A Learning Algorithm for Continually Running Fully Recurrent Neural Networks

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.

In Model Architectures

Citations in the corpus

A Deep Reinforced Model for Abstractive Summarization (2017, methodology): “The most widely used method to train a decoder RNN for sequence generation, called the teacher forcing” algorithm (Williams and Zipser, 1989), minimizes a maximum-likelihood loss at each decoding step.”
Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer (2019, methodology): “Specifically, the model is trained with a maximum likelihood objective (using “teacher forcing” [Williams and Zipser, 1989]) regardless of the task.”
ColdGANs: Taming Language GANs with Cautious Sampling Strategies (2020, methodology): “Under the most popular paradigm, sequence to sequence models [41] are trained with Maximum Likelihood Estimation (MLE) via Teacher Forcing [51].”
Discriminative Adversarial Search for Abstractive Summarization (2020, neutral): “Inspired by Venkatra-man et al. (2015), Bengio et al. (2015) proposed a variation of Teacher Forcing wherein the ground truth tokens are incrementally replaced by the predicted words.”
Learning Energy-Based Spatial-Temporal Generative ConvNets for Dynamic Patterns (2019, background): “For temporal data, a popular model is the recurrent neural network [27], [49].”
The Reversible Residual Network: Backpropagation Without Storing Activations (2017, methodology): “Martens and Sutskever [22] adopted this technique in the context of training recurrent neural networks on a sequence of length T using backpropagation through time [33], storing every d √ T e layers and recomputing the intermediate activations between each during the backward pass.”
Learning to Reason Deductively: Math Word Problem Solving as Complex Relation Extraction (2022, methodology): “, 2015), we adopt the teacherforcing strategy (Williams and Zipser, 1989) to guide the model with gold expressions during training.”
UNICORN on RAINBOW: A Universal Commonsense Reasoning Model on a New Multitask Benchmark (2021, methodology): “Following Raffel et al. (2019), we converted all tasks to text-to-text and used teacher forcing (Williams and Zipser 1989) as the training objective, with greedy decoding for predictions.”
On the difficulty of training recurrent neural networks (2012, methodology): “ train properly. A plethora of training algorithms have been proposed in the literature, like Backpropagation Through Time (BPTT) (Rumelhart et al., 1986; Werbos, 1988), Real Time Recurrent Learning (Williams and Zipser, 1989), AtiyaParlos learning rule (Atiya and Parlos, 2000), etc. Most of these a”
Long-term recurrent convolutional networks for visual recognition and description (2014, background): “Such “deep” models predated deep spatial convolution models in the literature [31, 45].”
Offline Reinforcement Learning as One Big Sequence Modeling Problem (2021, methodology): “(A full architectural description is provided in Appendix A.) Training is performed with the standard teacher-forcing procedure (Williams & Zipser, 1989) used to train sequence models.”
Professor Forcing: A New Algorithm for Training Recurrent Networks (2016, background): “In the RNN literature, this form of training is also known as teacher forcing (Williams and Zipser, 1989), due to the use of the ground-truth samples yt being fed back into the model to be conditioned on for the prediction of later outputs.”
Deep Interest Network for Click-Through Rate Prediction (2017, methodology): “Similar to NMT, they use RNN[24] to model text, then learn one global hidden vector to help paying attention on the key words in each query.”
Federated Learning for Mobile Keyboard Prediction (2018, background): “Neural models — in particular word and character-level recurrent neural networks (RNNs) [5] — have been shown to perform well on language modeling tasks [6, 7, 8].”
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.”
A practical tutorial on autoencoders for nonlinear feature fusion: Taxonomy, models, software and guidelines (2018, background): “These include DBNs ( Deep Belief Networks ) [10], CNNs ( Convolutional Neural Networks ) [11], RNNs ( Recurrent Neural Networks ) [12] as well as LSTMs ( Long Short-Term Memory ) [13] or AEs ( autoencoders ).”
Forward and Reverse Gradient-Based Hyperparameter Optimization (2017, methodology): “6 Forward-mode Computation Linked to RTRL [Williams and Zipser, 1989]”
Meta-Gradient Reinforcement Learning (2018, positive): “To compute the gradient of the updates, dθ′/dη, we note that the parameters form an additive sequence, and the gradient can therefore be accumulated online [Williams and Zipser, 1989], dθ′ dη = dθ dη + ∂f(τ, θ, η) ∂η + ∂f(τ, θ, η) ∂θ dθ dη = ( I + ∂f(τ, θ, η) ∂θ ) dθ dη + ∂f(τ, θ, η) ∂η (3)”
Recurrent Independent Mechanisms (2019, methodology): “During training, we predict the next frame at each time step using teacher forcing (Williams & Zipser, 1989).”