Recurrent Neural Networks and Related Machines That Learn Algorithms

Project Information

AlgoRNN

Grant agreement ID: 742870

DOI 10.3030/742870

Project closed

EC signature date 12 June 2017

Start date 1 October 2017

End date 31 March 2024

Funded under

• EXCELLENT SCIENCE - European Research Council (ERC)

Total cost € 2 500 000,00

EU contribution € 2 500 000,00

Coordinated by

UNIVERSITA DELLA SVIZZERA ITALIANA
Switzerland

Publications

Conference proceedings (29)

The Neural Data Router: Adaptive Control Flow in Transformers Improves Systematic Generalization

Author(s): R Csordás, K Irie, J Schmidhuber
Published in: ICLR 2022, 2022
Publisher: ICLR 2022

​Linear Transformers Are Secretly Fast Weight Programmers

Author(s): ​I. Schlag*, K. Irie*, J. Schmidhuber
Published in: ICML 2021, Proceedings of the 38th International Conference on Machine Learning, 2021
Publisher: ICML 2021

Parameter-based value functions

Author(s): F. Faccio, L. Kirsch, J. Schmidhuber
Published in: NeurIPS 2020 Workshop on Offline Reinforcement Learning, 2020
Publisher: NeurIPS 2020 Workshop on Offline Reinforcement Learning

​Going Beyond Linear Transformers with Recurrent Fast Weight Programmers

Author(s): ​K. Irie*, I. Schlag*, R. Csordás, J. Schmidhuber
Published in: NeurIPS 2021, 2021
Publisher: NeurIPS 2021

​Training and Generating Neural Networks in Compressed Weight Space

Author(s): ​K. Irie, J. Schmidhuber
Published in: ​ICLR 2021 Workshop on Neural Compression, 2021
Publisher: ​ICLR 2021 Workshop on Neural Compression

​A Modern Self-Referential Weight Matrix That Learns to Modify Itself

Author(s): ​K. Irie, I. Schlag, R. Csordás, J. Schmidhuber
Published in: ​​NeurIPS 2021 Workshop on Deep Reinforcement Learning, 2021
Publisher: ​​NeurIPS 2021 Workshop on Deep Reinforcement Learning

Reward-Weighted Regression Converges to a Global Optimum

Author(s): M. Štrupl*, F. Faccio*, D. R. Ashley, R. K. Srivastava, J. Schmidhuber
Published in: AAAI 2022, in press, 2022
Publisher: AAAI 2022

Learning Associative Inference Using Fast Weight Memory

Author(s): I. Schlag, T. Munkhdalai, J. Schmidhuber
Published in: ICLR 2021, 2021
Publisher: ICLR 2021

The Devil is in the Detail: Simple Tricks Improve Systematic Generalization of Transformers

Author(s): R Csordás, K Irie, J Schmidhuber
Published in: EMNLP 2021, Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, 2021, Page(s) 619–634
Publisher: EMNLP 2021

Bayesian Brains and the Rényi Divergence

Author(s): N. Sajid*, F. Faccio*, L. Da Costa, T. Parr, J. Schmidhuber, K. Friston
Published in: CNS*2021, 2021
Publisher: CNS*2021

Improving Stateful Premise Selection with Transformers

Author(s): K. Prorokovic, M. Wand, J. Schmidhuber
Published in: CICM 2021, 2021
Publisher: CICM 2021

​Improving Baselines in the Wild

Author(s): ​K. Irie, I. Schlag, R. Csordás, J. Schmidhuber
Published in: ​NeurIPS 2021 Workshop on Distribution Shifts, 2021
Publisher: ​NeurIPS 2021 Workshop on Distribution Shifts

Unsupervised Object Keypoint Learning using Local Spatial Predictability

Author(s): A. Gopalakrishnan, S. van Steenkiste, J. Schmidhuber
Published in: ICML 2020 workshop on Object-Oriented Learning: Perception, Representation and Reasoning, 2020
Publisher: ICML

Parameter-based value functions

Author(s): F. Faccio, L. Kirsch, J. Schmidhuber
Published in: ICLR 2021, 2021
Publisher: ICLR 2021

Reward-Weighted Regression Converges to a Global Optimum

Author(s): M. Štrupl*, F. Faccio*, D. R. Ashley, R. K. Srivastava, J. Schmidhuber
Published in: ICML 2021 Workshop on Reinforcement Learning Theory, 2021
Publisher: ICML 2021 Workshop on Reinforcement Learning Theory

Policy Optimization via Importance Sampling

Author(s): A. M. Metelli, M. Papini, F. Faccio, M. Restelli
Published in: NeurIPS 2018, 2018
Publisher: published at the Neural Information Processing Systems Conference

Improving Differentiable Neural Computers Through Memory Masking, De-allocation, and Link Distribution Sharpness Control

Author(s): R. Csordas, J. Schmidhuber
Published in: ICLR 2019, 2019
Publisher: Published as a conference paper at ICLR 2019

Modular Networks: Learning to Decompose Neural Computation

Author(s): L. Kirsch, J. Kunze, D. Barber
Published in: NeurIPS 2018, 2018
Publisher: published at the Neural Information Processing Systems Conference

Learning to Reason with Third Order Tensor Products

Author(s): I. Schlag, J. Schmidhuber
Published in: NeurIPS 2018, 2018
Publisher: published at the Neural Information Processing Systems Conference

Recurrent World Models Facilitate Policy Evolution

Author(s): D. Ha, J. Schmidhuber
Published in: NeurIPS 2018, 2018
Publisher: published at the Neural Information Processing Systems Conference

Enhancing the Transformer with Explicit Relational Encoding for Math Problem Solving

Author(s): I. Schlag, P. Smolensky, R. Fernandez, N. Jojic, J. Schmidhuber, J. Gao
Published in: Under review, 2020
Publisher: Not disclosed yet

Improving Generalization in Meta Reinforcement Learning using Neural Objectives

Author(s): L. Kirsch, S. van Steenkiste, J. Schmidhuber
Published in: ICLR 2020, 2020
Publisher: Published as a conference paper at ICLR 2020

Learning Adaptive Control Flow in Transformers for Improved Systematic Generalization

Author(s): R. Csordás, K. Irie, J. Schmidhuber
Published in: NeurIPS 2021 Workshop on Advances in Programming Languages and Neurosymbolic Systems, 2021
Publisher: NeurIPS 2021 Workshop on Advances in Programming Languages and Neurosymbolic Systems

Augmenting Classic Algorithms with Neural Components for Strong Generalisation on Ambiguous and High-Dimensional Data

Author(s): I. Schlag, J. Schmidhuber
Published in: NeurIPS 2021 Workshop on Advances in Programming Languages and Neurosymbolic Systems, 2021
Publisher: NeurIPS 2021 Workshop on Advances in Programming Languages and Neurosymbolic Systems

Parameter-based value functions

Author(s): F. Faccio, L. Kirsch, J. Schmidhuber
Published in: NeurIPS 2020 Workshop on Deep Reinforcement Learning, 2021
Publisher: NeurIPS 2020 Workshop on Deep Reinforcement Learning

Unsupervised Object Keypoint Learning using Local Spatial Predictability

Author(s): A. Gopalakrishnan, S. van Steenkiste, J. Schmidhuber
Published in: ICLR 2021, 2021
Publisher: ICLR 2021

Are Neural Nets Modular? Inspecting Functional Modularity Through Differentiable Weight Masks

Author(s): R Csordás, S van Steenkiste, J Schmidhuber
Published in: ICLR 2021, 2021
Publisher: ICLR 2021

Exploring through Random Curiosity with General Value Functions

Author(s): A. Ramesh, L. Kirsch, S. van Steenkiste, J. Schmidhuber
Published in: ​​NeurIPS 2021 Workshop on Deep Reinforcement Learning, 2021
Publisher: ​​NeurIPS 2021 Workshop on Deep Reinforcement Learning

Meta learning backpropagation and improving it

Author(s): L. Kirsch, J. Schmidhuber
Published in: NeurIPS 2021, 2021
Publisher: NeurIPS 2021

Peer reviewed articles (2)

Bayesian Brains and the Rényi Divergence

Author(s): N. Sajid*, F. Faccio*, L. Da Costa, T. Parr, J. Schmidhuber, K. Friston
Published in: Neural Computation, Issue 34, 829–855 (2022), 2021, ISSN 1530-888X
Publisher: Neural computation, Cambridge: MIT Press

Generative Adversarial Networks are Special Cases of Artificial Curiosity (1990) and also Closely Related to Predictability Minimization (1991)

Author(s): J. Schmidhuber
Published in: Neural Networks, 2020, ISSN 0893-6080
Publisher: Pergamon Press Ltd.

Other (3)

Automatic Embedding of Stories Into Collections of Independent Media

Author(s): D. R. Ashley*, V. Herrmann*, Z. Friggstad, K. W. Mathewson, J. Schmidhuber
Published in: ArXiv, 2021
Publisher: ArXiv

All You Need Is Supervised Learning: From Imitation Learning to Meta-RL With Upside Down RL

Author(s): K. Arulkumaran, D. R. Ashley, J. Schmidhuber, R. K. Srivastava
Published in: RLDM 2022, 2022
Publisher: RLDM 2022

Learning Relative Return Policies With Upside-Down Reinforcement Learning

Author(s): D. R. Ashley, K. Arulkumaran, J. Schmidhuber, R. K. Srivastava
Published in: RLDM 2022, 2022
Publisher: RLDM 2022