首页|Study Data from Purdue University Update Understanding of Artificial Intelligenc e (MixTrain: accelerating DNN training via input mixing)
Study Data from Purdue University Update Understanding of Artificial Intelligenc e (MixTrain: accelerating DNN training via input mixing)
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News – Investigators publish new report on ar tificial intelligence. According to news reporting originating from West Lafayet te, Indiana, by NewsRx correspondents, research stated, “Training Deep Neural Ne tworks (DNNs) places immense compute requirements on the underlying hardware pla tforms, expending large amounts of time and energy. An important factor contribu ting to the long training times is the increasing dataset complexity required to reach state-of-the-art performance in real-world applications.” Our news correspondents obtained a quote from the research from Purdue Universit y: “To address this challenge, we explore the use of input mixing, where multipl e inputs are combined into a single composite input with an associated composite label for training. The goal is for training on the mixed input to achieve a si milar effect as training separately on each the constituent inputs that it repre sents. This results in a lower number of inputs (or mini-batches) to be processe d in each epoch, proportionally reducing training time. We find that naive input mixing leads to a considerable drop in learning performance and model accuracy due to interference between the forward/backward propagation of the mixed inputs . We propose two strategies to address this challenge and realize training speed ups from input mixing with minimal impact on accuracy. First, we reduce the impa ct of inter-input interference by exploiting the spatial separation between the features of the constituent inputs in the network’s intermediate representations . We also adaptively vary the mixing ratio of constituent inputs based on their loss in previous epochs.”
Purdue UniversityWest LafayetteIndia naUnited StatesNorth and Central AmericaArtificial IntelligenceMachine L earning
NETL
NSTL
