Experimental Analysis of the Real-time Recurrent Learning Algorithm

By Ronald J. Williams, David Zipser. 1989

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

Deep learning in neural networks: An overview (2014, neutral): “…1995; Gherrity, 1989; Kremer & Kolen, 2001; Pearlmutter, 1989, 1995; Robinson & Fallside, 1987; Rohwer, 1989; Schmidhuber, 1992a; Werbos, 1988;Williams, 1989;Williams & Peng, 1990;Williams & Zipser, 1988, 1989a, 1989b), also for equilibrium RNNs (Almeida, 1987; Pineda, 1987) with stationary inputs.”
Learning Simple Algorithms from Examples (2015, background): “The simple feed-forward controller generalizes perfectly on the copy, reverse and walk tasks but completely fails on the remaining ones, due to a lack of required memory † .”
Adapting Bias by Gradient Descent: An Incremental Version of Delta-Bar-Delta (1992, methodology): “A similar technique is used in gradient-descent analyses of recurrent connectionist networks (c.f., e.g., Williams & Zipser 1989).”
Learning Complex, Extended Sequences Using the Principle of History Compression (1992, unknown): “Pi denotes the ith level network, which is trained to predict its own next inputfrom its previous inputs.' We take Pi to be a conventional dynamic recurrent neural network (Robinson and Fallside 1987; Williams and Zipser 1989; Williams and Peng 1990); however, it might be some other adaptive seque”
Symbiotic Evolution of Neural Networks in Sequential Decision Tasks (1997, background): “In more complex architectures such as networks with recurrent connections, computing the gradient informationnecessary for hill-climbing is very costly (Williams and Zipser 1989).”
Gradient-based learning algorithms for recurrent networks and their computational complexity (1995, methodology): “…described here, both subgrouped RT'RL and truncated BPTT were tested for their ability to train fully recurrent networks to emulate the finite state machine part of a Turing machine for balancing parentheses, a task that had previously been shown to be learnable by RTRL (Williams & Zipser, 1989b).”
Mind's eye: A recurrent visual representation for image caption generation (2015, methodology): “For learning we use the Backpropagation Through Time (BPTT) algorithm [43].”
A Recurrent Neural Network that Learns to Count (1999, methodology): “It has also been shown that an RNN can implement FSM computations, such as push± pop transitions for a PDA (Sun et al., 1990) and the read/ write /shift transitions for a Turing machine (Cottrell & Tsung, 1993; Williams & Zipser, 1989).”
An Efficient Gradient-Based Algorithm for On-Line Training of Recurrent Network Trajectories (1990, methodology): “Using BPTT(16;8) gave an additional factor of 2 speedup: making BPTT(16;8) well over 50 times faster than RTRL on this task.”
Networks adjusting networks (1990, methodology): “(Immediate weight changes are also employed by Williams and Zipser 22] who tested another learning algorithm for fully recurrent networks -rst described by Robinson and Fallside 6] -which is local in time but not in space.)”
Learning Algorithms for Networks with Internal and External Feedback (1990, methodology): “Notation (the reader may nd it convenient to compare with (Williams and Zipser, 1989)): C is the set of all non-input units of the control network, A is the set of its output units, I is the set of its`normal' input units, P is the set of its pain and reinforcement units, M is the set of all units…”
Artificial Neural Networks-Based Machine Learning for Wireless Networks: A Tutorial (2017, methodology): “ed based on the complete training set, which may become time consuming if the size of the training set is very large. To overcome these drawbacks in BPTT training, real-time recurrent learning (RTRL) [88] can be used to compute the exact error gradient at every time step, which is suitable for onlin”
High-Order Graph Convolutional Recurrent Neural Network: A Deep Learning Framework for Network-Scale Traffic Learning and Forecasting (2018, background): “This kind of methods can be described as the real-time recurrent learning method [22].”
Thirty Years of Machine Learning: The Road to Pareto-Optimal Wireless Networks (2019, methodology): “ts “memory” inherited from previous layers for processing the future inputs as shown in Fig. 20 (b). Popular algorithms used for training the RNN include the real time recurrent learning technique of [238], the causal recursive backpropagation algorithm of [239], the backpropagation through time alg”
Gradient-based learning algorithms for recurrent connectionist networks (1990, methodology): “This algorithm has beenindependently derived in various forms by Robinson and Fallside (1987), Kuhn (1987), Bachrach(1988), Mozer (1988), and Williams and Zipser (1989a), and continuous-time versions have beenproposed by Gherrity (1989) and by Doya and Yoshizawa (1989).5.1 The AlgorithmFor each k 2…”