mlsanityData to 7 Oct 2026

Exploring the Promise and Limits of Real-Time Recurrent Learning

arXiv 2305.19044 ↗First posted 30 May 2023cs.LGcited by 19

Not on the map. Why?

Abstract (arXiv)

Real-time recurrent learning (RTRL) for sequence-processing recurrent neural networks (RNNs) offers certain conceptual advantages over backpropagation through time (BPTT). RTRL requires neither caching past activations nor truncating context, and enables online learning. However, RTRL's time and space complexity make it impractical. To overcome this problem, most recent work on RTRL focuses on approximation theories, while experiments are often limited to diagnostic settings. Here we explore the practical promise of RTRL in more realistic settings. We study actor-critic methods that combine RTRL and policy gradients, and test them in several subsets of DMLab-30, ProcGen, and Atari-2600 environments. On DMLab memory tasks, our system trained on fewer than 1.2 B environmental frames is competitive with or outperforms well-known IMPALA and R2D2 baselines trained on 10 B frames. To scale to such challenging tasks, we focus on certain well-known neural architectures with element-wise recurrence, allowing for tractable RTRL without approximation. Importantly, we also discuss rarely addressed limitations of RTRL in real-world applications, such as its complexity in the multi-layer case.

Key facts. Cited by 19 papers in the map (data through 7 Oct 2026). Builds on 28 of its 64 references that are in the map. First posted on arXiv 30 May 2023.

Builds on

28 of 64 references.

References · 28

  1. 2016Asynchronous Methods for Deep Reinforcement Learningcited by 2,218
  2. 2020Transformers are RNNs: Fast Autoregressive Transformers with Linear Attentioncited by 1,979
  3. 2018IMPALA: Scalable Distributed Deep-RL with Importance Weighted Actor-Learner Architecturescited by 866
  4. 2019Leveraging Procedural Generation to Benchmark Reinforcement Learningcited by 478
  5. 2021Linear Transformers Are Secretly Fast Weight Programmerscited by 471
  6. 2015Deep Recurrent Q-Learning for Partially Observable MDPscited by 457
  7. 2023Resurrecting Recurrent Neural Networks for Long Sequencescited by 397
  8. 2022Diagonal State Spaces are as Effective as Structured State Spacescited by 350
  9. 2015LSTM: A Search Space Odysseycited by 315
  10. 2019Stabilizing Transformers for Reinforcement Learningcited by 262
  11. 2018Independently Recurrent Neural Network (IndRNN): Building A Longer and Deeper RNNcited by 142
  12. 2018Multi-task Deep Reinforcement Learning with PopArtcited by 138
16 more
  1. 2016Quasi-Recurrent Neural Networkscited by 87
  2. 2021Going Beyond Linear Transformers with Recurrent Fast Weight Programmerscited by 84
  3. 2019Towards Interpretable Reinforcement Learning Using Attention Augmented Agentscited by 82
  4. 2020A Formal Hierarchy of RNN Architecturescited by 61
  5. 2017Simple Recurrent Units for Highly Parallelizable Recurrencecited by 60
  6. 2017Dynamic Evaluation of Neural Sequence Modelscited by 53
  7. 2017Unbiased Online Recurrent Optimizationcited by 46
  8. 2019TorchBeast: A PyTorch Platform for Distributed RLcited by 41
  9. 2022History Compression via Language Models in Reinforcement Learningcited by 39
  10. 2018Approximating Real-Time Recurrent Learning with Random Kronecker Factorscited by 29
  11. 2023Online learning of long-range dependenciescited by 25
  12. 2019Optimal Kronecker-Sum Approximation of Real Time Recurrent Learningcited by 21
  13. 2019Biologically inspired alternatives to backpropagation through time for learning in recurrent neural netscited by 20
  14. 2023Scalable Real-Time Recurrent Learning Using Columnar-Constructive Networkscited by 11
  15. 2019IndyLSTMs: Independently Recurrent LSTMscited by 3
  16. 2021Scalable Online Recurrent Learning Using Columnar Neural Networkscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 3; 2024: 8; 2025: 5; 2026: 3.

19 papers.

Sort:

  1. 2024Streaming Deep Reinforcement Learning Finally Workscited by 24
  2. 2023Practical Computational Power of Linear Transformers and Their Recurrent and Self-Referential Extensionscited by 20
  3. 2024On the Resurgence of Recurrent Models for Long Sequences -- Survey and Research Opportunities in the Transformer Eracited by 17
  4. 2024Real-Time Recurrent Learning using Trace Units in Reinforcement Learningcited by 10
  5. 2023Real-Time Recurrent Reinforcement Learningcited by 6
  6. 2025Fast weight programming and linear transformers: from machine learning to neurobiologycited by 6
  7. 2024Recurrent Reinforcement Learning with Memoroidscited by 5
  8. 2023Estimating Post-Synaptic Effects for Online Training of Feed-Forward SNNscited by 2
  9. 2024A Unified Framework for Neural Computation and Learning Over Timecited by 1
  10. 2026Streaming Reinforcement Learning under Partial Observability with Real-Time Recurrent Learningcited by 1
  11. 2024Nature-Inspired Local Propagationcited by 0
  12. 2024ODEStream: A Buffer-Free Online Learning Framework with ODE-based Adaptor for Streaming Time Series Forecastingcited by 0
7 more
  1. 2024Recurrent Joint Embedding Predictive Architecture with Recurrent Forward Propagation Learningcited by 0
  2. 2025Convergence Analysis of Real-time Recurrent Learning (RTRL) for a class of Recurrent Neural Networkscited by 0
  3. 2025A Scalable Hybrid Training Approach for Recurrent Spiking Neural Networkscited by 0
  4. 2025Who invented deep residual learning?cited by 0
  5. 2025Generalising E-prop to Deep Networkscited by 0
  6. 2026Learning Long-Range Dependencies with Temporal Predictive Codingcited by 0
  7. 2026Massive Redundancy in Gradient Transport Enables Sparse Online Learningcited by 0