Mathstodon

0 posts0 participants0 posts today

Tero Keski-ValkamaHow to formulate exploration-exploitation trade-off better than all the hacks on top of Bellman equation?We can first of all simply estimate the advantage of exploration by Monte-Carlo in a swarm setting: Pitting fully exploitative agents against fully exploitative agents which have the benefit of recent exploration. This can be easily done by lagging policy models.Of course the advantage of exploration needs to be divided by the cost of exploration, which is linear to the number of agents used in the swarm to explore at a particular state.Note that the advantage of exploration depends on the state of the agent, so we might want to define an explorative critic to estimate this.What's beautiful in this formulation is that we can incorporate autoregressive <a href="https://rukii.net/tags/WorldModels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#WorldModels</a> naturally, as the exploitative agents only learn from rewards, but the explorative agents choose their actions in a way which maximizes the improvement of the auto-regressive World Model.It brings these two concepts together as sides of the same coin.Exploitation is reward-guided action, exploration is auto-regressive state transition model improvement guided action.Balancing the two is a swarm dynamic which encourages branching where exploration has an expected value in reward terms. This can be estimated by computing the advantage of exploitative agents utilizing recent exploration versus agents which do not, and returning this advantage to the points of divergence between the two.<a href="https://rukii.net/tags/mathematics" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#mathematics</a> <a href="https://rukii.net/tags/ReinforcementLearning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#ReinforcementLearning</a> <a href="https://rukii.net/tags/RL" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#RL</a> <a href="https://rukii.net/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#AI</a> <a href="https://rukii.net/tags/LLMs" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#LLMs</a>

LavX NewsGoogle DeepMind's Bold Leap into AI World SimulationIn a groundbreaking move, Google DeepMind is assembling a new team led by Tim Brooks to develop AI models capable of simulating the physical world. This initiative aims to push the boundaries of artif...<a href="https://news.lavx.hu/article/google-deepmind-s-bold-leap-into-ai-world-simulation" rel="nofollow noopener noreferrer" target="_blank">https://news.lavx.hu/article/google-deepmind-s-bold-leap-into-ai-world-simulation</a><a href="https://mastodon.cloud/tags/news" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#news</a> <a href="https://mastodon.cloud/tags/tech" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#tech</a> <a href="https://mastodon.cloud/tags/ArtificialIntelligence" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#ArtificialIntelligence</a> <a href="https://mastodon.cloud/tags/WorldModels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#WorldModels</a> <a href="https://mastodon.cloud/tags/GoogleDeepMind" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#GoogleDeepMind</a>

LavX NewsGoogle DeepMind's Bold Leap into AI World SimulationIn a groundbreaking move, Google DeepMind is assembling a new team led by Tim Brooks to develop AI models capable of simulating the physical world. This initiative aims to push the boundaries of artif...<a href="https://news.lavx.hu/article/google-deepmind-s-bold-leap-into-ai-world-simulation" rel="nofollow noopener noreferrer" translate="no" target="_blank">https://news.lavx.hu/article/google-deepmind-s-bold-leap-into-ai-world-simulation</a><a href="https://mastodon.social/tags/news" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#news</a> <a href="https://mastodon.social/tags/tech" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#tech</a> <a href="https://mastodon.social/tags/ArtificialIntelligence" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#ArtificialIntelligence</a> <a href="https://mastodon.social/tags/WorldModels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#WorldModels</a> <a href="https://mastodon.social/tags/GoogleDeepMind" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#GoogleDeepMind</a>

WinbuzzerGoogle DeepMind is creating a new team to develop AI world models, focusing on simulating real-world dynamics and AGI <a href="https://mastodon.social/tags/AI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#AI</a> <a href="https://mastodon.social/tags/DeepMind" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#DeepMind</a> <a href="https://mastodon.social/tags/GoogleDeepMind" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#GoogleDeepMind</a> <a href="https://mastodon.social/tags/WorldModels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#WorldModels</a> <a href="https://mastodon.social/tags/GenAI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#GenAI</a> <a href="https://mastodon.social/tags/AGI" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#AGI</a> <a href="https://mastodon.social/tags/Robotics" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#Robotics</a> <a href="https://mastodon.social/tags/Gaming" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#Gaming</a> <a href="https://mastodon.social/tags/AIResearch" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#AIResearch</a> <a href="https://mastodon.social/tags/MachineLearning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#MachineLearning</a> <a href="https://mastodon.social/tags/AIModels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#AIModels</a> <a href="https://mastodon.social/tags/Alphabet" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#Alphabet</a><a href="https://winbuzzer.com/2025/01/08/google-deepmind-forms-specialized-team-for-ai-world-models-xcxwbn" rel="nofollow noopener noreferrer" translate="no" target="_blank">https://winbuzzer.com/2025/01/08/google-deepmind-forms-specialized-team-for-ai-world-models-xcxwbn</a>

PKPs Powerfromspace1@wesroth Ep 12-20-2024 🤖📰😅Watch " <a href="https://mstdn.social/tags/Genesis" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#Genesis</a> Project Just UNLEASHED Legions of <a href="https://mstdn.social/tags/Robots" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#Robots</a> from SIMULATION to REALITY..."<a href="https://youtu.be/IAmrSaDW88I?feature=shared" rel="nofollow noopener noreferrer" translate="no" target="_blank">https://youtu.be/IAmrSaDW88I?feature=shared</a><a href="https://mstdn.social/tags/ai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#ai</a> <a href="https://mstdn.social/tags/llm" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#llm</a> <a href="https://mstdn.social/tags/robotics" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#robotics</a> <a href="https://mstdn.social/tags/worldmodels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#worldmodels</a> combined with <a href="https://mstdn.social/tags/agi" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#agi</a> <a href="https://mstdn.social/tags/o3" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#o3</a> Coming soon 🔜 to <a href="https://mstdn.social/tags/skynet" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#skynet</a> v0.1 😈

Joe GANIO-MEGO<a href="https://xn--baw-joa.social/@unikonstanz" class="u-url mention" rel="nofollow noopener noreferrer" target="_blank">@unikonstanz</a> Pretty cool initiative!!!If anybody is looking for inspirations about subjects feel free to use and are re-use any of my models about human population and competitiveness. Totally open science models with code listings available here:<a href="https://osf.io/mg82f/" rel="nofollow noopener noreferrer" translate="no" target="_blank">https://osf.io/mg82f/</a><a href="https://mastodon.social/tags/openscience" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#openscience</a> <a href="https://mastodon.social/tags/worldmodels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#worldmodels</a> <a href="https://mastodon.social/tags/population" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#population</a>

Michal ValkoBYOL-Explore getting the apple! <a href="https://www.deepmind.com/publications/byol-explore-exploration-by-bootstrapped-prediction" rel="nofollow noopener noreferrer" target="_blank">https://www.deepmind.com/publications/byol-explore-exploration-by-bootstrapped-prediction</a> w/ Zhaohan Daniel Guo, Shantanu Thakoor, Miruna Pislar, Corentin Tallec, Florent Altché, Bernardo Avila Pires, Robin (Yunhao) Tang, Alaa Saade, Jean-Bastien Grill, Mohammad Gheshlaghi Azar, Bilal Piot, Remi Munos, Daniele Calandriello <a href="https://mastodon.social/tags/neurips2022" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#neurips2022</a> <a href="https://mastodon.social/tags/worldmodels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#worldmodels</a> <a href="https://mastodon.social/tags/reinforcementlearning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#reinforcementlearning</a> <a href="https://mastodon.social/tags/ai" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#ai</a>

Shiwali Mohan | शिवाली मोहनHowever, as soon as you start thinking about <a href="https://sigmoid.social/tags/language" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#language</a> as communication between <a href="https://sigmoid.social/tags/intelligent" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#intelligent</a> <a href="https://sigmoid.social/tags/agents" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#agents</a>, you have to start thinking about how language is connected with <a href="https://sigmoid.social/tags/worldmodels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#worldmodels</a> Lawrence Barsalou wrote about this phenomenon in 1999 while critiquing <a href="https://sigmoid.social/tags/language" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#language</a> <a href="https://sigmoid.social/tags/research" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#research</a> in <a href="https://sigmoid.social/tags/psychology" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#psychology</a>. He argued that <a href="https://sigmoid.social/tags/language" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#language</a> evolved in humans for co-coordinating collaborative actions in human teams. But, <a href="https://sigmoid.social/tags/psycholinguistics" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#psycholinguistics</a> approached language as if its primary function was archival.<a href="https://web.archive.org/web/20031206190916id_/http://userwww.service.emory.edu:80/~barsalou/Papers/Disc_Proc_Files/disc_proc_98.pdf" rel="nofollow noopener noreferrer" target="_blank">https://web.archive.org/web/20031206190916id_/http://userwww.service.emory.edu:80/~barsalou/Papers/Disc_Proc_Files/disc_proc_98.pdf</a>

Manuel Baltieri<a href="https://mathstodon.xyz/tags/Introduction" class="mention hashtag" rel="tag">#Introduction</a> time.TL;dr: maths to formally study <a href="https://mathstodon.xyz/tags/agents" class="mention hashtag" rel="tag">#agents</a>, <a href="https://mathstodon.xyz/tags/worldmodels" class="mention hashtag" rel="tag">#worldmodels</a> and <a href="https://mathstodon.xyz/tags/representations" class="mention hashtag" rel="tag">#representations</a>I'm currently a Chief Researcher at Araya, a Tokyo based startup with the goal to understand (artificial) <a href="https://mathstodon.xyz/tags/consciousness" class="mention hashtag" rel="tag">#consciousness</a>.My work focuses on general principles for the <a href="https://mathstodon.xyz/tags/origins" class="mention hashtag" rel="tag">#origins</a> of <a href="https://mathstodon.xyz/tags/agency" class="mention hashtag" rel="tag">#agency</a>, <a href="https://mathstodon.xyz/tags/life" class="mention hashtag" rel="tag">#life</a> and <a href="https://mathstodon.xyz/tags/cognition" class="mention hashtag" rel="tag">#cognition</a>. I previously used the <a href="https://mathstodon.xyz/tags/freeenergyprinciple" class="mention hashtag" rel="tag">#freeenergyprinciple</a> (+ <a href="https://mathstodon.xyz/tags/activeinference" class="mention hashtag" rel="tag">#activeinference</a>) as my main framework, but I'm now looking into other directions, mostly using <a href="https://mathstodon.xyz/tags/categorytheory" class="mention hashtag" rel="tag">#categorytheory</a> applied to <a href="https://mathstodon.xyz/tags/systemstheory" class="mention hashtag" rel="tag">#systemstheory</a>.

Michal ValkoAnd now it comes, presenting BYOL-Explore! <a href="https://arxiv.org/abs/2206.08332" rel="nofollow noopener noreferrer" target="_blank">https://arxiv.org/abs/2206.08332</a> With Zhaohan Daniel Guo, Shantanu Thakoor, Miruna Pislar, Bernardo Ávila Pires, Florent Altché, Corentin Tallec, Alaa Saade, Daniele Calandriello, Jean-Bastien Grill, Robin (Yunhao) Tang, Rémi Munos, Mohammad Gheshlaghi Azar, Bilal Piot Comenius University in Bratislava DeepMind Photo (c) Stanislav Griguš <a href="https://mastodon.social/tags/artificialintelligence" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#artificialintelligence</a> <a href="https://mastodon.social/tags/worldmodels" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#worldmodels</a> <a href="https://mastodon.social/tags/reinforcementlearning" class="mention hashtag" rel="nofollow noopener noreferrer" target="_blank">#reinforcementlearning</a>

Recent searches

Search options

Administered by:

Server stats:

#worldmodels