Moreno Álvarez, SergioHaut, Juan M.Paoletti, Mercedes EugeniaRico Gallego, Juan Antonio2024-11-182024-11-182021-06-21Sergio Moreno-Alvarez, Juan M. Haut, Mercedes E. Paoletti, Juan A. Rico-Gallego, Heterogeneous model parallelism for deep neural networks, Neurocomputing, Volume 441, 2021, Pages 1-12, ISSN 0925-2312, https://doi.org/10.1016/j.neucom.2021.01.1250925-2312 | eISSN 1872-8286https://doi.org/10.1016/j.neucom.2021.01.125https://hdl.handle.net/20.500.14468/24396The registered version of this article, first published in “Neurocomputing, Volume 441", is available online at the publisher's website: Elsevier, https://doi.org/10.1016/j.neucom.2021.01.125 La versión registrada de este artículo, publicado por primera vez en “Neurocomputing, Volume 441", está disponible en línea en el sitio web del editor: Elsevier, https://doi.org/10.1016/j.neucom.2021.01.125Deep neural networks (DNNs) have transformed computer vision, establishing themselves as the current state-of-the-art for image processing. Nevertheless, the training of current large DNN models is one of the main challenges to be solved. In this sense, data-parallelism has been the most widespread distributed training strategy since it is easy to program and can be applied to almost all cases. However, this solution suffers from several limitations, such as its high communication requirements and the memory constraints when training very large models. To overcome these limitations model-parallelism has been proposed, solving the most substantial problems of the former strategy. However, describing and implementing the parallelization of the training of a DNN model across a set of processes deployed on several devices is a challenging task. Current proposed solutions assume a homogeneous distribution, being impractical when working with devices of different computational capabilities, which is quite common on high performance computing platforms. To address previous shortcomings, this work proposes a novel model-parallelism technique considering heterogeneous platforms, where a load balancing mechanism between uneven devices of an HPC platform has been implemented. Our proposal takes advantage of the Google Brain’s Mesh-TensorFlow for convolutional networks, splitting computing tensors across filter dimension in order to balance the computational load of the available devices. Conducted experiments show an improvement in the exploitation of heterogeneous computational resources, enhancing the training performance. The code is available on: https://github.com/mhaut/HeterogeneusModelDNN.eninfo:eu-repo/semantics/openAccess12 Matemáticas::1203 Ciencia de los ordenadores ::1203.17 Informáticaartículodeep learninghigh performance computingdistributed trainingheterogeneous platformsmodel parallelism