Machine Learning - Research (EN)

NucleusResearch.com 5 Document Number: T176 December 2019 INFORMATION TECHNOLOGY SERVICES PROVIDER The company is a consultancy for AWS that helps customers build applications and execute projects on AWS. Among other projects, the company helps customers implement machine learning. After helping customers establish best practices and infrastructure around collecting and preparing data, the company focuses on implementing classification and forecasting projects. The company is currently working on seven machine learning projects and it runs all of them on Amazon SageMaker. It's best-suited for companies without many data science resources or experience operationalizing data. It provides a well-architected way for visualizing and evaluating results from models in a centralized location, allowing leaders to choose the most accurate models for the task. The automated training is a huge benefit for the company as it can be done with a few clicks and doesn't need to be supervised, allowing the developers to complete other tasks simultaneously. Using SageMaker Ground Truth, the time spent labeling and preparing data has reduced by over 50 percent. On average, the consultancy allocated one to two weeks to labeling datasets for training machine learning models, with SageMaker Ground Truth it is able to complete labeling in less than a week it is expected to accelerate further. Altogether, SageMaker allows the consultancy to take on approximately 3 times the machine learning projects as it could when the projects were all self-managed, allowing for higher margins and greater expertise in machine learning applications for future projects. BUSINESS PROCESS AUTOMATION VENDOR The company creates web tools for automating administrative tasks. It is building a document understanding and data extraction platform for customers to input documents and automatically record relevant fields and insights from it, allowing customers to automate tedious manual tasks like document review. The platform leverages machine learning for natural language processing and computer vision, and the company manages all the cooperating machine learning models with Amazon SageMaker. The machine learning algorithms are compute-heavy, so the company leverages GPU instances for training and inference; it is easy to these attach GPU instances to the cluster in Amazon SageMaker, increasing performance while remaining relatively straightforward for developers. There are five models in production at once, so having a centralized location to manage them all is key. The company estimates that by building with AWS, it has accelerated the overall development cycle by 30 percent. It formerly maintained infrastructure on-premises, and has since been able to retire those machines and redeploy the IT resources to other value-add tasks. The peace of mind from no longer worrying about the performance or security of the on-site hardware is difficult to quantify, but one of the biggest benefits to the company leadership.

• Cover