Can AI disover new physics? https://www.youtube.com/watch?v=XRL56YCfKtA https://atlas.cern/Updates/Feature/Machine-Learning Resources for state of the art techniques in AI for fundamental physics https://grok.com/share/bGVnYWN5_800708de-0c3c-4b87-9c28-690cf4466a3e https://home.cern/news/news/physics/how-can-ai-help-physicists-search-new-particles https://physics.mit.edu/news/provably-exact-artificial-intelligence-for-nuclear-and-particle-physics/ https://www.symmetrymagazine.org/article/symmetrys-guide-to-ai-in-particle-physics-and-astrophysics?language_content_entity=und https://arxiv.org/abs/1905.01023 https://www.cmu.edu/news/stories/archives/2021/may/machine-learning-cosmology.html AI for particle physics in CERN CMS develops new AI algorithm to detect anomalies https://home.cern/news/news/experiments/cms-develops-new-ai-algorithm-detect-anomalies https://blog.google/technology/google-deepmind/alphaqubit-quantum-error-correction/ https://en.wikipedia.org/wiki/Machine_learning_in_physics https://youtu.be/MO6ZvA7U3F0?si=qL_zKlya8vqDC5jD https://arxiv.org/abs/2302.04919 Sparse Identification of Nonlinear Dynamics (SINDy): Sparse Machine Learning Models 5 Years Later! https://www.youtube.com/watch?v=NxAn0oglMVw https://www.pnas.org/doi/10.1073/pnas.1517384113 https://journals.aps.org/prl/abstract/10.1103/PhysRevLett.133.233601 https://en.wikipedia.org/wiki/Machine_learning_in_physics https://youtu.be/-zrY7P2dVC4?si=8USsUS0H_U2X-I4r Physics Informed Neural Networks (PINNs) Adds physics bias to the loss function to penalize the model for violating physics, e.g. the divergence of the velocity field should be 0 in incompressible Navier-Stokes equations in fluid dynamics. https://youtu.be/AEOcss20nDA?si=o518qTiUb6PPx0_6 https://www.youtube.com/watch?v=HLUIx6FqAvg https://arxiv.org/abs/2003.04630 https://arxiv.org/abs/1906.01563 learning Langrangians or Hamiltonians using NNs and enforcing them in loss function to conserve energy more and they actually do much better, i love it Physicist Mario Krenn uses artificial intelligence to inspire and accelerate scientific progress. He runs the Artificial Scientist Lab at the Max Planck Institute for the Science of Light, where he develops machine-learning algorithms that discover new experimental techniques at the frontiers of physics and microscopy. He also develops algorithms that predict and suggest personalized research questions and ideas. https://www.youtube.com/watch?v=T_2ZoMNzqHQ https://youtu.be/XRL56YCfKtA?si=cEepwmpkRH9hYAfZ . https://www.nature.com/articles/s43588-022-00281-6 I really like what they do in this paper. They throttle the size of the autoencoder latent space to find how many variables are probably needed to model various dynamical systems, including dynamical systems for which we do not know the ground truth. And the dimension of that latent space corresponds to ground truth number of variables in known equations. Plus it finds some size of latent space even for systems for which we do not have ground truth equations, from which those equations could possibly be derived. We present Panda: a foundation model for nonlinear dynamics pretrained on 20,000 chaotic ODE discovered via evolutionary search. Panda zero-shot forecasts unseen ODE best-in-class, and can forecast PDE despite having never seen them during training (1/8) https://x.com/wgilpin0/status/1925164094010609809 " AI x physics intersection is an endless rabbithole. You can try to: - make better AI/ML systems for modeling physical systems (ferminet <https://deepmind.google/discover/blog/ferminet-quantum-physics-and-chemistry-from-first-principles/>, <https://www.microsoft.com/en-us/research/blog/breaking-bonds-breaking-ground-advancing-the-accuracy-of-computational-chemistry-with-deep-learning/> , cosmology <https://www.pnas.org/doi/full/10.1073/pnas.2022038118>, <https://x.com/wgilpin0/status/1925164094010609809>, lagrangian neural networks <https://arxiv.org/abs/2003.04630>, <https://en.wikipedia.org/wiki/Physics-informed_neural_networks> ,...), - do automated discovery of physical laws ( latent space squeezing https://www.youtube.com/watch?v=XRL56YCfKtA , Physics-tailored machine learning reveals unexpected physics in dusty plasmas <https://www.pnas.org/doi/10.1073/pnas.2505725122>,...) - study existing AI/ML systems using methods from physics (the principles of deep learning theory book <https://arxiv.org/abs/2106.10165> statistical physics <https://arxiv.org/abs/2505.10559>, <https://arxiv.org/abs/2501.19281> , <https://arxiv.org/abs/2506.04374>, <https://www.lesswrong.com/s/mqwA5FcL6SrHEQzox>,...) - design and advance better AI/ML architectures that use insights/methods from physics (diffusion models, <https://en.wikipedia.org/wiki/Diffusion_model#Non-equilibrium_thermodynamics>, flow matching <https://arxiv.org/abs/2210.02747> , <https://en.wikipedia.org/wiki/Quantum_machine_learning>, liquid neural networks <https://arxiv.org/abs/2006.04439>,...), - make neurosymbolic systems to do physics algebraically (formal theorem proving <https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/>, neural search in Lean <https://physlean.com/> <https://arxiv.org/abs/2405.14333>, or outside of Lean <https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/> <https://deepmind.google/discover/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/> , DreamCoder <https://arxiv.org/abs/2006.08381>,...), - do machine designed physics experiments <https://www.youtube.com/watch?v=T_2ZoMNzqHQ> - do machine learning for particle physics <https://atlas.cern/Updates/Feature/Machine-Learning> - discovery of bigger novel theories (což je totálně v plenkách) <https://x.com/MLStreetTalk/status/1957535063660474492> More here: <https://en.wikipedia.org/wiki/Machine_learning_in_physics> etc. " https://interestingengineering.com/innovation/ai-decodes-dusty-plasma-new-forces-physics Physics-tailored machine learning reveals unexpected physics in dusty plasmas https://www.pnas.org/doi/10.1073/pnas.2505725122 "Dusty plasma, a mixture of ions, electrons, and charged dust particles, is common throughout the universe. Understanding and modeling dusty plasma require precise knowledge of the complex interactions between dust particles, but machine learning (ML) is a promising approach for learning such interactions. In simulated data where the ground truth is known, ML models excel at making short-term predictions or inferring governing equations, yet there are few examples where new physical laws have been learned from real experimental data. Here, we demonstrate a scalable ML model that learns the complex interparticle forces from motion of particles in a laboratory dusty plasma. Our experiments and model reveal important discrepancies from common theoretical assumptions in dusty plasmas." https://arxiv.org/abs/2509.13805 https://x.com/burny_tech/status/1968601645018742843 >"Our key insight is that transformers can learn to infer governing dynamics from context, enabling a single model to simulate fluid-solid interactions, shock waves, thermal convection, and multi-phase dynamics without being told the underlying equations" >"GPhyT achieves three critical breakthroughs: (1) superior performance across multiple physics domains, outperforming specialized architectures by up to 29x, (2) zero-shot generalization to new, unseen physical systems (e.g., new boundary conditions, entirely new physics) by inferring the dynamics from the input alone through in-context learning, and (3) stable long-term predictions through 50-timestep rollouts" "With our novel AI methods, we presented the first systematic discovery of new families of unstable singularities across three different fluid equations." https://deepmind.google/discover/blog/discovering-new-solutions-to-century-old-problems-in-fluid-dynamics/ A physics-informed graph neural network conserving linear and angular momentum for dynamical systems "In this paper, we propose DYNAMI-CAL GRAPHNET, a Physics-Informed Graph Neural Network that integrates the learning capabilities of GNNs with physics-based inductive biases to address these limitations. DYNAMI-CAL GRAPHNET enforces pairwise conservation of linear and angular momentum for interacting nodes using edge-local reference frames that are equivariant to rotational symmetries, invariant to translations, and equivariant to node permutations." https://www.nature.com/articles/s41467-025-67802-5 https://x.com/i/status/2012134176678465951 https://openai.com/index/new-result-theoretical-physics/ https://x.com/getjonwithit/status/2023600575565152336/ https://arxiv.org/abs/2602.14853 https://arxiv.org/abs/2510.10713 https://www.youtube.com/watch?v=6tKNwIhHk9s Terrence Tao is one of the best geniuses of our age I love his nuanced perspectives on everything, including AI, and how it relates to math and science So much intelligence juice in him https://www.youtube.com/watch?v=Q8Fkpi18QXU https://www.arenaphysica.com/publications/rf-studio Foundation Model for electromagnetism https://youtu.be/Q8Fkpi18QXU?is=IawpEawv6Am_Icqp Transformer Neural Networks for Identifying Boosted Higgs Bosons decaying into and in ATLAS https://atlas.web.cern.ch/Atlas/GROUPS/PHYSICS/PUBNOTES/ATL-PHYS-PUB-2023-021/