NORDICS20 - Assets

Deep Learning on AWS

Amazon Web Services Resources EMEA

Issue link: https://emea-resources.awscloud.com/i/1242450

Contents of this Issue

Navigation

Page 27 of 50

Amazon Web Services Deep Learning on AWS Page 23 • Models, input tensors and batch sizes influence the amount of accelerator memory you need. Start with an accelerator type that provides at least as much memory as the file size of your trained model. • Demands on CPU compute resources, GPU-based acceleration, and CPU memory vary significantly between different kinds of deep learning models. The latency and throughput requirements of the application also determine the amount of compute and acceleration you need. Thoroughly test different configurations of instance types and Amazon Elastic Inference accelerator sizes to make sure you choose the configuration that best fits the performance needs of your application. Solutions Amazon SageMaker Ground Truth for Data Labeling Relative to other forms of machine learning, supervised learning continues to dominate the machine learning space. Feeding more data into the model training cycle continues to improve machine learning model performance. However, building a training dataset with accurate labels is a challenging and cost prohibitive task. Amazon SageMaker Ground Truth helps in the first step of the machine learning process when data is collected and labeled. Amazon SageMaker Ground Truth combines automated data labeling techniques based on active learning with crowdsourced data labeling processes using Mechanical Turk. You can use active learning to identify attributes that must be learned and then use crowdsourced workforce to perform the labeling. Active learning is a methodology that can sometimes significantly reduce the amount of labeled data required to train a model. It does this by prioritizing the labeling work for the experts. Active learning model looks at unlabeled data and calculates answers ranked by confidence. Next, the model compares its least confident scores against the labeled data. Last, the model tweaks itself so that if it sees the same data again, it is be more likely to calculate the correct answer. Besides active learning capability and access to Mechanical Turk workforce, Amazon SageMaker Ground Truth helps you with label management and workflow management. Optionally, you also set up private and hybrid workforces for the labeling task.

Articles in this issue

Links on this page

view archives of NORDICS20 - Assets - Deep Learning on AWS