ALEPH · Autonomous learning agents in Physics
Horizon Europe — Marie Skłodowska-Curie Actions
- Duration
- 2023-03-01 → 2025-02-28
- EU contribution
- €183,601
- Participants
- 1
- Scheme
- HORIZON-TMA-MSCA-PF-EF
Lines connect the coordinator with its partners.
Results in brief
Autonomous learning agents in Physics
It is now widely understood that machine learning (ML) holds enormous potential across almost every area of society. One particularly promising field is the application of these models to the study of Sciences, and Physics in particular. With these machines, we can create better predictions when analyzing experimental data or, for instance, perform improved simulations of physical systems, as needed in areas like weather forecasting. However, ML holds an even greater potential: the ability to discover new Physics directly from experimental data. This means that, given access to experimental observations—or even direct control over the experiment—the machine could learn to understand and describe the underlying physical processes, connecting them to our existing theories. To achieve this, we must develop models that are not only good at predicting outcomes but also capable of providing insights in a language that humans can understand. The main objective of the project “Active Learning Agents in Physics” (ALEPH) was to develop autonomous learning agents for Physics, with a particular focus on their interpretability. This refers to designing learning machines that are not only powerful and robust in the tasks they are trained for but that can also assist in the process of scientific discovery. In ALEPH, I aimed to link scientific discovery to the interpretability of ML models. A model is considered interpretable if we can easily understand why and how a given prediction was made. Thus, if an ML model can perform accurate predictions about a physical process, it implies that important information about the process is encoded within the model. If the model is interpretable, we can access this information and use it to deepen our understanding of the physics behind the studied system.
Data: CORDIS, © European Union
Project objective
In recent years, the use of machine learning (ML) for the study of physics has experienced a strong boost. However, most of the machines used are black boxes, and the causal relation between inputs and outputs is often impossible to extract. Nonetheless, a critical aspect when dealing with physical systems is not only to make correct predictions, but to understand the physical laws which underlie these assessments. Recently, an increasing number of works aim at developing interpretable ML methods, from which such hidden laws can be extracted. However, their application to physics has been often limited to supervised and unsupervised learning approaches.The aim of this project is: 1) construct an interpretable reinforcement learning method; 2) extract hidden rules and features in timely and paramount problems in physics. The method combines three well-established concepts of ML: projective simulation, graph neural networks (GNN) and hidden variable disentanglement. PS provides interpretable RL agents that can be trained for a variety of tasks, from the construction of quantum experiments, via skill acquisition in robotics, to the modelling of honeybee colonies. By enhancing their learning power and interpretability with GNNs and variable disentanglement, we will extract the hidden features of the systems the RL agents have interacted with and ultimately, the physical laws governing them. In particular, we will tackle problems in the field of condensed matter, where particles diffuse either passively or actively, to reach a target state. Moreover, we will consider ensembles of RL agents, so as to analyze not only the physical properties of the systems, but also their interactions and communication dynamics in the quest of a common target.The originality of the proposal is directly related to: 1) the methods that will be developed; 2) the systems of study; 3) most importantly, the information we will access and discover with the interpretable RL agents.
Original text from CORDIS.
Participants
- UNIVERSITAET INNSBRUCK · InnsbruckCoordinatorAustria
Links
- View on CORDIS
- DOI: 10.3030/101063794
- https://ec.europa.eu/research/participants/documents/downloadPublic?documentIds=080166e517d13b67&appId=PPGMS
- https://ec.europa.eu/research/participants/documents/downloadPublic?documentIds=080166e5ff7be482&appId=PPGMS
Data: CORDIS, © European Union