TU Darmstadt / ULB / TUprints

Evolutionary training and abstraction yields algorithmic generalization of neural computers

Tanneberg, Daniel ; Rueckert, Elmar ; Peters, Jan (2023)
Evolutionary training and abstraction yields algorithmic generalization of neural computers.
In: Nature Machine Intelligence, 2020, 2 (12)
doi: 10.26083/tuprints-00020535
Article, Secondary publication, Postprint

[img] Text
Copyright Information: In Copyright.

Download (8MB)
Item Type: Article
Type of entry: Secondary publication
Title: Evolutionary training and abstraction yields algorithmic generalization of neural computers
Language: English
Date: 17 October 2023
Place of Publication: Darmstadt
Year of primary publication: 16 November 2020
Place of primary publication: London
Publisher: Springer
Journal or Publication Title: Nature Machine Intelligence
Volume of the journal: 2
Issue Number: 12
Collation: 14, v Seiten
DOI: 10.26083/tuprints-00020535
Corresponding Links:
Origin: Secondary publication service

A key feature of intelligent behaviour is the ability to learn abstract strategies that scale and transfer to unfamiliar problems. An abstract strategy solves every sample from a problem class, no matter its representation or complexity—similar to algorithms in computer science. Neural networks are powerful models for processing sensory data, discovering hidden patterns and learning complex functions, but they struggle to learn such iterative, sequential or hierarchical algorithmic strategies. Extending neural networks with external memories has increased their capacities to learn such strategies, but they are still prone to data variations, struggle to learn scalable and transferable solutions, and require massive training data. We present the neural Harvard computer, a memory-augmented network-based architecture that employs abstraction by decoupling algorithmic operations from data manipulations, realized by splitting the information flow and separated modules. This abstraction mechanism and evolutionary training enable the learning of robust and scalable algorithmic solutions. On a diverse set of 11 algorithms with varying complexities, we show that the neural Harvard computer reliably learns algorithmic solutions with strong generalization and abstraction, achieves perfect generalization and scaling to arbitrary task configurations and complexities far beyond seen during training, and independence of the data representation and the task domain.

Status: Postprint
URN: urn:nbn:de:tuda-tuprints-205359
Classification DDC: 000 Generalities, computers, information > 004 Computer science
Divisions: 20 Department of Computer Science > Intelligent Autonomous Systems
TU-Projects: EC/H2020|640554|SKILLS4ROBOTS
Date Deposited: 17 Oct 2023 11:31
Last Modified: 23 Oct 2023 09:24
URI: https://tuprints.ulb.tu-darmstadt.de/id/eprint/20535
PPN: 512617023
Actions (login required)
View Item View Item