CORDIS - EU research results
CORDIS

INFORMATION-THEORETIC LIMITS FOR DEEP NEURAL NETWORKS

Project description

Towards a greater understanding of deep neural networks

Deep neural networks (DNN) are based on a web of multiple units bound together by mathematical functions that allow for learning. In recent decades, these algorithms have assisted in computer vision, speech and audio recognition and natural language processing. A common criticism though remains that deep-learning algorithms are often used as black box, which is unsatisfactory in all applications for which performance guarantees are critical. The EU-funded IT-DNN project aims to enhance the understanding of DNNs. This will be done by developing novel information-theoretic bounds on the generalisation error attainable using DNN and by demonstrating how such bounds can guide the design of such networks.

Objective

Over the last decade, deep-learning algorithms have dramatically improved the state of the art in many machine-learning problems, including computer vision, speech recognition, natural language processing, and audio recognition. Despite their success, however, there is no satisfactory mathematical theory that explains the functioning of such algorithms. Indeed, a common critique is that deep-learning algorithms are often used as black box, which is unsatisfactory in all applications for which performance guarantees are critical (e.g. traffic-safety applications).

The purpose of this project is to increase our theoretical understanding of deep neural networks (DNN). This will be done by developing novel information-theoretic bounds on the generalization error attainable using DNN and by demonstrating how such bounds can guide the design of such network.

Coordinator

CHALMERS TEKNISKA HOGSKOLA AB
Net EU contribution
€ 203 852,16
Address
-
412 96 GOTEBORG
Sweden

See on map

Region
Södra Sverige Västsverige Västra Götalands län
Activity type
Higher or Secondary Education Establishments
Links
Total cost
€ 203 852,16