Incremental self-improvement for life-time multi-agent reinforcement learning

By Jian Zhao, J. Urgen Schmidhuber. 1996

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

Citations in the corpus

Supermasks in Superposition (2020, background): “In continual learning, a model aims to solve a number of tasks sequentially [47, 56] without catastrophic forgetting [10, 23, 33].”
BatchEnsemble: An Alternative Approach to Efficient Ensemble and Lifelong Learning (2020, methodology): “In lifelong learning, the model trains on a number of tasks in a sequential (online) order, without access to entire previous tasks’ data (Thrun, 1998; Zhao & Schmidhuber, 1996).”
Online Meta-Learning (2019, neutral): “Continual learning: Our problem setting is related to (but distinct from) continual, or lifelong learning (Thrun, 1998; Zhao & Schmidhuber, 1996).”
Meta-Learning Improves Lifelong Relation Extraction (2019, background): “Lifelong Learning In the lifelong learning setting, also referred to as continual learning (Ring, 1994; Thrun, 1996; Zhao and Schmidhuber, 1996), a model f θ is presented with a sequence of tasks {T t } t =1 , 2 , 3 ..,T , one task per round, and the goal is to learn model parameters { θ t } t =1 ,…”
Cooperative Multi-Agent Learning: The State of the Art (2005, methodology): “This idea was used in early multi-agent learning literature [228, 229, 312].”
A GENERAL METHOD FOR MULTI-AGENT REINFORCEMENT LEARNING IN UNRESTRICTED ENVIRONMENTS (1996, background): “See (Schmidhuber, 1995; Zhao and Schmidhuber, 1996; Wiering and Schmidhuber, 1996) for more recent work along these lines.”
Shifting Inductive Bias with Success-Story Algorithm, Adaptive Levin Search, and Incremental Self-Improvement (1997, methodology): “It is easy to show (Schmidhuber, 1994, 1996) that the current SSA call will have enforced the following “success-story criterion” SSC (t is thet in the most recent stepSSA.2): R(t) t R(t)−R(t1) t− t1 R(t)−R(t2) t− t2 . . .”
Probabilistic Incremental Program Evolution (1997, methodology): “PIPE is partly inspired by recent work on reinforcement learning with probabilistic programming languages (Schmidhuber, 1994; Wiering & Schmidhuber, 1996b; Zhao & Schmidhuber, 1996; Schmidhuber, Zhao, & Schraudolph, 1997a; Schmidhuber, Zhao, & Wiering, 1997b).”
Simple Principles of Metalearning (1996, background): “The current section exploits the fact that it is also possible to use many, much simpler actions that can be combined to form more complex learning strategies, or metalearning strategies (Schmidhuber, 1994, 1996; Zhao and Schmidhuber, 1996).”
Planning and Acting in Partially Observable Stochastic Domains (1998, methodology): “Several techniques have been proposed recently for searching for good program-like controllers in pomdps [44, 21] We restrict our attention to the simpler nite-horizon case and a small set of in nite-horizon problems that have optimal nite-state plans.”
Reinforcement Learning with Self-Modifying Policies (1998, background): “This does include certain zero-sum games though [42].”