Dynamische neuronale Netze und das fundamentale raumzeitliche Lernproblem

By Jürgen Schmidhuber. 1990

Citations in the corpus

• Deep learning in neural networks: An overview (2014, methodology): “While the RNN methods above were designed for DL of temporal sequences, the Neural Heat Exchanger (Schmidhuber, 1990c) consists of two parallel deep FNNs with opposite flow directions.”
• PowerPlay: Training an Increasingly General Problem Solver by Continually Searching for the Simplest Still Unsolvable Problem (2011, methodology): “They can compute any function computable by a standard PC (Schmidhuber, 1990).”
• Self-Delimiting Neural Networks (2012, background): “There is a trivial way of seeing this [51]: a traditional microprocessor can be modeled as a very sparsely connected RNN consisting of simple neurons implementing nonlinear AND and NAND gates.”
• Formal Theory of Creativity, Fun, and Intrinsic Motivation (1990–2010) (2010, methodology): “Early work [67], [71] describes a predictor based on an adaptive world model implemented as a recurrent neural network (RNN) (in principle a rather powerful computationa l device, even by today’s machine learning standards), predi cting sensory inputs including reward signals from the entire previou”
• Maximizing Fun by Creating Data with Easily Reducible Subjective Complexity (2013, methodology): “…without restrictive traditional assumptions (Schmidhuber 1991d)) based on an adaptive recurrent neural network as a predictive data encoder (Schmidhuber 1990b) is used to maximize intrinsic reward for the creator in proportion to the model's prediction errors (1990) (Schmidhuber 1990a, 1991c).”
• Adaptive confidence and adaptive curiosity (1991, background): “In [10], [6] and [7] the following basic idea foròn-line state space exploration by implementing dynamic curiosity and boredom' has been formulated: Spend reinforcement for a model-building controller whenever there is a mismatch between the expectations of the adaptive world model and reality.”