Our work on tracking time evolution of stochastic reaction networks by neural network is @NatMachIntell, https://t.co/iwTxbrcwGd, free-read link https://t.co/1GQ8qKJsDS It shows again the great potential of neural network to solve stochastic dynamics in many fields.

Code: https://t.co/j4IonVurzw We will continuously update it, aiming to establish it as a standard alternative to the Gillespie algorithm, with higher efficiency for rare events.

We push the frontier of neural-network methods for solving chemical master equations, tackling large biochemical networks and spatially extended systems with rare events. https://t.co/FRs43Xx1OC

Our new paper: "The geometry and dynamics of annealed optimization in the coherent Ising machine with hidden and planted solutions" https://t.co/mWIURYUQwS How do algorithms like gradient descent negotiate the complex geometry of high dimensional loss landscapes to find near

“Everyone knows” what an autoencoder is… but there's an important complementary picture missing from most introductory material. In short: we emphasize how autoencoders are implemented—but not always what they represent (and some of the implications of that representation).🧵

Flow matching has become a dominant paradigm in generative modeling. We develop Quantum Flow Matching, with diverse applications: generate states with target magnetization and entanglement entropy; probe nonequilibrium free energy and superdiffusion. https://t.co/heKjAvyfLr

📣 We are hiring! Want to move to beautiful Stockholm 🇸🇪 and work on cutting-edge ML research? Join our group and help push the frontiers of machine learning! https://t.co/ir8qWYfkYV 📍 Apply now / spread the word! #ML #AI #Postdoc #Nordita #KTH #Stockholm

Autoregressive model: alphabets, actions, and atoms https://t.co/UQvHvzcq0M This lecture tried to offer a unified perspective to LLM, RL, and atomistic modeling!

🚨 1 week left for registering for the Les Houches workshop "Towards a theory for typical-case algorithmic hardness" 🌐 https://t.co/OVM3nsRVAj Also a good occasion to remind the important role that Les Houches played in the history of Statistical Physics & Computer Science 🤓👇

📣 We are organizing a winter school on "Physics of ML & ML for Physics" from Jan 13-24 at Nordita in Stockholm, quite in line with this year's Nobel Prize! Application deadline is Nov 10: https://t.co/5FMWU7gflc Help me to spread the words! #ML #AI #NobelPrize #winterschool

This work is more about opening a problem: even for the simplest noise-induced transition in bistable potential, the prevailing methods SINDy, FORCE are found inadequate. Leveraging Reservoir Computing is one way out, motivating to extend the other methods https://t.co/ks1R1jSB0E

CrystalFormer generates crystalline materials with an autoregressive transformer paper: https://t.co/KpRP5BLpUE codes: https://t.co/zVAzILMx22

The tensor-network message-passing method reduces the error in the calculation of local observables by several orders of magnitude compared with state-of-the-art techniques https://t.co/qIfcjXbPt3

🚨Deadline approaching soon🚨 Apply for our four-year, fully funded PhD programme at @Sissaschool before Feb 23rd; details below 👇 or via DM !

The exploration of the neural-network approach to uncover dynamical phase transition of nonequilibrium statistical mechanics is out with some updates:

Can large language models (LLMs) like ChatGPT help advance quantum computing? Yes! In a paper released today, PI's @rgmelko and visiting fellow @carrasqu describe how the same algorithmic structure used in LLMs might do just that! Full article:

Great work by Aleksandra Walczak and Thierry Mora's group. Glad that our previous work on how dynamics of signaling transmit information helps motivate more theoretical studies on this topic.

It is achieved by focusing on tuning a single hyperpara controlling the time scale of reservoir dynamics. It applies to white/colored noise, nonequilibrium dynamics, and experimental data of protein conformational transition, which helps reduce the need for entensive measurement.

Noise is usually regarded adversarial and mitigated for learning effective dynamics from time series. However, it can have functional roles of driving transitions between stable states. We leverage Reservoir Computing to predict noise-induced transitions.

We wondered how sampling with flow-based, diffusion-based or autoregressive networks compares to basic Monte Carlo or Langevin sampling. This is what came out for two classes of probability distributions well understood in statistical physics: https://t.co/qFTz13xLn4

We also take the chance to publish the code: https://t.co/P6XQtFhctp Dear Gillespie users, try it out and give us feedbacks. We will keep updating it to be efficient for any chemical reactions, maybe by asking #chatgpt4 for help😂