This Tutorial presents Denario, an AI multiagent system that can assist in the full scientific process, and provides examples and guidance for its application.

This Roadmap envisions how to prioritize research in hardware-based machine learning methods and more to adapt to challenging real-time data collection environments in the forthcoming era of particle physics experiments.

Comprehensive benchmarking demonstrates the power and robustness of two convolutional backflow architectures for strongly correlated systems.

An approach that uses prediction-error methodology to train universal differential equations can learn unknown states in chaotic systems by removing the sensitive dependence of conventional methods on parameter choices.

Transformer models trained for protein structure prediction are analyzed from a statistical mechanics perspective, leading to a better understanding of how they work and why they can efficiently retrieve spatial and chemical relationships between amino acids.

AI Foundation Models trained on collider data can translate to cosmology.

A denoising network that reports a calibrated uncertainty alongside every data point allows 12-second photoemission scans to yield superconducting gap values with meaningful error bars, turning deep learning from a visualization aid into a quantitative measurement tool.

Several stress tests are utilized to assess the accuracy of two widely used machine-learned potential approaches.

A Transformer-based neural quantum state unifies spin and fermionic degrees of freedom in a single variational framework, achieving high accuracy in one dimension and extending naturally to challenging two-dimensional systems.

Analysis of the transition probabilities between large language model generated states shows that the dynamics of LLMs tend toward detailed balance.

A nonperturbative approach to Landau-level mixing in fractional quantum Hall liquids achieves competitive accuracy with high computational efficiency.

A tailored AI workflow learns latent patterns in tensor-network simulated snapshots of strongly correlated electrons in the 2D Hubbard model, enabling estimates of temperature and electron density with potential applications to ultracold atom simulators.

Neural simulation-based inference makes possible the identification and differentiation of hypotheses for how cell-scale organization emerges from molecular factors.

Machine learning models developed for core physics tasks, such as particle reconstruction, can double as foundation models.

A general framework which introduces the entropy production for generative autoregressive models establishes a bridge between these models and stochastic thermodynamics.

Physics-informed neural networks enable numerical simulations of relativistic spin hydrodynamics with strict angular-momentum conservation, which is essential for studying the rotational viscous effect that drives the mutual conversion between spin and orbital angular momentum.

A validation-driven agentic framework is demonstrated that enables reliable workflow generation for atomistic simulations on high performance computing systems.