster, better organized, and more secure than email.
join.slack.com
hadivafaii.bsky.social
Not all 'active' learning is 'reinforcement' learning.

➡️ This suggests we need to upgrade—or even replace—RL with a more general theory of active learning that's not solely reward-based

✅Which is why we are launching the RL Debates

More info: sensorimotorai.github.io/debates/

🧵[4/5]

🧠🤖🧠📈
hadivafaii.bsky.social
For decades, reinforcement learning (RL) has been synonymous with "active learning".

This figure from Sutton & Barto's classic textbook summarizes the standard view👇

🤔 But what if the premise of an 'external, scalar reward' is a simplification we need to move past?

🧵[3/5]
hadivafaii.bsky.social
🌍 More info on our website: sensorimotorai.github.io/debates/
💬 Join us on Slack: join.slack.com/t/sensorimot...

💡 The main motivation: recognizing how central "action" is in all things intelligence

Perception, cognition, and knowledge are fundamentally intertwined with "action"

🧵[2/5]
hadivafaii.bsky.social
What drives behavior in living organisms? And how can we design artificial agents that learn interactively?

📢 To address these, the Sensorimotor AI Journal Club is launching the "RL Debate Series"👇

w/ @elisennesh.bsky.social, @noreward4u.bsky.social, @tommasosalvatori.bsky.social

🧵[1/5]

🧠🤖🧠📈
hadivafaii.bsky.social
Interested in attending future meetings?

Join 150+ researchers from around the world who want a deeper understanding of AI 🧠🤖

Here's our Slack invite:

join.slack.com/t/sensorimot...

🧠🤖🧠📈
Join Sensorimotor AI on Slack
Slack is a new way to communicate with your team. It’s faster, better organized, and more secure than email.
join.slack.com
hadivafaii.bsky.social
In the second meeting of 'Sensorimotor AI Journal Club' we covered:

📜 (B)ayesian (O)nline learning in (N)on-stationary (E)nvironments (BONE; openreview.net/forum?id=ose...)

Many thanks to the first author Gerardo for an amazing presentation!

🎥 watch it here: www.youtube.com/watch?v=49PP...

🧠🤖🧠📈
Bayesian online learning in non-stationary environments (BONE) by Duran-Martin et al., 2025
YouTube video by Sensorimotor AI
www.youtube.com
hadivafaii.bsky.social
"2000s Dad Rock" is a thing 🙂
hadivafaii.bsky.social
We need bio-magic for consciousness… until we build it in silico
hadivafaii.bsky.social
Interested? Join our Slack, along with 110+ researchers from around the world who want a deeper understanding of AI 🧠🤖

Here's our Slack: join.slack.com/t/sensorimot...

🧵[4/4]
Join Sensorimotor AI on Slack
Slack is a new way to communicate with your team. It’s faster, better organized, and more secure than email.
join.slack.com
hadivafaii.bsky.social
This time we covered "Bayesian Online Natural Gradient (BONG)".

📅 Next meeting (September 4):

We explore BONE (openreview.net/forum?id=ose...), which tackles a crucial problem:

💡 How can AI systems adapt when the world keeps changing? (Non-stationary environments)

🧵[3/4]
A unifying framework for generalised Bayesian online learning in...
We propose a unifying framework for methods that perform probabilistic online learning in non-stationary environments. We call the framework BONE, which stands for generalised (B)ayesian (O)nline...
openreview.net
hadivafaii.bsky.social
btw, I know everyone knows these points about real-world deployment, energy efficiency, etc.

But the incentive structures often push us toward short-term wins over fundamental understanding.

In our JC, we read papers that take the longer view, exploring deeper principles.

🧵[2/4]
hadivafaii.bsky.social
Our first 'Sensorimotor AI Journal Club' meeting was a blast!

(...which, we might as well just call the "AI Club for Non-conformists"👇😎)

📽️ full presentation: youtube.com/watch?v=efc7...

🧵[1/4]

🧠🤖🧠📈