Proceedings of the 2nd ACM International Workshop on Distributed Machine Learning

Rachid Guerraoui was the rst keynote speaker, and he got things o to a great start by discussing the broad relevance of the research done in our community relative to both industry and academia. He rst argued that, in some sense, the fact that distributed computing is so pervasive nowadays could end up sti ing progress in our community by inducing people to work on marginal problems, and becoming isolated. His rst suggestion was to try to understand and incorporate new ideas coming from applied elds into our research, and argued that this has been historically very successful. He illustrated this point via the distributed payment problem, which appears in the context of blockchains, in particular Bitcoin, but then turned out to be very theoretically interesting; furthermore, the theoretical understanding of the problem inspired new practical protocols. He then went further to discuss new directions in distributed computing, such as the COVID tracing problem, and new challenges in Byzantine-resilient distributed machine learning. Another source of innovation Rachid suggested was hardware innovations, which he illustrated with work studying the impact of RDMA-based primitives on fundamental problems in distributed computing. The talk concluded with a very lively discussion.

Download Full-text

MODES: model-based optimization on distributed embedded systems

Machine Learning ◽

10.1007/s10994-021-06014-6 ◽

2021 ◽

Author(s):

Junjie Shi ◽

Jiang Bian ◽

Jakob Richter ◽

Kuan-Hsun Chen ◽

Jörg Rahnenführer ◽

...

Keyword(s):

Machine Learning ◽

Embedded Systems ◽

Learning Model ◽

Black Box ◽

Distributed Embedded Systems ◽

Data Set ◽

Individual Model ◽

Model Based ◽

Machine Learning Model ◽

Distributed Machine Learning

AbstractThe predictive performance of a machine learning model highly depends on the corresponding hyper-parameter setting. Hence, hyper-parameter tuning is often indispensable. Normally such tuning requires the dedicated machine learning model to be trained and evaluated on centralized data to obtain a performance estimate. However, in a distributed machine learning scenario, it is not always possible to collect all the data from all nodes due to privacy concerns or storage limitations. Moreover, if data has to be transferred through low bandwidth connections it reduces the time available for tuning. Model-Based Optimization (MBO) is one state-of-the-art method for tuning hyper-parameters but the application on distributed machine learning models or federated learning lacks research. This work proposes a framework $$\textit{MODES}$$ MODES that allows to deploy MBO on resource-constrained distributed embedded systems. Each node trains an individual model based on its local data. The goal is to optimize the combined prediction accuracy. The presented framework offers two optimization modes: (1) $$\textit{MODES}$$ MODES -B considers the whole ensemble as a single black box and optimizes the hyper-parameters of each individual model jointly, and (2) $$\textit{MODES}$$ MODES -I considers all models as clones of the same black box which allows it to efficiently parallelize the optimization in a distributed setting. We evaluate $$\textit{MODES}$$ MODES by conducting experiments on the optimization for the hyper-parameters of a random forest and a multi-layer perceptron. The experimental results demonstrate that, with an improvement in terms of mean accuracy ($$\textit{MODES}$$ MODES -B), run-time efficiency ($$\textit{MODES}$$ MODES -I), and statistical stability for both modes, $$\textit{MODES}$$ MODES outperforms the baseline, i.e., carry out tuning with MBO on each node individually with its local sub-data set.

Download Full-text