SubsidieMeestersDeepLearning 2.0: Meta-Learning Qualitatively New Components
Develop meta-learning methods to create customized deep learning pipelines that enhance accuracy, reduce training time, and improve usability across various applications.
Projectdetails
Introduction
Deep learning has revolutionized many fields, such as computer vision, speech recognition, natural language processing, and reinforcement learning. This success is based on replacing domain-specific hand-crafted features with features that are learned for the particular task at hand.
Objective
The logical step to take deep learning to the next level is to also (meta-)learn other hand-crafted elements of the deep learning pipeline. We therefore propose to develop meta-level learning methods for the creation of novel customized deep learning pipelines, by means of:
- Hierarchical neural architecture search for learning qualitatively new architectures and architectural building blocks from scratch.
- Learning of optimizers and hyperparameter adaptation policies that adapt to their context in order to converge faster and more robustly.
- Learning the data to train on, to remove the need for large sets of labeled data.
- Bootstrapping from prior design efforts to increase efficiency and make an integrative design of architectures, optimizers, hyperparameter adaptation policies, and pretraining tasks feasible in practice.
Expected Outcomes
These advances will allow the next generation of deep learning pipelines to achieve higher accuracy, lower training time, and improved ease-of-use (democratization of deep learning). They will also allow customization to particular design contexts, including additional objectives next to accuracy, such as:
- Robustness
- Memory requirements
- Energy consumption
- Latency
- Interpretability
- Training cost
- Uncertainty estimation
- Algorithmic fairness
This customization aims to facilitate trustworthy AI.
Demonstration
In order to demonstrate the effectiveness of these methods, we plan to develop:
- New state-of-the-art customized deep learning pipelines for various applications, including EEG decoding, RNA folding, and improving the reinforcement learning pipeline and deep learning on tabular data.
Financiële details & Tijdlijn
Financiële details
| Subsidiebedrag | € 2.000.000 |
| Totale projectbegroting | € 2.000.000 |
Tijdlijn
| Startdatum | 1-5-2022 |
| Einddatum | 30-4-2027 |
| Subsidiejaar | 2022 |
Partners & Locaties
Projectpartners
- ALBERT-LUDWIGS-UNIVERSITAET FREIBURGpenvoerder
Land(en)
Vergelijkbare projecten binnen European Research Council
| Project | Regeling | Bedrag | Jaar | Actie |
|---|---|---|---|---|
MANUNKIND: Determinants and Dynamics of Collaborative ExploitationThis project aims to develop a game theoretic framework to analyze the psychological and strategic dynamics of collaborative exploitation, informing policies to combat modern slavery. | ERC STG | € 1.497.749 | 2022 | Details |
Elucidating the phenotypic convergence of proliferation reduction under growth-induced pressureThe UnderPressure project aims to investigate how mechanical constraints from 3D crowding affect cell proliferation and signaling in various organisms, with potential applications in reducing cancer chemoresistance. | ERC STG | € 1.498.280 | 2022 | Details |
Uncovering the mechanisms of action of an antiviral bacteriumThis project aims to uncover the mechanisms behind Wolbachia's antiviral protection in insects and develop tools for studying symbiont gene function. | ERC STG | € 1.500.000 | 2023 | Details |
The Ethics of Loneliness and SociabilityThis project aims to develop a normative theory of loneliness by analyzing ethical responsibilities of individuals and societies to prevent and alleviate loneliness, establishing a new philosophical sub-field. | ERC STG | € 1.025.860 | 2023 | Details |
MANUNKIND: Determinants and Dynamics of Collaborative Exploitation
This project aims to develop a game theoretic framework to analyze the psychological and strategic dynamics of collaborative exploitation, informing policies to combat modern slavery.
Elucidating the phenotypic convergence of proliferation reduction under growth-induced pressure
The UnderPressure project aims to investigate how mechanical constraints from 3D crowding affect cell proliferation and signaling in various organisms, with potential applications in reducing cancer chemoresistance.
Uncovering the mechanisms of action of an antiviral bacterium
This project aims to uncover the mechanisms behind Wolbachia's antiviral protection in insects and develop tools for studying symbiont gene function.
The Ethics of Loneliness and Sociability
This project aims to develop a normative theory of loneliness by analyzing ethical responsibilities of individuals and societies to prevent and alleviate loneliness, establishing a new philosophical sub-field.
Vergelijkbare projecten uit andere regelingen
| Project | Regeling | Bedrag | Jaar | Actie |
|---|---|---|---|---|
Next-Generation Natural Language GenerationThis project aims to enhance natural language generation by integrating neural models with symbolic representations for better control, adaptability, and reliable evaluation across various applications. | ERC STG | € 1.420.375 | 2022 | Details |
Algorithmic Bias Control in Deep learningThe project aims to develop a theory of algorithmic bias in deep learning to improve training efficiency and generalization performance for real-world applications. | ERC STG | € 1.500.000 | 2022 | Details |
Dynamics-Aware Theory of Deep LearningThis project aims to create a robust theoretical framework for deep learning, enhancing understanding and practical tools to improve model performance and reduce complexity in various applications. | ERC STG | € 1.498.410 | 2022 | Details |
Understanding Deep LearningThe project aims to establish a solid theoretical foundation for deep learning by investigating optimization, statistical complexity, and representation, enhancing understanding and algorithm development. | ERC STG | € 1.499.750 | 2022 | Details |
Next-Generation Natural Language Generation
This project aims to enhance natural language generation by integrating neural models with symbolic representations for better control, adaptability, and reliable evaluation across various applications.
Algorithmic Bias Control in Deep learning
The project aims to develop a theory of algorithmic bias in deep learning to improve training efficiency and generalization performance for real-world applications.
Dynamics-Aware Theory of Deep Learning
This project aims to create a robust theoretical framework for deep learning, enhancing understanding and practical tools to improve model performance and reduce complexity in various applications.
Understanding Deep Learning
The project aims to establish a solid theoretical foundation for deep learning by investigating optimization, statistical complexity, and representation, enhancing understanding and algorithm development.