Learning to Learn Using Gradient Descent

By Sepp Hochreiter, A. S. Younger, Peter R. Conwell. 2001

339 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Computational Theory and Mathematics, Meta & Continual Learning

How researchers see this paper

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

In Artificial Intelligence

In Computational Theory and Mathematics

In Meta & Continual Learning

Citations in the corpus

Language Models are Few-Shot Learners (2020, methodology): “As indicated by the name, few-shot learning as described here for language models is related to few-shot learning as used in other contexts in ML [HYC01, VBL + 16] – both involve learning based on a broad distribution of tasks and then rapidly adapting to a new task.”
Few-shot Autoregressive Density Estimation: Towards Learning to Learn Distributions (2017, background): “Learning to learn or meta-learning has been studied in cognitive science and machine learning for decades (Harlow, 1949; Thrun & Pratt, 1998; Hochreiter et al., 2001).”
Auto-Meta: Automated Gradient Based Meta Learner Search (2018, background): “The other major category of meta learning is to learn an optimizer as the meta-learner which enables the learner to learn a new task more effectively [10, 1].”
Meta-Gradient Reinforcement Learning (2018, background): “Hochreiter et al. [2001] used recurrent neural networks like Long Short-Term Memory (LSTM) [Hochreiter and Schmidhuber, 1997] as meta-learners.”
Probabilistic Model-Agnostic Meta-Learning (2018, background): “A number of works have approached the problem of few-shot learning from a meta-learning perspective [35, 19], including black-box [33, 5, 42] and optimization-based approaches [31, 9].”
Meta-Learning with Temporal Convolutions (2017, methodology): “A number of methods consider a meta-learner that makes updates to the parameters of a traditional learner [10, 4].”
Embracing Change: Continual Learning in Deep Neural Networks. (2020, background): “In [86,87] both loops use gradient descent , while inner-loops can also be nearest-neighbor matching [88,89], recurrent neural representations [8,90,91], probabilistic inference [82,92], or mixed approaches [93 – 96].”
Learning to Continually Learn (2020, background): “Two promising ones are combining ANML with the style of meta-learning wherein the entire learning algorithm is meta-learned within a recurrent neural network (either in a supervised [14] or RL context, e.g. RL 2 [6] aka Learning to Reinforcement Learn [43]).”
Meta-Learning: A Survey (2018, neutral): “[65] use LSTMs [66] as a meta-learner to train multi-layer perceptrons.”
Online Meta-Learning (2019, neutral): “Meta-learning: Prior meta-learning works have proposed learning an update rule or optimizer for fast adaptation (Hochreiter et al., 2001; Bengio et al., 1992; Andrychowicz et al., 2016; Li & Malik, 2017; Ravi & Larochelle, 2017) by having one model output the weights of another (Ha et al., 2017),…”
Discovering objects and their relations from entangled scene representations (2017, background): “Sequences – or episodes – consisted of 50 random samples generated from ﬁve unique graphs, from a pool of 1900 total classes, presented jointly with time-offset label identiﬁers, as per Hochreiter et al. (2001) and Santoro et al. (2016).”
Optimal Ordered Problem Solver (2002, neutral): “The stack used by the iterative equivalent of procedure Try for storage management (Section 4.1) never held more than 20,000 elements though.”
A Baseline for Few-Shot Image Classification (2019, background): “This mapping can be learnt using temporal models [12, 20], or one can back-propagate the gradient across the base-level updates [2, 21].”
One Big Net For Everything (2018, methodology): “Many RNN-like models can be used to build general computers, e.g., RNNs controlling push-down automata [9, 41] or other types of differentiable memory [20] including differentiable fast weights [63, 65], as well as closely related RNN-based meta-learners [66, 25, 51].”
Generalizing from a Few Examples (2019, background): “Meta-learning [59] improves P of the new task T by the provided data set and the meta-knowledge extracted across tasks by a meta-learner (Figure 16).”
Natural Language to Structured Query Generation via Meta-Learning (2018, methodology): “This direction has been applied to learning to optimize deep neural networks (Hochreiter et al., 2001; Andrychowicz et al., 2016; Li and Malik, 2017; Ha et al., 2017).”
Meta-Learning with Warped Gradient Descent (2019, methodology): “Hochreiter et al. (2001) introduced gradient descent methods to meta-learning, speciﬁcally for recurrent meta-learning algorithms.”
Been There, Done That: Meta-Learning with Episodic Recall (2018, methodology): “More precisely, the LSTM cell update equation (Hochreiter et al., 2001) uses multiplicative gates i t and f t , deﬁned as follows, to coordinate between the contributions of current working memory and incoming perceptual inputs To the above update rule we add a new term for the contribution of…”
Practical tradeoffs between memory, compute, and performance in learned optimizers (2022, methodology): “…optimizers specify parameter update rules using a ﬂexible parametric form and learn the parameters of this function from a “dataset” of optimization tasks—a procedure typically referred to as meta-training or meta-learning (Andrychowicz et al., 2016; Finn et al., 2017; Hochreiter et al., 2001).”
Reinforcement Learning, Fast and Slow. (2019, positive): “As discussed in the main text, one implementation of this idea employs a form of recurrent neural network commonly used in machine learning: long-short term memory units (LSTM) [83, 88].”