Simple Principles of Metalearning

By J. Urgen Schmidhuber, Jieyu Zhao, Marco A. Wiering. 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

Learning to reinforcement learn (2016, background): “…for meeting these challenges, which we refer to as deep meta-reinforcement learning , a label that is intended to both link it with and distinguish it from previous work employing the term “meta-reinforcement learning” (e.g. Schmidhuber et al., 1996; Schweighofer and Doya, 2003, discussed later).”
Machine Theory of Mind (2018, methodology): “We consider the challenge of building a Theory of Mind as essentially a meta-learning problem (Schmidhuber et al., 1996; Thrun & Pratt, 1998; Hochreiter et al., 2001; Vilalta & Drissi, 2002).”
Optimal Ordered Problem Solver (2002, neutral): “A heuristic attempt ( Schmidhuber, Zhao, & Wiering, 1996, 1997b ) to greatly reduce the constants through experience was called Adaptive LSEARCH or ALS—compare related ideas by Solomonoff (1986, 1989).”
Generalizing from a Few Examples (2019, methodology): “ch as neural activity encoding [84]. For example, in image classification, this relationship can be annotated by human, mined from text corpus or extracted from lexical database [133]. •Meta-learning [100] or learning-to-learn [47] that improves performance P on taskT by data set of the taskT and th”
Shifting Inductive Bias with Success-Story Algorithm, Adaptive Levin Search, and Incremental Self-Improvement (1997, background): “Schmidhuber et al. (1996) describe two agents A and B living in a partially observable600 × 500 pixel environment with obstacles.”
Goedel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements (2003, methodology): “In previous work we used the success-story algorithm (SSA) to force some (stochastic) SMP to trigger better and better self-modifications [32, 44, 43, 45].”
Transformers are Meta-Reinforcement Learners (2022, unknown): “Meta-Learning is an established Machine Learning (ML) principle to learn inductive biases from the distribution of tasks to produce a data-efficient learning system (Bengio et al., 1991; Schmidhuber et al., 1996; Thrun & Pratt, 1998).”
Reinforcement Learning, Fast and Slow. (2019, background): “Meta-learning occurs when one learning system progressively adjusts the operation of a second learning system, such that the latter operates with increasing speed and efficiency [83, 84, 85, 86, 87].”
Learning Gradient Descent: Better Generalization and Longer Horizons (2017, background): “From a reinforcement learning perspective, the optimizer can be viewed as a policy which takes the current state as input and output the next action (Schmidhuber et al., 1999).”
Learning to Learn: Hierarchical Meta-Critic Networks (2019, background): “One of sources of such biases has been explored in the machine learning literature under the rubric of meta learning [5], [6], which also has various alternative names, including life-long learning, learning to learn, etc.”
Meta-learning by the Baldwin effect (2018, background): “ey perform poorly when only a small amount of data is available. Meta-learning aims to mitigate this limitation by broadening the learner’s scope from a single task to a distribution of related tasks [25, 26]. The goal of metalearning is then to learn a learning strategy that generalizes to similar ”
Learning to Learn Using Gradient Descent (2001, background): “Meta-learning is known in the reinforcement learning framework [ 11 ,12].”
What Can Learned Intrinsic Rewards Capture? (2019, methodology): “Meta-learning for Exploration Meta-learning (Schmidhuber et al., 1996; Thrun & Pratt, 1998) has recently received considerable attention in RL. Recent advances include few-shot adaptation (Finn et al., 2017a), few-shot imitation (Finn et al., 2017b; Duan et al., 2017), model adaptation (Clavera et a”
Gödel Machines: Fully Self-referential Optimal Universal Self-improvers (2007, methodology): “In previous work we used the success-story algorithm (SSA) to force some (stochastic) SMPs to trigger better and better self-modifications [35, 49, 48, 50].”
Been There, Done That: Meta-Learning with Episodic Recall (2018, background): “First introduced as a core topic in AI research in the 1990s (Thrun & Pratt, 1998; Schmidhuber et al., 1996), meta-learning has recently resurged as a front-line agenda item (Santoro et al., 2016; Andrychowicz et al., 2016; Vinyals et al., 2016).”
Multi-agent deep reinforcement learning: a survey (2021, background): “This learning-to-learn approach is known as meta-learning (Finn and Levine 2018; Schmidhuber et al. 1996).”
Discovery of Useful Questions as Auxiliary Tasks (2019, background): “A meta-learner progressively improves the learning process of a learner (Schmidhuber et al., 1996; Thrun & Pratt, 1998) that is attempting to solve some task.”
What learning algorithm is in-context learning? Investigations with linear models (2022, background): “Such “meta-learning” approaches typically treat learning as a bi-level optimization problem (Schmidhuber et al., 1996; Andrychowicz et al., 2016; Finn et al., 2017): they define “inner” and “outer” models and learning procedures, then train an outer model to set parameters for an inner procedure (e.”