DICE: Data Influence Cascade in Decentralized Learning

Main Results What scientific problem does this paper study? Q: In decentralized financial systems, proof of work (PoW) ensures security and consensus through computational effort. How can PoW be formally defined and quantified in the context of decentralized distributed machine learning? Data Influence Cascade in Decentralized Learning Main Results The influence of data "cascades" through the communication graph, resembling "ripples in water". What phenomena does this paper uncover? What scientific problem does this paper study? Q: In decentralized financial systems, proof of work (PoW) ensures security and consensus through computational effort. How can PoW be formally defined and quantified in the context of decentralized distributed machine learning? Data Influence Cascade in Decentralized Learning Main Results The influence of data "cascades" through the communication graph, resembling "ripples in water". What phenomena does this paper uncover? What scientific problem does this paper study? Q: In decentralized financial systems, proof of work (PoW) ensures security and consensus through computational effort. How can PoW be formally defined and quantified in the context of decentralized distributed machine learning? Data Influence Cascade in Decentralized Learning Main Results In decentralized learning, the influence of data "cascades" through the communication graph, resembling "ripples in water".