Abstract: A method (100) for synthesising multi-core magnetic metal oxide nanoparticles is disclosed. The method comprises providing a first precursor mixture (102) comprising a first metal-containing precursor, a first solvent, and a first nanoparticle clustering agent, and heating the first precursor mixture to thermally decompose the first metal-containing precursor to produce a nanoparticle mixture (104) comprising multi-core magnetic metal oxide nanoparticles. The method further comprises performing at least one seeded growth step (106), each comprising a feeding step in which a further precursor mixture (108) is added to the nanoparticle mixture, the further precursor mixture comprising a further metal-containing precursor and a further solvent, and a heating step in which the nanoparticle mixture is heated to thermally decompose the second metal-containing precursor to achieve growth of the multi-core magnetic metal oxide nanoparticles.

Abstract: A system and method provides efficient parallel training of a neural network model on multiple graphics processing units. A training module reduces the time and communication overhead of gradient accumulation and parameter updating of the network model in a neural network by overlapping processes in an advantageous way. In a described embodiment, a training module overlaps backpropagation, gradient transfer and accumulation in a Synchronous Stochastic Gradient Decent algorithm on a convolution neural network. The training module collects gradients of multiple layers during backpropagation of training from a plurality of graphics processing units (GPUs), accumulates the gradients on at least one processor and then delivers the gradients of the layers to the plurality of GPUs during the backpropagation of the training. The whole model parameters can then be updated on the GPUs after receipt of the gradient of the last layer.

Abstract: A system and method provides efficient parallel training of a neural network model on multiple graphics processing units. A training module reduces the time and communication overhead of gradient accumulation and parameter updating of the network model in a neural network by overlapping processes in an advantageous way. In a described embodiment, a training module overlaps backpropagation, gradient transfer and accumulation in a Synchronous Stochastic Gradient Decent algorithm on a convolution neural network. The training module collects gradients of multiple layers during backpropagation of training from a plurality of graphics processing units (GPUs), accumulates the gradients on at least one processor and then delivers the gradients of the layers to the plurality of GPUs during the backpropagation of the training. The whole model parameters can then be updated on the GPUs after receipt of the gradient of the last layer.