Machine Learning

April 2019, IDC #US44941919 White Paper Developing Applications for an AI-Enabled World Sponsored by: Amazon Web Services Ritu Jyoti David Schubmehl April 2019 IN THIS WHITE PAPER In today's hypercompetitive business environment, organizations are continually seeking to improve and provide better value to their customers, shareholders, and workers. Specifically, they're looking for new ways to increase sales, reduce costs, streamline business processes, and understand their customers better by using various types of automation coupled with the ever-increasing amount of data available to them. To that end, many organizations have started to look at machine learning and deep learning as methods to build real-world artificial intelligence (AI) into their applications and business processes. Machine learning is the process of creating a statistical model from various types of data that perform various functions without having to be programmed by a human. Deep learning is a type of machine learning based on neural network algorithms, used to produce more accurate insights, recommendations, and predictions, trained on large amounts of data. Organizations are using deep learning models to recommend products, predict pricing, recognize images, and improve decision making as well as for a host of other use cases. Until recently, developing deep learning models took significant amounts of time, effort, and knowledge and required expertise in this field. Recently vendors like Amazon Web Services (AWS) have developed services and tools for deep learning, that allow data scientists and developers to develop and deploy deep learning models more quickly and easily. There are numerous machine/deep learning tools and frameworks such as TensorFlow, Apache MXNet, and PyTorch — all have valuable attributes that make them useful in developing intelligent applications. However, there are many factors involved that inhibit the development of machine learning applications: ▪ Choosing the right machine/deep learning framework for the job at hand ▪ Choosing the right machine/deep learning algorithm ▪ Adjusting and tuning the machine algorithm and data for the most accurate predictions ▪ Identifying, locating, and curating training data for machine learning models ▪ Having the right amount of compute resources for both model training and generating predictions in production (inferences) ▪ Integrating machine/deep learning models into existing enterprise applications ▪ Operationalizing models to perform at scale in production

• Cover