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 9 of 50

Amazon Web Services Deep Learning on AWS Page 5 We will discuss ways to mitigate these unique challenges to deep learning in the Highly Optimized AWS Technology Building Blocks for Deep Learning section of this paper. As a first step, use the diagram below to assess your data collection process. Figure 3: Data collection assessment 1 Data Preprocessing Data preprocessing comprises data cleaning, data integration, data transformation, and data reduction, with the intent to mitigate inaccurate, missing, noisy, and inconsistent data before starting the training process. AWS provides a variety of tools and services that you can use to perform the data preprocessing steps in addition to performing feature engineering: AWS Glue, Amazon EMR, AWS Lambda, Amazon SageMaker, AWS Batch, and AWS Marketplace. The use of these tools is described in detail in the Big Data Analytics Options on AWS whitepaper. Most important, with the widespread availability of many open source deep learning frameworks, a broad variety of file formats have emerged to accommodate the individual frameworks. The choice of file format for your data ingestion process is an important step in the data preprocessing phase and greatly depends on the framework chosen to perform the deep learning implementation. Some of the standard formats include RecordIO, TFRecords, Hierarchical Data Format (HDF5), pth, N5, and light memory mapped database (LMDB). Step 2. Choose and Optimize Your Algorithm Within deep learning implementations, we differentiate between various network architectures and deep learning algorithms. Discussing every available network architecture and learning algorithm is outside the scope of this paper. For brevity, we briefly discuss three of the most commonly used network architectures and some popular learning algorithms used today.

Articles in this issue

Links on this page

view archives of NORDICS20 - Assets - Deep Learning on AWS