Filters








21,149 Hits in 6.7 sec

Quantifying the effects of environment and population diversity in multi-agent reinforcement learning [article]

Kevin R. McKee and Joel Z. Leibo and Charlie Beattie and Richard Everett
2022 arXiv   pre-print
In this paper, we investigate and quantify the relationship between generalization and diversity in the multi-agent domain.  ...  Generalization is a major challenge for multi-agent reinforcement learning. How well does an agent perform when placed in novel environments and in interactions with new co-players?  ...  We are also indebted to Mary Cassin for designing and creating the sprite art for the DeepMind Lab2D implementation of Overcooked.  ... 
arXiv:2102.08370v2 fatcat:s2akt5hcbrhepdwefolq73yd4u

Effective Diversity in Population Based Reinforcement Learning [article]

Jack Parker-Holder and Aldo Pacchiano and Krzysztof Choromanski and Stephen Roberts
2020 arXiv   pre-print
Exploration is a key problem in reinforcement learning, since agents can only learn from data they acquire in the environment.  ...  With that in mind, maintaining a population of agents is an attractive method, as it allows data be collected with a diverse set of behaviors.  ...  The authors want to thank anonymous reviewers for constructive feedback which helped improve the paper.  ... 
arXiv:2002.00632v3 fatcat:5ayehk64rjd4tijxtuzwl4gh7e

Towards Learning Multi-agent Negotiations via Self-Play [article]

Yichuan Charlie Tang
2020 arXiv   pre-print
In contrast, deep reinforcement learning (Deep RL) has been very effective at finding policies by simultaneously exploring, interacting, and learning from environments.  ...  Leveraging the powerful Deep RL paradigm, we demonstrate that an iterative procedure of self-play can create progressively more diverse environments, leading to the learning of sophisticated and robust  ...  Acknowledgements We thank Barry Theobald, Hanlin Goh, Ruslan Salakhutdinov, Jian Zhang, Nitish Srivastava, Alex Druinsky, and the anonymous reviewers for making this a better manuscript.  ... 
arXiv:2001.10208v1 fatcat:dp4ucamn5bektnd36ombdmxyfy

Towards Learning Multi-Agent Negotiations via Self-Play

Yichuan Tang
2019 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW)  
In contrast, deep reinforcement learning (Deep RL) has been very effective at finding policies by simultaneously exploring, interacting, and learning from environments.  ...  Leveraging the powerful Deep RL paradigm, we demonstrate that an iterative procedure of self-play can create progressively more diverse environments, leading to the learning of sophisticated and robust  ...  Acknowledgements We thank Barry Theobald, Hanlin Goh, Ruslan Salakhutdinov, Jian Zhang, Nitish Srivastava, Alex Druinsky, and the anonymous reviewers for making this a better manuscript.  ... 
doi:10.1109/iccvw.2019.00297 dblp:conf/iccvw/Tang19 fatcat:5vyk4zkwcrdozmk3cfn75ttzpu

Giving Up Control: Neurons as Reinforcement Learning Agents [article]

Jordan Ott
2020 arXiv   pre-print
At the same time, they must cooperate, so the population and organism can perform high-level functions. To this end, we introduce modeling neurons as reinforcement learning agents.  ...  Populations of neurons must compete with their neighbors for resources, inhibition, and activity representation.  ...  This can be framed as neurons in the population composing the environment of other neurons. Figure 3 depicts the multi-agent network.  ... 
arXiv:2003.11642v1 fatcat:5qzfkdcys5hstd4cvnj6upy6pi

Co-Evolution of Predator-Prey Ecosystems by Reinforcement Learning Agents

Jeongho Park, Juwon Lee, Taehwan Kim, Inkyung Ahn, Jooyoung Park
2021 Entropy  
Here we frame the co-evolution of predators and preys in an ecosystem as allowing agents to learn and evolve toward better ones in a manner appropriate for multi-agent reinforcement learning.  ...  The problem of finding adequate population models in ecology is important for understanding essential aspects of their dynamic nature.  ...  Conflicts of Interest: The authors declare no conflict of interest.  ... 
doi:10.3390/e23040461 pmid:33924723 pmcid:PMC8069842 fatcat:44qbdgtbfnamfpy4bhyw4wby24

Pick Your Battles: Interaction Graphs as Population-Level Objectives for Strategic Diversity [article]

Marta Garnelo, Wojciech Marian Czarnecki, Siqi Liu, Dhruva Tirumala, Junhyuk Oh, Gauthier Gidel, Hado van Hasselt, David Balduzzi
2021 arXiv   pre-print
We provide evidence for the importance of diversity in multi-agent training and analyse the effect of applying different interaction graphs on the training trajectories, diversity and performance of populations  ...  In this paper we study how to construct diverse populations of agents by carefully structuring how individuals within a population interact.  ...  Also related to our line of work is research on diversity in reinforcement learning, which is often studied in the context of intrinsic motivation and skill discovery [5, 8, 11, 13] .  ... 
arXiv:2110.04041v1 fatcat:fc54thv25jdtbakzfuxtqvt4de

Measuring Diversity in Populations Employing Cultural Learning in Dynamic Environments [chapter]

Dara Curran, Colm O'Riordan
2005 Lecture Notes in Computer Science  
We compare the genotypic and phenotypic diversity of populations employing only population learning and of populations using both population and cultural learning in two types of dynamic environment: one  ...  This paper examines the effect of cultural learning on a population of neural networks.  ...  Acknowledgements This research is funded by the Irish Research Council for Science, Engineering and Technology.  ... 
doi:10.1007/11553090_39 fatcat:frgrojpucvfpfpdl7uhydj33oq

Emergent Coordination Through Competition [article]

Siqi Liu, Guy Lever, Josh Merel, Saran Tunyasuvunakool, Nicolas Heess, Thore Graepel
2019 arXiv   pre-print
We study the emergence of cooperative behaviors in reinforcement learning agents by introducing a challenging competitive multi-agent soccer environment with continuous simulated physics.  ...  Our study highlights several of the challenges encountered in large scale multi-agent training in continuous control.  ...  In any such game each agent in the population effectively treats the other agents as part of their environment and learns a policy π θ to optimize their expected return, averaged over such games.  ... 
arXiv:1902.07151v2 fatcat:dbhxgyk5lzb4zhhoflvjt642sy

Diverse Behavior Is What Game AI Needs: Generating Varied Human-Like Playing Styles Using Evolutionary Multi-Objective Deep Reinforcement Learning [article]

Ruimin Shen, Yan Zheng, Jianye Hao, Yinfeng Chen, Changjie Fan
2020 arXiv   pre-print
This demonstrates the effectiveness of EMO-DRL in learning complex style automatically.  ...  In the beginning, EMO-DRL initializes and maintains a population of candidates (black dots).  ... 
arXiv:1910.09022v6 fatcat:aolmaskxsberdfyofof2j4z3hu

Dynamics-aware novelty search with behavior repulsion

Kang Xu, Yan Ma, Wei Li
2022 Proceedings of the Genetic and Evolutionary Computation Conference  
Searching solutions for the task with sparse or deceptive rewards is a fundamental problem in Evolutionary Algorithms (EA) and Reinforcement Learning (RL).  ...  The novelty of a single solution is defined as the prediction error of an approximate dynamic model in the task-agnostic behavior space.  ...  As a promising policy search method in recent years, Reinforcement Learning agents can typically only learn from experience acquired in the environment, making training policy with deceptive or sparse  ... 
doi:10.1145/3512290.3528761 fatcat:ocwbnxzqana4xazfctsr6gezbm

Diverse Auto-Curriculum is Critical for Successful Real-World Multiagent Learning Systems [article]

Yaodong Yang, Jun Luo, Ying Wen, Oliver Slumbers, Daniel Graves, Haitham Bou Ammar, Jun Wang, Matthew E. Taylor
2021 arXiv   pre-print
Multiagent reinforcement learning (MARL) has achieved a remarkable amount of success in solving various types of video games.  ...  A cornerstone of this success is the auto-curriculum framework, which shapes the learning process by continually creating new challenging tasks for agents to adapt to, thereby facilitating the acquisition  ...  Challenges include the lack of an accurate simulator [13] , the high cost of environmental interaction, and the difficulty in learning both effective and diverse policies.  ... 
arXiv:2102.07659v2 fatcat:nlijvc7yizflvdz5iqxibw4tfa

Policy-focused Agent-based Modeling using RL Behavioral Models [article]

Osonde A. Osoba, Raffaele Vardavas, Justin Grana, Rushil Zutshi, Amber Jaycocks
2020 arXiv   pre-print
This paper examines the value of reinforcement learning (RL) models as adaptive, high-performing, and behaviorally-valid models of agent decision-making in ABMs.  ...  We run some analytic experiments on our AI-equipped ABMs e.g. explorations of the effects of behavioral heterogeneity in a population and the emergence of synchronization in a population.  ...  A good ABM can enable policy analysts and decision-makers to explore the potential macro-effects of diverse policy interventions on a population of real-world agents.  ... 
arXiv:2006.05048v3 fatcat:2xdlvcxtajfu5ekpj2zhqzmnai

An Introduction to Multi-Agent Reinforcement Learning and Review of its Application to Autonomous Mobility [article]

Lukas M. Schmidt, Johanna Brosig, Axel Plinge, Bjoern M. Eskofier, Christopher Mutschler
2022 arXiv   pre-print
Recent advances in behavioral planning use Reinforcement Learning to find effective and performant behavior strategies.  ...  Multi-Agent Reinforcement Learning (MARL) is a research field that aims to find optimal solutions for multiple agents that interact with each other.  ...  MULTI-AGENT REINFORCEMENT LEARNING In MARL, multiple agents are concurrently optimized to find optimal policies.  ... 
arXiv:2203.07676v2 fatcat:jedez77btbehzmlsahfl6x56wm

Embodied intelligence via learning and evolution

Agrim Gupta, Silvio Savarese, Surya Ganguli, Li Fei-Fei
2021 Nature Communications  
The intertwined processes of learning and evolution in complex environmental niches have resulted in a remarkable diversity of morphological forms.  ...  Here, we introduce Deep Evolutionary Reinforcement Learning (DERL): a computational framework which can evolve diverse agent morphologies to learn challenging locomotion and manipulation tasks in complex  ...  In addition, the repository also includes code for creating UNIMALS, generating evolutionary environments, and evaluation task suite.  ... 
doi:10.1038/s41467-021-25874-z pmid:34615862 fatcat:x32gdnis45b2zbia7zvaxc5ahi
« Previous Showing results 1 — 15 out of 21,149 results