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

We consider a sequence of related multivariate time series learning tasks,such as predicting failures for different instances of a machine from timeseries of multi-sensor data, or activity recognition tasks over differentindividuals from multiple wearable sensors. We focus on two under-exploredpractical challenges arising in such settings: (i) Each task may have adifferent subset of sensors, i.e., providing different partial observations ofthe underlying 'system'. This restriction can be due to different manufacturersin the former case, and people wearing more or less measurement devices in thelatter (ii) We are not allowed to store or re-access data from a task once ithas been observed at the task level. This may be due to privacy considerationsin the case of people, or legal restrictions placed by machine owners.Nevertheless, we would like to (a) improve performance on subsequent tasksusing experience from completed tasks as well as (b) continue to perform betteron past tasks, e.g., update the model and improve predictions on even the firstmachine after learning from subsequently observed ones. We note that existingcontinual learning methods do not take into account variability in inputdimensions arising due to different subsets of sensors being available acrosstasks, and struggle to adapt to such variable input dimensions (VID) tasks. Inthis work, we address this shortcoming of existing methods. To this end, welearn task-specific generative models and classifiers, and use these to augmentdata for target tasks. Since the input dimensions across tasks vary, we proposea novel conditioning module based on graph neural networks to aid a standardrecurrent neural network. We evaluate the efficacy of the proposed approach onthree publicly available datasets corresponding to two activity recognitiontasks (classification) and one prognostics task (regression).

Continual Learning for Multivariate Time Series Tasks with Variable Input Dimensions