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

The world needs diverse and unbiased data to train deep learning models.Currently data comes from a variety of sources that are unmoderated to a largeextent. The outcomes of training neural networks with unverified data yieldsbiased models with various strains of homophobia, sexism and racism. Anothertrend observed in the world of deep learning is the rise of distributedtraining. Although cloud companies provide high performance compute fortraining models in the form of GPU's connected with a low latency network,using these services comes at a high cost. We propose Hydra, a system thatseeks to solve both of these problems in a novel manner by proposing adecentralized distributed framework which utilizes the substantial amount ofidle compute of everyday electronic devices like smartphones and desktopcomputers for training and data collection purposes. Hydra couples aspecialized distributed training framework on a network of these low powereddevices with a reward scheme that incentivizes users to provide high qualitydata to unleash the compute capability on this training framework. Such asystem has the ability to capture data from a wide variety of diverse sourceswhich has been an issue in the current scenario of deep learning. Hydra bringsin several new innovations in training on low powered devices including a faulttolerant version of the All Reduce algorithm. Furthermore we introduce areinforcement learning policy to decide the size of training jobs on differentmachines on a heterogeneous cluster of devices with varying network latenciesfor Synchronous SGD. The novel thing about such a network is the ability ofeach machine to shut down and resume training capabilities at any point of timewithout restarting the overall training. To enable such an asynchronousbehaviour we propose a communication framework inspired by the Bittorrentprotocol and the Kademlia DHT.

Hydra: A Peer to Peer Distributed Training & Data Collection Framework