SubsidieMeestersContinual and Sequential Learning for Artificial Intelligence
Develop a Continual and Sequential Learning AI that integrates RL with advanced data gathering to autonomously adapt and explore in dynamic environments for scientific breakthroughs.
Projectdetails
Introduction
Machine Learning systems, while promising, lack the autonomy needed for many real-world applications beyond mere proof-of-concept stages. The key challenge lies in enabling AI to continuously adapt to shifting data distributions and proactively seek information under high uncertainty.
Application Fields
Fields such as drug discovery and micro-chemistry are expecting breakthroughs from AI, given the vast and intricate search spaces they deal with, coupled with expensive data acquisition. It is vital for algorithms to steer this search, assimilate new data, and strategically explore promising zones.
Reinforcement Learning Challenges
Reinforcement Learning (RL) offers tools and methods for agents to autonomously learn from their actions, but its efficacy has been largely confined to stationary, single-task settings.
Project Vision
ConSequentIAL's vision is a Continual and Sequential Learning AI that marries supervised and unsupervised learning with advanced data gathering and RL-driven discovery mechanisms.
Proposed Approach
To achieve these goals, I propose to bridge the theory of constrained and non-stationary RL to build a sound and useful mathematical formulation of the problem.
Algorithmic Development
On these new solid grounds, I develop novel algorithmic principles that allow the agent to:
- Detect and respond to external shifts.
- Remain aware of her own impact on the system she interacts with.
Memory and Stability Trade-off
I address the memory-versus-stability trade-off central to continual learning by enabling agents to actively plan their skill acquisition in accordance with their long-term goals.
Project Ambition
The ambition of this project is to position AI to tackle the consequential scientific challenges ahead.
Financiële details & Tijdlijn
Financiële details
| Subsidiebedrag | € 1.259.375 |
| Totale projectbegroting | € 1.259.375 |
Tijdlijn
| Startdatum | 1-5-2025 |
| Einddatum | 30-4-2030 |
| Subsidiejaar | 2025 |
Partners & Locaties
Projectpartners
- EBERHARD KARLS UNIVERSITAET TUEBINGENpenvoerder
Land(en)
Vergelijkbare projecten binnen European Research Council
| Project | Regeling | Bedrag | Jaar | Actie |
|---|---|---|---|---|
Deep Bayesian Reinforcement Learning -- Unifying Perception, Planning, and ControlDevelop an algorithmic framework using deep learning and Bayesian reinforcement learning to enhance robotic manipulation in unstructured environments by effectively managing uncertainty. | ERC Starting... | € 1.500.000 | 2022 | Details |
Control for Deep and Federated LearningCoDeFeL aims to enhance machine learning methods through control theory, developing efficient ResNet architectures and federated learning techniques for applications in digital medicine and recommendations. | ERC Advanced... | € 2.499.224 | 2024 | Details |
Koopman-Operator-based Reinforcement Learning Control of Partial Differential EquationsThis project aims to enhance reinforcement learning for large-scale engineering systems by developing performance-guaranteed controllers, addressing safety in energy-efficient technologies. | ERC Starting... | € 1.499.000 | 2025 | Details |
Model-based Reinforcement Learning for Versatile Robots in the Real WorldREAL-RL aims to create versatile autonomous robots that learn from experience using a model-based approach for efficient task adaptation and behavior planning. | ERC Consolid... | € 1.998.500 | 2023 | Details |
Data-Driven Verification and Learning Under UncertaintyThe DEUCE project aims to enhance reinforcement learning by developing novel verification methods that ensure safety and correctness in complex, uncertain environments through data-driven abstractions. | ERC Starting... | € 1.500.000 | 2023 | Details |
Deep Bayesian Reinforcement Learning -- Unifying Perception, Planning, and Control
Develop an algorithmic framework using deep learning and Bayesian reinforcement learning to enhance robotic manipulation in unstructured environments by effectively managing uncertainty.
Control for Deep and Federated Learning
CoDeFeL aims to enhance machine learning methods through control theory, developing efficient ResNet architectures and federated learning techniques for applications in digital medicine and recommendations.
Koopman-Operator-based Reinforcement Learning Control of Partial Differential Equations
This project aims to enhance reinforcement learning for large-scale engineering systems by developing performance-guaranteed controllers, addressing safety in energy-efficient technologies.
Model-based Reinforcement Learning for Versatile Robots in the Real World
REAL-RL aims to create versatile autonomous robots that learn from experience using a model-based approach for efficient task adaptation and behavior planning.
Data-Driven Verification and Learning Under Uncertainty
The DEUCE project aims to enhance reinforcement learning by developing novel verification methods that ensure safety and correctness in complex, uncertain environments through data-driven abstractions.