Alice Dauphin
banner
monomisphere.bsky.social
Alice Dauphin
@monomisphere.bsky.social
PhD student, TU Graz
Links:
๐Ÿง  My neuro-RHEL abstract:
openreview.net/attachment?i...
๐Ÿ“„ Original RHEL paper: openreview.net/forum?id=w1i...
๐ŸŽฎ Interactive demo by Guillaume Pourcel to understand RHEL: guillaumepourcel.github.io/rhel-interac...
December 3, 2025 at 12:08 AM
๐Ÿ“† And if you want to hear more (or rather all) about RHEL Guillaume Pourcel will present the full framework in San Diego ๐Ÿ‡บ๐Ÿ‡ธ on Friday 5:
- Oral at 4:10 pm PST
- Poster from 4:30 pm PST
Maxence Ernoult presents in Copenhagen ๐Ÿ‡ฉ๐Ÿ‡ฐ at #EurIPS2025:
- Oral on Friday 5 at 9 am
- Poster on Thursday 4
December 2, 2025 at 11:39 PM
๐Ÿค“ If you're interested in our latest results, comparison with existing approaches or just curious about new theories for learning in the brain, stop by!
December 2, 2025 at 11:39 PM
๐Ÿ” The key result: Our learning rule is a form of contrastive Hebbian learning and recovers the exact gradients through time!
December 2, 2025 at 11:39 PM
Amazingly there is a new alternative to BPTT in town: RHEL==Recurrent Hamiltonian Echo Learning (Pourcel & Ernoult, 2025)! We took inspiration from Hopfield energy models to design a novel Hamiltonian NN. In our model, neurons are now second-order oscillating ODEs that can be trained using RHEL...
December 2, 2025 at 11:38 PM
๐Ÿ“ข Presenting our work at #NeurIPS2025 ๐Ÿ‡บ๐Ÿ‡ธ today (6-9 pm) @wimlworkshop.bsky.social!

๐Ÿง  The question: How can the brain learn temporal dependencies without an artificial backward pass? BPTT is notoriously harder than BP and is still uncharted territory (Lillicrap & @adamsantoro.bsky.social , 2019).
December 2, 2025 at 11:37 PM
Reposted by Alice Dauphin
๐—ช๐—ต๐—ฒ๐—ฟ๐—ฒ ๐—ฑ๐—ผ ๐—ฟ๐—ฒ๐˜„๐—ฎ๐—ฟ๐—ฑ๐˜€ (๐˜ƒ๐—ฎ๐—น๐˜‚๐—ฒ) ๐—ฐ๐—ผ๐—บ๐—ฒ ๐—ณ๐—ฟ๐—ผ๐—บ?
Implications for reinforcement learning.
Very interesting paper.
doi.org/10.1016/j.ti...
#neuroskyence
September 3, 2025 at 4:27 PM
If you are curious about AI safety/alignment or just look for a good series picturing a future with AI at scale, check my post: open.substack.com/pub/aisig/p/...
What a Swedish Series (Real Humans) Teaches Us About AI Safety
A cool series that brings AI safety concepts to life...
open.substack.com
September 3, 2025 at 8:01 AM
Reposted by Alice Dauphin
Et je viens de voir passer de strip de @unpied.bsky.social qui rรฉsume pas mal ce qu'on peut penser de cette histoire.
June 19, 2025 at 7:17 AM
Reposted by Alice Dauphin
With @eliamascolo.bsky.social we developed an interactive class on biological evolution (focusing on mutations ๐Ÿ‘พ) for science #teacher through a card game ๐ŸŽฒ We crash-tested it with #HighSchool students from Milan and made it available online for broader use: replicards.netlify.app #SciCom #EvoBio
The Replicards โ€“ Replicards
replicards.netlify.app
February 25, 2025 at 11:36 AM