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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Personal mobile sensing is fast permeating our daily lives to enable activitymonitoring, healthcare and rehabilitation. Combined with deep learning, theseapplications have achieved significant success in recent years. Different fromconventional cloud-based paradigms, running deep learning on devices offersseveral advantages including data privacy preservation and low-latency responsefor both model inference and update. Since data collection is costly inreality, Google's Federated Learning offers not only complete data privacy butalso better model robustness based on multiple user data. However, personalmobile sensing applications are mostly user-specific and highly affected byenvironment. As a result, continuous local changes may seriously affect theperformance of a global model generated by Federated Learning. In addition,deploying Federated Learning on a local server, e.g., edge server, may quicklyreach the bottleneck due to resource constraint and serious failure by attacks.Towards pushing deep learning on devices, we present MDLdroid, a noveldecentralized mobile deep learning framework to enable resource-aware on-devicecollaborative learning for personal mobile sensing applications. To addressresource limitation, we propose a ChainSGD-reduce approach which includes anovel chain-directed Synchronous Stochastic Gradient Descent algorithm toeffectively reduce overhead among multiple devices. We also design anagent-based multi-goal reinforcement learning mechanism to balance resources ina fair and efficient manner. Our evaluations show that our model training onoff-the-shelf mobile devices achieves 2x to 3.5x faster than single-devicetraining, and 1.5x faster than the master-slave approach.

MDLdroid: a ChainSGD-reduce Approach to Mobile Deep Learning for Personal Mobile Sensing