Policy manifold search: exploring the manifold hypothesis for diversity-based neuroevolution
Open accessThe DOI matches its Crossref record; 46 of the 46 references it cites were found in a registry or an index.
Abstract
Neuroevolution is an alternative to gradient-based optimisation that has the potential to avoid local minima and allows parallelisation. The main limiting factor is that usually it does not scale well with parameter space dimensionality. Inspired by recent work examining neural network intrinsic dimension and loss landscapes, we hypothesise that there exists a low-dimensional manifold, embedded in the policy network parameter space, around which a high-density of diverse and useful policies are located. This paper proposes a novel method for diversity-based policy search via Neuroevolution, that leverages learned representations of the policy network parameters, by performing policy search in this learned representation space. Our method relies on the Quality-Diversity (QD) framework which provides a principled approach to policy search, and maintains a collection of diverse policies, used as a dataset for learning policy representations. Further, we use the Jacobian of the inverse-mapping function to guide the search in the representation space. This ensures that the generated samples remain in the high-density regions, after mapping back to the original space. Finally, we evaluate our contributions on four continuous-control tasks in simulated environments, and compare to diversity-based baselines.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“there are a variety of off-the-shelf techniques for generating diverse policies 70,71”
From Learning few-shot imitation as cultural transmission (2023) · cites this paper as [70]
In the citing paper
“Stable Diffusionfoot_3 is a publicly available latent diffusion model [95] that supports CLIP-guided [91] text-to-image synthesis.”
From Language Model Crossover: Variation through Few-Shot Prompting (2024) · cites this paper as [95]
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Reducing the dimensionality of data with neural networks
Hinton, G. E., & Salakhutdinov, R. R. (2006) · Science · doi:10.1126/science.1127647 · Cited by 20,928
(2010) · Cited by 14,134
Representation Learning: A Review and New Perspectives
Bengio, Y., Courville, A., & Vincent, P. (2013) · IEEE Transactions on Pattern Analysis and Machine Intelligence · doi:10.1109/tpami.2013.50 · Cited by 13,001 · cited 2×
“This notion has been thoroughly explored under representation learning [3], and applied in RL [1, 8, 15], with important insights on how to exploit the structure of the manifold to improve an algorithm’s robustness [38, 39].”
BackgroundProximal Policy Optimization Algorithms
Schulman, J., Wolski, F., Dhariwal, P., Radford, A., & Klimov, O. (2017) · arXiv · doi:10.48550/arxiv.1707.06347 · Cited by 10,625 · cited 2×
“The architecture of the policy network is kept the same across all experiments, and is based on the Proximal Policy Optimisation [41] policy implementation for continuous control tasks.”
BackgroundMethodNonlinear principal component analysis using autoassociative neural networks
Kramer, M. A. (1991) · AIChE Journal · doi:10.1002/aic.690370209 · Cited by 3,037 · cited 1×
“Then, we use the parameters that have been added to the policy collection to train a dimensionality reduction algorithm, like a deep AutoEncoder (AE) [26, 28].”
MethodContractive Auto-Encoders: Explicit Invariance During Feature Extraction
Rifai, S., Vincent, P., Muller, X., Glorot, X., & Bengio, Y. (2011) · International Conference on Machine Learning · Cited by 1,603
The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networks
Frankle, J., & Carbin, M. (2018) · arXiv · doi:10.48550/arxiv.1803.03635 · Cited by 1,529 · cited 2×
“formance as the original network, when trained in isolation [16, 32].”
BackgroundEvolution Strategies as a Scalable Alternative to Reinforcement Learning
Salimans, T., Ho, J., Chen, X., Sidor, S., & Sutskever, I. (2017) · arXiv · doi:10.48550/arxiv.1703.03864 · Cited by 1,027 · cited 6×
“In this paper, we propose a novel approach, Policy Manifold Search (PoMS), that learns a mapping to a manifold in the policy network parameter space, which is used for Neuroevolution-based policy search.”
BackgroundInfluentialRobots that can adapt like animals
Cully, A., Clune, J., Tarapore, D., & Mouret, J.-B. (2015) · Nature · doi:10.1038/nature14422 · Cited by 979 · cited 5×
“Each iteration potentially adds new policies to the collection, following the MAP-Elites framework [11], which are then used to refine the latent representation in the next iteration.”
BackgroundMethodInfluentialSelf-citationPredicting Parameters in Deep Learning
Denil, M., Shakibi, B., Dinh, L., Ranzato, M., & de Freitas, N. (2013) · arXiv · doi:10.48550/arxiv.1306.0543 · Cited by 853 · cited 1×
“on the existence of an intrinsic dimension of the network parameter space, that is dependant on the particular task’s complexity and is sufficient for solving such a task [13, 19, 30, 37, 48].”
Background
Research works printed but not linked: 3
- Erin Catto . 2011. Box2d: A 2d physics engine for games. URL: http://www.box2d.org ( 2011 ). Erin Catto. 2011. Box2d: A 2d physics engine for games. URL: http://www.box2d.org (2011).
- Erwin Coumans and Yunfei Bai . 2016. Pybullet , a python module for physics simulation for games, robotics and machine learning. URL: http://pybullet.org ( 2016 ). Erwin Coumans and Yunfei Bai. 2016. Pybullet, a python module for physics simulation for games, robotics and machine learning. URL: http://pybullet.org (2016).
- David Ha , Andrew Dai , and Quoc V Le . 2017 . HyperNetworks. In International Conference on Learning Representations. David Ha, Andrew Dai, and Quoc V Le. 2017. HyperNetworks. In International Conference on Learning Representations.
Cited by
Evolutionary Reinforcement Learning: A Survey
Bai, H., Cheng, R., & Jin, Y. (2023) · Intelligent Computing · doi:10.34133/icomputing.0025 · Cited by 136
“Therefore, a number of automated BCs methods have been proposed to improve exploration efficiency, such as using dimensionality reduction methods to autonomously learn BCs [162, 163], or mapping the high-dimensional parameter space into a low-dimensional manifold in which a high-density of good policies is located [16…”
MethodLanguage Model Crossover: Variation through Few-Shot Prompting
Meyerson, E., Nelson, M. J., Bradley, H., Gaier, A., Moradi, A., Hoover, A. K., & Lehman, J. (2024) · ACM Transactions on Evolutionary Learning and Optimization · doi:10.1145/3694791 · Cited by 75
“The LMX operator in this paper can be seen similarly as building a probabilistic model of individuals (here of parents, rather than the whole population), and doing so implicitly in the forward-pass of the LLM (through in-context learning).”
Language model crossover: Variation through few-shot prompting
Meyerson, E., Nelson, M. J., Bradley, H., Gaier, A., Moradi, A., Hoover, A. K., & Lehman, J. (2023) · arXiv · doi:10.48550/arxiv.2302.12170 · Cited by 75
“The LMX operator in this paper can be seen similarly as building a probabilistic model of individuals (here of parents, rather than the whole population), and doing so implicitly in the forward-pass of the LLM (through in-context learning).”
Approximating gradients for differentiable quality diversity in reinforcement learning
Tjanaka, B., Fontaine, M. C., Togelius, J., & Nikolaidis, S. (2022) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3512290.3528705 · Cited by 74
Covariance Matrix Adaptation MAP-Annealing
Fontaine, M. C., & Nikolaidis, S. (2023) · Genetic and Evolutionary Computation Conference (GECCO) · doi:10.1145/3583131.3590389 · Cited by 68
“Later work [66], connected the elite hypervolume hypothesis to the”
BackgroundMethod
Metadata
- Authors
- Rakicevic, N., Cully, A., & Kormushev, P.
- Affiliations
- Imperial College London (GB)
- Venue
- Proceedings of the Genetic and Evolutionary Computation Conference
- Publisher
- ACM
- Published
- 2021-06-26
- Pages
- 901-909
- OpenAlex
- W3145430300
- Cited by
- 32
- References
- 46
- Citations by year
- 2026: 4 · 2024: 6 · 2023: 10 · 2022: 10 · 2021: 1
- FWCI
- 3.12
- Type
- conference paper
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Genetic and Evolutionary Computation Conference (GECCO) (publishedVersion)
- arXiv (Cornell University) (submittedVersion)
- Spiral (Imperial College London) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Engineering and Physical Sciences Research Council (EP/V006673/1 project REcoVER)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Evolutionary Algorithms and Applications
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Physical Sciences › Computer Science › Artificial Intelligence › Metaheuristic Optimization Algorithms Research
- Keywords
- neuroevolution, policy diversity, inverse mapping, continuous control, loss landscape, intrinsic dimension
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Policy manifold search: exploring the manifold hypothesis for diversity-based neuroevolution. https://app.cochat.ai/papers/10.1145/3449639.3459320
- The paper
- Rakicevic, N., Cully, A., & Kormushev, P. (2021). Policy manifold search: exploring the manifold hypothesis for diversity-based neuroevolution. Proceedings of the Genetic and Evolutionary Computation Conference. https://doi.org/10.1145/3449639.3459320
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.