English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Humans and animals can learn complex predictive models that allow them toaccurately and reliably reason about real-world phenomena, and they can adaptsuch models extremely quickly in the face of unexpected changes. Deep neuralnetwork models allow us to represent very complex functions, but lack thiscapacity for rapid online adaptation. The goal in this paper is to develop amethod for continual online learning from an incoming stream of data, usingdeep neural network models. We formulate an online learning procedure that usesstochastic gradient descent to update model parameters, and an expectationmaximization algorithm with a Chinese restaurant process prior to develop andmaintain a mixture of models to handle non-stationary task distributions. Thisallows for all models to be adapted as necessary, with new models instantiatedfor task changes and old models recalled when previously seen tasks areencountered again. Furthermore, we observe that meta-learning can be used tometa-train a model such that this direct online adaptation with SGD iseffective, which is otherwise not the case for large function approximators. Inthis work, we apply our meta-learning for online learning (MOLe) approach tomodel-based reinforcement learning, where adapting the predictive model iscritical for control; we demonstrate that MOLe outperforms alternative priormethods, and enables effective continuous adaptation in non-stationary taskdistributions such as varying terrains, motor failures, and unexpecteddisturbances.

Deep Online Learning via Meta-Learning: Continual Adaptation for Model-Based RL