Augmentative Topology Agents For Open-Ended Learning
Not on the map. Why?
Abstract (arXiv)
In this work, we tackle the problem of open-ended learning by introducing a method that simultaneously evolves agents and increasingly challenging environments. Unlike previous open-ended approaches that optimize agents using a fixed neural network topology, we hypothesize that generalization can be improved by allowing agents' controllers to become more complex as they encounter more difficult environments. Our method, Augmentative Topology EPOET (ATEP), extends the Enhanced Paired Open-Ended Trailblazer (EPOET) algorithm by allowing agents to evolve their own neural network structures over time, adding complexity and capacity as necessary. Empirical results demonstrate that ATEP results in general agents capable of solving more environments than a fixed-topology baseline. We also investigate mechanisms for transferring agents between environments and find that a species-based approach further improves the performance and generalization of agents.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 41 of its 77 references that are in the map. First posted on arXiv 20 Oct 2022.
Builds on
41 of 77 references.
References · 41
- 2017Proximal Policy Optimization Algorithmscited by 14,441
- 2016Learning without Forgettingcited by 2,465
- 2016OpenAI Gymcited by 1,684
- 2016Progressive Neural Networkscited by 1,332
- 2018Continual Lifelong Learning with Neural Networks: A Reviewcited by 1,313
- 2018Diversity is All You Need: Learning Skills without a Reward Functioncited by 734
- 2020It's Not Just Size That Matters: Small Language Models Are Also Few-Shot Learnerscited by 668
- 2017Evolution Strategies as a Scalable Alternative to Reinforcement Learningcited by 555
- 2022MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledgecited by 469
- 2015Illuminating search spaces by mapping elitescited by 406
- 2019Dynamics-Aware Unsupervised Discovery of Skillscited by 308
- 2017Grammar Variational Autoencodercited by 283
29 more
- 2020Towards Continual Reinforcement Learning: A Review and Perspectivescited by 268
- 2019Continual learning with hypernetworkscited by 267
- 2020Emergent Complexity and Zero-shot Transfer via Unsupervised Environment Designcited by 260
- 2016CAD2RL: Real Single-Image Flight without a Single Real Imagecited by 227
- 2019Go-Explore: a New Approach for Hard-Exploration Problemscited by 207
- 2014Robots that can adapt like animalscited by 203
- 2020Prioritized Level Replaycited by 190
- 2019Paired Open-Ended Trailblazer (POET): Endlessly Generating Increasingly Complex and Diverse Learning Environments and Their Solutionscited by 182
- 2021Open-Ended Learning Leads to Generally Capable Agentscited by 177
- 2016Variational Lossy Autoencodercited by 175
- 2022Evolving Curricula with Regret-Based Environment Designcited by 166
- 2020Explore, Discover and Learn: Unsupervised Discovery of State-Covering Skillscited by 120
- 2020Enhanced POET: Open-Ended Reinforcement Learning through Unbounded Invention of Learning Challenges and their Solutionscited by 103
- 2019V-MPO: On-Policy Maximum a Posteriori Policy Optimization for Discrete and Continuous Controlcited by 85
- 2022Deep Surrogate Assisted Generation of Environmentscited by 40
- 2021Adversarial Reinforcement Learning for Procedural Content Generationcited by 36
- 2021Illuminating Diverse Neural Cellular Automata for Level Generationcited by 35
- 2016Neurogenesis Deep Learningcited by 32
- 2018Relative Entropy Regularized Policy Iterationcited by 32
- 2018Talakat: Bullet Hell Generation through Constrained Map-Elitescited by 26
- 2020CPPN2GAN: Combining Compositional Pattern Producing Networks and GANs for Large-scale Pattern Generationcited by 19
- 2022When, where, and how to add new neurons to ANNscited by 18
- 2020Co-generation of game levels and game-playing agentscited by 14
- 2022Open-Ended Evolution for Minecraft Building Generationcited by 9
- 2022Transfer Dynamics in Emergent Evolutionary Curriculacited by 6
- 2022Open-Ended Reinforcement Learning with Neural Reward Functionscited by 5
- 2021OPEn: An Open-ended Physics Environment for Learning Without a Taskcited by 4
- 2021Co-optimising Robot Morphology and Controller in a Simulated Open-Ended Environmentcited by 3
- 2022Open-Ended Learning Strategies for Learning Complex Locomotion Skillscited by 1
Built on by
Citing papers per year: 2023: 2; 2024: 1.
3 papers.