A Resilient Distributed Boosting Algorithm
This work addresses communication efficiency in distributed learning for parties handling noisy data, but it is incremental as it builds on classical boosting with a new robustness component.
The authors tackled the problem of minimizing communication in distributed learning with noisy data by presenting a resilient distributed boosting algorithm, achieving resilience to a limited amount of noise while proving that higher noise levels are not feasible with communication-efficient methods.
Given a learning task where the data is distributed among several parties, communication is one of the fundamental resources which the parties would like to minimize. We present a distributed boosting algorithm which is resilient to a limited amount of noise. Our algorithm is similar to classical boosting algorithms, although it is equipped with a new component, inspired by Impagliazzo's hard-core lemma [Impagliazzo95], adding a robustness quality to the algorithm. We also complement this result by showing that resilience to any asymptotically larger noise is not achievable by a communication-efficient algorithm.