The Task Rehearsal Method of Life-Long Learning: Overcoming Impoverished Data

By Daniel L. Silver, Robert E. Mercer. 2002

144 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Meta & Continual Learning

Citations in the corpus

Incremental Task Learning with Incremental Rank Updates (2022, background): “Incremental task learning (ITL) [10,34] aims to train a single model on a sequence of different tasks and perform well on all the trained tasks once the training is finished.”
ConTinTin: Continual Learning from Task Instructions (2022, background): “(2)Continual learning in the literature is also referred to as: lifelong learning (Silver and Mercer, 2002), incremental learning (Solomonoff, 1989), sequential learning (McCloskey ing problem was mainly studied in computer vision or robotics domains, and most work concentrated on mitigating catastr”
Continual Lifelong Learning in Natural Language Processing: A Survey (2020, background): “Continual learning in the literature is also referred to as: lifelong learning (Silver and Mercer, 2002; Silver et al., 2013; Chen and Liu, 2018; Chaudhry et al., 2019a; Parisi et al., 2019; Aljundi et al., 2017), incremental learning (Solomonoff, 1989; Chaudhry et al.”
Rotate your Networks: Better Weight Consolidation and Less Catastrophic Forgetting (2018, background): “In a lifelong learning setting, however, learning is considered as a sequence of tasks to be learned [1], which is more similar to how biological systems learn in the real world.”
Memory Aware Synapses: Learning what (not) to forget (2017, background): “With tasks being dealt with in a sequential manner [25, 26], the absence of the data from the previous tasks introduces the risk of catastrophic forgetting [14, 20, 13, 6, 7] of the previously acquired knowledge.”
Encoder Based Lifelong Learning (2017, background): “It occurs when the datasets or the tasks are presented to the model separately and sequentially, as is the case in a lifelong learning setup [25, 26, 24].”
Lifelong Generative Modeling (2017, background): “[33, 71, 91] relaxes the need for model persistence, but requires preservation of all data [33, 71], or a growing core-set of data [116, 114, 91].”
Continual learning: A comparative study on how to defy forgetting in classification tasks (2019, methodology): “This results in an ordering from hard to easy consisting of task sequence [5, 7, 10, 2, 9, 8, 6, 4, 3, 1] from the random ordering, with the inverse order equal to the easy to hard ordering.”
Class-Incremental Learning: Survey and Performance Evaluation on Image Classification (2020, background): “Some regularize on the weights and estimate an importance metric for each parameter in the network [5], [7], [33], [34], [38], [39], while others focus on the importance of remembering feature representations [20], [35], [40], [41], [42], [43].”
Continual Learning in Neural Networks (2019, methodology): “Several directions have been explored such as reducing representation overlap [39, 40, 78, 79, 144], replay samples or virtual samples from previous history [124, 142] or dual architectures [41, 130, 12].”
A Continual Learning Survey: Defying Forgetting in Classification Tasks (2019, methodology): “Silver et al. [8] ﬁrst proposed to use previous task model outputs given new task input images, mainly for improving new task performance.”
Differentiable Programs with Neural Libraries (2016, background): “…base allows either (1) rehearsal on historical examples (Robins, 1995), (2) rehearsal on virtual examples generated by the frozen networks (Silver & Mercer, 2002; Silver & Poirier, 2006) or (3) creation of new networks containing frozen sub networks from the historical tasks (Rusu et al.,…”
Inductive transfer with context-sensitive neural networks (2008, methodology): “The Band domain, described in Silver and Mercer (2002), consists of seven synthetic tasks.”
Drinking From a Firehose: Continual Learning With Web-Scale Natural Language (2020, background): “al., 2017; Zenke et al., 2017; Nguyen et al., 2018; Ritter et al., 2018; Chaudhry et al., 2018; Schwarz et al., 2018; Aljundi et al., 2019a; Zhang et al., 2020; Jung et al., 2016; Triki et al., 2017; Silver and Mercer, 2002; Aljundi et al., 2017; Swaroop et al., 2019; Ahn et al., 2019; Chen et al., ”
Continual Test-Time Domain Adaptation (2022, background): “Continual learning methods can often be categorized into replaybased [49] and regularization-based [53, 72] methods.”
On Training Recurrent Neural Networks for Lifelong Learning (2018, background): “This rehearsal setup (Silver and Mercer, 2002) can help in two ways - if the tasks are related, training on multiple tasks helps in transferring knowledge across the tasks.”