kirill_makarov -

Manage Learn to apply best practices and optimize your operations.

Discover practical applications of machine learning on AWS

Many organizations still don't know how to get value out of AI, but industries ranging from finance to manufacturing have found practical applications for Amazon SageMaker.

Machine learning continues to evolve -- thanks to the explosive growth of on-demand public cloud resources -- but the technology requires much more than just compute instances.

Machine learning relies on complex models that developers must train and tweak in response to expansive reservoirs of real-world data. This process can be painfully slow, expensive and filled with complications. Amazon SageMaker is a managed service intended to alleviate much of that complexity, but how exactly can enterprises use it?

Let's examine what's under the hood of SageMaker, as well as some potential use cases for machine learning on AWS.

Amazon SageMaker basics

SageMaker helps developers gather raw data from in-house or publicly available repositories and then clean it for consistency. The data is often transformed or preprocessed in a Jupyter notebook so it can be combined or to make initial data relationships.

A developer can then use SageMaker to train a model and employ a selected algorithm to solve a problem. SageMaker offers a range of algorithms, but developers can also use custom algorithms for training. The actual scope of the compute deployment for this step depends on the size of the data set and how quickly the training must be completed. It can be accomplished with just a few instances but can also grow to involve a large distributed cluster of GPU instances, if needed. Once trained, developers test the model to determine its accuracy, either with an AWS software development kit for Python or another SageMaker Python library through a Jupyter notebook.

Then, the model is ready for deployment. A developer can move the model into an application without affecting the application itself, because SageMaker separates the model from the application code. Data scientists can then check the model's decision-making and integrate the verified results to retrain the model for further accuracy.

Potential SageMaker use cases

Though still relatively new, SageMaker has already demonstrated its predictive analytics capabilities in several industries. This isn't an exhaustive list of all its potential uses, but it does illustrate a wide range of practical situations where SageMaker, and other machine learning platforms, can support real-world tasks on the public cloud.

Medical image analysis: Physicians can spend considerable time searching X-rays and other medical images to look for signs of disease and illness. The medical industry increasingly uses machine learning models to compare patient images against extensive libraries to identify anomalies or even suggest diagnoses based on a model's analysis. This could potentially provide faster screenings, earlier and more accurate diagnoses and better patient outcomes.

Sports performance predictions: Major League Baseball and Formula One use machine learning on AWS to drive stats for their live broadcasts and applications. For example, SageMaker can process pitcher and hitter stats against other current conditions to suggest future pitch selection and its success or failure rate against the current batter.

Product forecasting: Companies can also use machine learning on AWS to analyze detailed manufacturing and sales data and predict demand based on seasonal, economic and other factors. SageMaker and other platforms can also provide detailed predictions and other recommendations across the product supply chain.

Bank/credit fraud prediction: The financial industry has made extensive use of machine learning technology. Platforms such as SageMaker can prepare models that monitor banking and credit activities for signs of abuse or fraud.

Marketing predictions: Organizations can use SageMaker models to make better marketing decisions, based on past behaviors or choices that the models have identified. For example, a marketing team could analyze a customer's web browsing habits and, based on those habits, render tailored ads in the hope of more sales.

Data analysis: Machine learning helps data analysts spot correlations, cause and effect, and other relationships in complex and disparate data sets. For example, SageMaker could support models for population analysis and use census data to identify population trends that help a government better allocate funding.

Dig Deeper on AWS machine learning