{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,1,5]],"date-time":"2026-01-05T18:28:47Z","timestamp":1767637727864,"version":"3.48.0"},"reference-count":62,"publisher":"Maximum Academic Press","license":[{"start":{"date-parts":[[2021,4,16]],"date-time":"2021-04-16T00:00:00Z","timestamp":1618531200000},"content-version":"unspecified","delay-in-days":105,"URL":"https:\/\/www.cambridge.org\/core\/terms"}],"content-domain":{"domain":["cambridge.org"],"crossmark-restriction":true},"short-container-title":["The Knowledge Engineering Review"],"published-print":{"date-parts":[[2021]]},"abstract":"<jats:title>Abstract<\/jats:title>\n                  <jats:p>\n                    We propose a fully distributed actor-critic architecture, named diffusion-distributed-actor-critic\n                    <jats:italic>Diff-DAC<\/jats:italic>\n                    , with application to multitask reinforcement learning (MRL). During the learning process, agents communicate their value and policy parameters to their neighbours, diffusing the information across a network of agents with no need for a central station. Each agent can only access data from its local task, but aims to learn a common policy that performs well for the whole set of tasks. The architecture is scalable, since the computational and communication cost per agent depends on the number of neighbours rather than the overall number of agents. We derive Diff-DAC from duality theory and provide novel insights into the actor-critic framework, showing that it is actually an instance of the dual-ascent method. We prove almost sure convergence of Diff-DAC to a common policy under general assumptions that hold even for deep neural network approximations. For more restrictive assumptions, we also prove that this common policy is a stationary point of an approximation of the original problem. Numerical results on multitask extensions of common continuous control benchmarks demonstrate that Diff-DAC stabilises learning and has a regularising effect that induces higher performance and better generalisation properties than previous architectures.\n                  <\/jats:p>","DOI":"10.1017\/s0269888921000023","type":"journal-article","created":{"date-parts":[[2021,4,16]],"date-time":"2021-04-16T09:43:13Z","timestamp":1618566193000},"update-policy":"https:\/\/doi.org\/10.1017\/policypage","source":"Crossref","is-referenced-by-count":2,"title":["Fully distributed actor-critic architecture for multitask deep reinforcement learning"],"prefix":"10.48130","volume":"36","author":[{"ORCID":"https:\/\/orcid.org\/0000-0003-4551-5625","authenticated-orcid":false,"given":"Sergio","family":"Valcarcel Macua","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Ian","family":"Davies","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Aleksi","family":"Tukiainen","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Enrique Munoz","family":"de Cote","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"27968","published-online":{"date-parts":[[2021,4,16]]},"reference":[{"year":"2009","author":"Bertsekas","key":"S0269888921000023_ref5"},{"key":"S0269888921000023_ref35","unstructured":"Parisotto, E. , Ba, J. L. & Salakhutdinov, R. 2016. Actor-mimic: deep multitask and transfer reinforcement learning. In Proceedings of the International Conference on Learning Representations (ICLR)."},{"key":"S0269888921000023_ref17","unstructured":"Espeholt, L. , Soyer, H. , Munos, R. , Simonyan, K. , Mnih, V. , Ward, T. , Doron, Y. , Firoiu, V. , Harley, T. , Dunning, I. et al. 2018. Impala: scalable distributed deep-rl with importance weighted actor-learner architectures. In ICML."},{"first-page":"207","year":"2012","author":"Van Hasselt","key":"S0269888921000023_ref54"},{"year":"1990","author":"Horn","key":"S0269888921000023_ref22"},{"year":"2012","author":"Bertsekas","key":"S0269888921000023_ref6"},{"key":"S0269888921000023_ref56","doi-asserted-by":"crossref","unstructured":"Weinstein, A. & Littman, M. L. 2012. Bandit-based planning and learning in continuous-action markov decision processes. In International Conference on Automated Planning and Scheduling (ICAPS).","DOI":"10.1609\/icaps.v22i1.13507"},{"key":"S0269888921000023_ref44","first-page":"3802","article-title":"Almost sure convergence of two time-scale stochastic approximation algorithms","volume":"4","author":"Tadic","year":"2004","journal-title":"IEEE American Control Conference"},{"key":"S0269888921000023_ref48","unstructured":"Tomczak, M. B. , Valcarcel Macua, S. , de Cote, E. M. & Vrancx, P. 2019. Compatible features for monotonic policy improvement."},{"key":"S0269888921000023_ref23","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2013.2241057"},{"key":"S0269888921000023_ref46","unstructured":"Teh, Y. W. , Bapst, V. , Czarnecki, W. M. , Quan, J. , Kirkpatrick, J. , Hadsell, R. , Heess, N. & Pascanu, R. 2017. Distral: robust multitask reinforcement learning. arXiv preprint ."},{"year":"1958","author":"Arrow","key":"S0269888921000023_ref2"},{"key":"S0269888921000023_ref20","doi-asserted-by":"publisher","DOI":"10.1109\/TSMCC.2012.2218595"},{"key":"S0269888921000023_ref21","unstructured":"Heess, N. , Wayne, G. , Silver, D. , Lillicrap, T. , Erez, T. & Tassa, Y. 2015. Learning continuous control policies by stochastic value gradients. In Advances in Neural Information Processing Systems (NIPS), 2926\u20132934."},{"key":"S0269888921000023_ref60","unstructured":"Zhang, K. , Yang, Z. , Liu, H. , Zhang, T. & Basar, T. 2018. Fully decentralized multi-agent reinforcement learning with networked agents. In Proceedings International Conference on Machine Learning (ICML), 5872\u20135881."},{"key":"S0269888921000023_ref36","doi-asserted-by":"publisher","DOI":"10.1007\/s11768-011-0313-y"},{"key":"S0269888921000023_ref43","unstructured":"Sutton, R. S. , Mcallester, D. , Singh, S. & Mansour, Y. 1999. Policy gradient methods for reinforcement learning with function approximation. In Advances in Neural Information Processing Systems (NIPS), 1057\u20131063."},{"key":"S0269888921000023_ref59","doi-asserted-by":"publisher","DOI":"10.1287\/moor.2019.1037"},{"key":"S0269888921000023_ref62","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2014.2385046"},{"key":"S0269888921000023_ref58","doi-asserted-by":"publisher","DOI":"10.1007\/BF00992696"},{"key":"S0269888921000023_ref42","unstructured":"Sutton, R. S. , Maei, H. R. , Precup, D. , Bhatnagar, S. , Silver, D. , Szepesvari, C. & Wiewiora, E. 2009. Fast gradient-descent methods for temporal-difference learning with linear function approximation. In Proceedings of the International Conference on Machine Learning (ICML), 993\u20131000."},{"key":"S0269888921000023_ref12","unstructured":"Bou-Ammar, H. , Eaton, E. , Ruvolo, P. & Taylor, M. 2014. Online multi-task learning for policy gradient methods. In Proceedings of the International Conference on Machine Learning (ICML), 1206\u20131214."},{"key":"S0269888921000023_ref53","unstructured":"van der Meulen, R. 2015. Gartner says 6.4 billion connected \u2018things\u2019 will be in use in 2016, up 30 percent from 2015. http:\/\/www.gartner.com\/newsroom\/id\/3165317."},{"key":"S0269888921000023_ref50","unstructured":"Valcarcel Macua, S. 2017. Distributed Optimization, Control and Learning in Multiagent Networks. PhD thesis, Universidad Polit\u00e9cnica de Madrid."},{"year":"2005","author":"Puterman","key":"S0269888921000023_ref37"},{"key":"S0269888921000023_ref49","unstructured":"Tutunov, R. , Bou-Ammar, H. & Jadbabaie, A. 2016. An exact distributed newton method for reinforcement learning. In IEEE Conference on Decision and Control (CDC), 1003\u20131008."},{"key":"S0269888921000023_ref14","unstructured":"Brockman, G. , Cheung, V. , Pettersson, L. , Schneider, J. , Schulman, J. , Tang, J. & Zaremba, W. 2016. OpenAI Gym."},{"key":"S0269888921000023_ref11","doi-asserted-by":"publisher","DOI":"10.1137\/S0363012997331639"},{"key":"S0269888921000023_ref52","unstructured":"Valcarcel Macua, S. , Tukiainen, A. , Hern\u00e1ndez, D. G.-O. , Baldazo, D. , de Cote, E. M. & Zazo, S. 2017. Diff-DAC: distributed actor-critic for average multitask deep reinforcement learning. arXiv preprint ."},{"key":"S0269888921000023_ref28","doi-asserted-by":"publisher","DOI":"10.1016\/j.automatica.2016.12.014"},{"key":"S0269888921000023_ref25","unstructured":"Kingma, D. & Ba, J. L. 2015. Adam: a method for stochastic optimization. In Proceedings of the International Conference on Learning Representations (ICLR)."},{"key":"S0269888921000023_ref51","doi-asserted-by":"publisher","DOI":"10.1109\/TAC.2014.2368731"},{"key":"S0269888921000023_ref34","unstructured":"Packer, C. , Gao, K. , Kos, J. , Kr\u00e4henb\u00fchl, P. , Koltun, V. & Song, D. 2018. Assessing generalization in deep reinforcement learning."},{"key":"S0269888921000023_ref8","doi-asserted-by":"publisher","DOI":"10.1109\/TAC.2012.2209984"},{"key":"S0269888921000023_ref10","doi-asserted-by":"publisher","DOI":"10.1016\/S0167-6911(97)90015-3"},{"key":"S0269888921000023_ref3","unstructured":"Assran, M. , Romoff, J. , Ballas, N. , Pineau, J. & Rabbat, M. 2019. Gossip-based actor-learner architectures for deep reinforcement learning. In Advances in Neural Information Processing Systems (NIPS), 13320\u201313330."},{"key":"S0269888921000023_ref57","first-page":"949","article-title":"Natural evolution strategies","volume":"15","author":"Wierstra","year":"2014","journal-title":"Journal of Machine Learning Research"},{"first-page":"1022","year":"1999","author":"Ng","key":"S0269888921000023_ref33"},{"year":"1996","author":"Golub","key":"S0269888921000023_ref19"},{"first-page":"85","year":"1972","author":"Karp","key":"S0269888921000023_ref24"},{"key":"S0269888921000023_ref39","doi-asserted-by":"publisher","DOI":"10.1561\/2200000051"},{"key":"S0269888921000023_ref31","unstructured":"Mnih, V. , Badia, A. P. , Mirza, M. , Graves, A. , Lillicrap, T. , Harley, T. , Silver, D. & Kavukcuoglu, K. 2016. Asynchronous methods for deep reinforcement learning. In International Conference on Machine Learning, 1928\u20131937."},{"key":"S0269888921000023_ref38","doi-asserted-by":"publisher","DOI":"10.1287\/moor.2016.0821"},{"key":"S0269888921000023_ref7","doi-asserted-by":"publisher","DOI":"10.1016\/j.automatica.2009.07.008"},{"key":"S0269888921000023_ref45","first-page":"1633","article-title":"Transfer learning for reinforcement learning domains: a survey","volume":"10","author":"Taylor","year":"2009","journal-title":"Journal of Machine Learning Research"},{"year":"2008","author":"Borkar","key":"S0269888921000023_ref9"},{"key":"S0269888921000023_ref26","unstructured":"Kober, J. & Peters, J. R. 2009. Policy search for motor primitives in robotics. In Advances in Neural Information Processing Systems (NIPS), 849\u2013856."},{"key":"S0269888921000023_ref27","doi-asserted-by":"publisher","DOI":"10.1137\/S0363012901385691"},{"year":"2004","author":"Boyd","key":"S0269888921000023_ref13"},{"key":"S0269888921000023_ref16","unstructured":"El Bsat, S. , Bou-Ammar, H. & Taylor, M. E. 2017. Scalable multitask policy gradient reinforcement learning. In AAAI Conference on Artificial Intelligence (AAAI), 1847\u20131853."},{"key":"S0269888921000023_ref55","unstructured":"Wei, E. & Ozdaglar, A. 2012. Distributed alternating direction method of multipliers. In IEEE Annual Conf. Decision and Control (CDC), 5445\u20135450."},{"key":"S0269888921000023_ref47","unstructured":"Tieleman, T. & Hinton, G. 2012. Lecture 6.5- RMSProp: divide the gradient by a running average of its recent magnitude."},{"key":"S0269888921000023_ref4","unstructured":"Baird, L. C. III 1993. Advantage Updating, Technical report, Wright Lab Wright-Patterson AFB OH."},{"key":"S0269888921000023_ref61","doi-asserted-by":"publisher","DOI":"10.1109\/TSP.2012.2204985"},{"key":"S0269888921000023_ref18","unstructured":"Fu, J. , Levine, S. & Abbeel, P. 2016. One-shot learning of manipulation skills with online dynamics adaptation and neural network priors. In IEEE RSJ International Conference on Intelligent Robots and Systems (IROS), 4019\u20134026."},{"first-page":"66","year":"2008","author":"Melo","key":"S0269888921000023_ref30"},{"key":"S0269888921000023_ref32","unstructured":"Mnih, V. , Kavukcuoglu, K. , Silver, D. , Graves, A. , Antonoglou, I. , Wierstra, D. & Riedmiller, M. 2013. Playing atari with deep reinforcement learning. arXiv preprint ."},{"key":"S0269888921000023_ref40","unstructured":"Scherrer, B. 2010. Should one compute the temporal difference fix point or minimize the Bellman residual? The unified oblique projection view. In Proceedings of the International Conference on Machine Learning (ICML), 959\u2013966."},{"key":"S0269888921000023_ref15","doi-asserted-by":"publisher","DOI":"10.1109\/JSTSP.2013.2246763"},{"key":"S0269888921000023_ref29","unstructured":"Lillicrap, T. P. , Hunt, J. J. , Pritzel, A. , Heess, N. , Erez, T. , Tassa, Y. , Silver, D. & Wierstra, D. 2015. Continuous control with deep reinforcement learning."},{"key":"S0269888921000023_ref41","unstructured":"Schulman, J. , Moritz, P. , Levine, S. , Jordan, M. & Abbeel, P. 2015. High-dimensional continuous control using generalized advantage estimation. arXiv preprint ."},{"key":"S0269888921000023_ref1","unstructured":"Andreas, J. , Klein, D. & Levine, S. 2017. Modular multitask reinforcement learning with policy sketches. In Proceedings of the International Conference on Machine Learning (ICML), 166\u2013175."}],"container-title":["The Knowledge Engineering Review"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.cambridge.org\/core\/services\/aop-cambridge-core\/content\/view\/S0269888921000023","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,1,5]],"date-time":"2026-01-05T14:42:21Z","timestamp":1767624141000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.cambridge.org\/core\/product\/identifier\/S0269888921000023\/type\/journal_article"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2021]]},"references-count":62,"alternative-id":["S0269888921000023"],"URL":"https:\/\/doi.org\/10.1017\/s0269888921000023","relation":{},"ISSN":["0269-8889","1469-8005"],"issn-type":[{"type":"print","value":"0269-8889"},{"type":"electronic","value":"1469-8005"}],"subject":[],"published":{"date-parts":[[2021]]},"assertion":[{"value":"\u00a9 The Author(s), 2021. Published by Cambridge University Press","name":"copyright","label":"Copyright","group":{"name":"copyright_and_licensing","label":"Copyright and Licensing"}}],"article-number":"e6"}}