Joachim Winther Pedersen
@JoachimWintherPostdoc at IT University of Copenhagen, mostly interested in black-box meta-reinforcement learning, developmental algorithms, and structurally flexible ANNs.
Similar User
@risi1979
@mayalen_etc
@kvfrans
@EIiasNajarro
@stenichele
@err_more
@eplantec
@gallorob2
@SidneyPontesF
@kyrre2000
@PierroTweets
@malganius
@khoribe3
@adam_gaier
@andrea_ferigo
There's another work by @JoachimWinther with adaptable, trainable weights instead of fixed, scalar ones. "Each neuron in our network has a small matrix associated with it. ... making each neuron in the network into a tiny RNN." 4/ x.com/JoachimWinther…
Looking forward to presenting this project @GeccoConf next month. arxiv.org/pdf/2305.15945… In this video, two networks are performing in the same environment. One is a normal FFNN with optimized weights, and the other has random weights and optimized neuro-centric parameters:
Transformers can be slow for real-time applications like robotics. We study if modern recurrent architectures, like xLSTM and Mamba, can be faster alternatives. Experiments on 432 tasks show that they compare favourably in terms of performance and speed 🎃 arxiv.org/abs/2410.22391
Are LLMs able to exhibit traits usually associated with ALife approaches? How can ALife perspectives help improve the current limitations of LLMs? Read more in this review that examines LLMs from a different angle, featuring some nice graphics along the way!
New paper: "From Text to Life: On the Reciprocal Relationship between Artificial Life and Large Language Models"! We investigate the potential synergies between LLMs and Artificial Life 🤝 On the one hand, LLMs can serve as tools for ALife research 🛠️ On the other hand,…
❤️this! I dream of a computer that could do this after construction (I guess my brain is a bit like this, but SDK has a lot to be desired) developer.nvidia.com/blog/autodmp-o…
I'm on the job market! Looking for positions where I can leverage my background in AI/ML and computational neuroscience. Below you'll find posts for a few of my publications/interests. I'm open to relocating & industry or academic positions. Feel free to reach out for a chat :)
Looking forward to giving a keynote at this workshop on Guided Self-Organization guided-self.org/gso-2025.html Consider submitting an extended abstract. Deadline is September 27th.
If we want generally capable agents in the real world, maybe more research should be devoted to plastic neural networks 🤔😉 Cool paper shedding some more light on the remaining challenges of ANNs!
"It has been unclear whether or not deep learning methods work in continual learning settings. Here we show that they do not—that standard deep learning methods gradually lose plasticity in continual learning settings until they learn no better than a shallow network."
Online RL algorithms can solve problems for which we have no prior knowledge or models. Improvements in such algorithms would be huge. For example, robots won't have to be manufactured so precisely. An imprecise hotchpotch of sensors and actuators could learn to do a lot.
Do neural networks dream of internal goals? We confirm RNNs trained to play Sokoban with RL learn to plan. Our black-box analysis reveals novel behaviors such as agents “pacing” to gain thinking time. We open-source the RNNs as model organisms for interpretability research.
Thanks to @khoribe3 and @m_crosscombe for inviting me as speaker for the ECCCI workshop at this year's ALife @ALifeConf I got to talk about how we can build neural networks as collectives of RL agents. #ALIFE2024
Meta learning developmental encodings by @miltonllera 🧬 #ALife2024
Let’s grow some artificial neural networks 🧠🧬 #ALIFE2024 #GROWAI
Super excited to share our new work (and the first of my PhD) : "Evolving Self-Assembling Neural Networks: From Spontaneous Activity to Experience-Dependent Learning" We propose Lifelong Neural Developmental Programs for continually self-organizing artificial neural networks !
Michael Beukman introduces JaxLife, an evolutionary ecosystem simulator that focuses on agents' high-level behaviors #ALIFE2024 @ALifeConf
Riversdale Waldegrave applies Developmental Graph CA to re-create network motif profiles of real biological network data #ALIFE2024 @ALifeConf
Many outstanding & creative ideas on interlinking LLMs with ALife and vice versa 🤖🔄🧬 We need this type of 'translation' & cross-fertilization to drive progress! 1. Self-rewarding LLMs ⇔ Autonomy 🦾 2. Reflexion ⇔ Self-regulation 🤔 3. Tool use ⇔ Embodiment 🛠 4.…
New paper: "From Text to Life: On the Reciprocal Relationship between Artificial Life and Large Language Models"! We investigate the potential synergies between LLMs and Artificial Life 🤝 On the one hand, LLMs can serve as tools for ALife research 🛠️ On the other hand,…
We also noticed that more and more LLM papers are starting to incorporate ALife-related concepts as key components (even though they might not call them that ... yet!)
Proteins in cells in organs in organisms in ecosystems... Now imagine cellular automata (CA) within cells of a larger CA. That's a hierarchical CA (HCA). @KamBielawski et al. just demonstrated at @GeccoConf that such HCAs can be more evolvable than CAs. dl.acm.org/doi/pdf/10.114…
I really like this book and have also found it useful for explaining open-endedness to students and getting them to think about natural evolution in a different light, i.e. not as optimization, as is a common pitfall for CS students.
When you realize your book can still go viral 9 years after it was published. (Thanks @bmix012 !)
Can vision transformers learn something useful when you randomly shuffle it's layers during training? Turns out they can! We present "LayerShuffle: Enhancing Robustness in Vision Transformers by Randomizing Layer Execution Order" 🧵 arxiv.org/abs/2407.04513
Very nice indeed - learning a good representation for interestingness and novelty!
We are happy to present "Meta-Learning an Evolvable Developmental Encoding"! 🧬 Generative models can work as learnable representations for blackbox optimization but they are not designed to be easily searchable. We present a system that can meta-learn such representation by…
United States Trends
- 1. #JusticeforDogs N/A
- 2. $CUTO 9.165 posts
- 3. ICBM 186 B posts
- 4. $EFR 2.224 posts
- 5. The ICC 241 B posts
- 6. Netanyahu 519 B posts
- 7. Denver 32,5 B posts
- 8. Jussie Smollett 5.871 posts
- 9. Illinois Supreme Court 6.151 posts
- 10. #KashOnly 40,1 B posts
- 11. #AcousticGuitarCollection 2.277 posts
- 12. DeFi 127 B posts
- 13. Dearborn 6.431 posts
- 14. #ATSD 10,3 B posts
- 15. #AtinySelcaDay 9.963 posts
- 16. chenle 127 B posts
- 17. Volvo 5.465 posts
- 18. Katie Couric 2.343 posts
- 19. Flat 52,1 B posts
- 20. Bezos 39,6 B posts
Who to follow
-
Sebastian Risi
@risi1979 -
Mayalen Etcheverry
@mayalen_etc -
Kevin Frans
@kvfrans -
Elias Najarro
@EIiasNajarro -
Stefano Nichele
@stenichele -
L Soros
@err_more -
erwan plantec
@eplantec -
gallorob
@gallorob2 -
Sidney Pontes-Filho
@SidneyPontesF -
Kyrre Glette
@kyrre2000 -
Alessandro Pierro
@PierroTweets -
Jessica Mégane
@malganius -
Kazuya Horibe (@[email protected] 🐘)
@khoribe3 -
adam gaier
@adam_gaier -
Andrea Ferigo
@andrea_ferigo
Something went wrong.
Something went wrong.