AWS Machine Learning

ECS Machine Learning

At ECS, we provide the highest levels of expertise as an AWS Premier Consulting Partner, AWS Audited Managed Service Provider, AWS DevOps and Microsoft Workloads Competency Partner. We utilize AWS Machine Learning (ML) services (https://aws.amazon.com/machine-learning/) through our AWS ML expert solution architects and data scientists to assist customers to create ML solutions with intelligent features such as computer vision, speech, video and language analysis—all within an integrated and cybersecure cloud environment. We offer services and tools tailored to meet our customers’ mission needs, utilizing industry-leading frameworks. Leveraging our experience with the comprehensive AWS services platform, we offer AWS ML-based solutions that provide integrated, secure, and pay-as-you-go analytics capabilities for data analysis supporting such activities as business data intelligence, processing, and workflow orchestration.

 

Customer Use Cases

Sun Edison Leverages ECS to Deploy AWS ML Solution

Challenge

The customer was facing a large ticket volume over time and wanted to get additional insights into the types of tickets that were being created. This included both the sentiment of the users inputting the tickets as well as how the support engineers handled the requests. Statistical methods had been used previously but can only find trends given specific criteria. The challenge was to have an ML-based analysis that discovered data that the statistical methods were not seeing.

Solution

As part of our cloud solution support services, ECS deployed AWS Comprehend to perform sentiment analysis and topic analysis of batches of cases as part of a service-level optimization and continuous improvement framework. Using this approach, unstructured text data was analyzed using Natural Language Processing (NLP) to analyze if the incident sentiment was positive, negative or neutral and to prioritize the ticket based on sentiment. Collections of tickets were analyzed for topic analysis. This analysis looked at both the text of the request of the submitter as well as the text of the support engineer’s responses.

We delivered this solution using the following Amazon Web Services: Amazon S3 and Comprehend.

Benefits

Benefits included the enhanced capability to both prioritize and resolve service requests and insights into trends that were not previously seen. The analysis that Comprehend performed began showing a correlation between tickets of unrelated product groups that may have been experiencing similar issues and requesting similar changes. This analysis was fed into a feedback loop for continuous improvement and coordination between the product lines.

ECS Supports Comtech with AWS ML- Enabled Solutions

Challenge

The customer required lower cloud administration costs and the ability to analyze costs across multiple AWS accounts and multiple AWS payer accounts. With multiple payer accounts, there is not a consolidated estimation of what forecasted spend will be based on a collection of accounts. The customer’s requirement was to be able to look at cost projections and the effects of optimizations by either looking at a specific AWS account or a group of AWS accounts to then look at the trending of data over time.

Solution

In support of our AWS MSP and cloud support solutions, ECS deployed a regression-based linear learner model to support advanced financial analysis of trends/usage to deliver advanced forecasting and prediction models for monthly costs. This model has been worked in both Amazon Machine Learning and Amazon SageMaker and is continuously refined based on lessons learned analyzing the inference. Amazon Athena is being used to conduct statistical analysis on the data for comparison to the algorithmic models that are trained through Machine Learning.

We delivered this solution using the following Amazon Web Services: AWS ML, Amazon S3, Amazon Athena, and Amazon SageMaker.

Benefits

Utilizing AWS ML, we identified models to support term purchases (also known as “reserved instances” or RIs) for EC2 and more accurately forecasted spend trends to support advanced budget and forecasting. This model is also allowing for more accurate forecasting of information based on specific criteria that are of interest to the business. These insights play into a larger feedback loop of cost optimization strategies and the tracking of the effectiveness of technical and business decisions that are made.

Global Education Software Platform Leverages ECS to Deploy AWS ML Solution

Challenge

Our customer operated a complex web service in AWS that did not support auto-scaling or automated responses to variable usage patterns. These usage patterns impacted the availability of the system and degraded the user experience, even causing losses in data. Because the system could not easily be re-architected or modernized to take advantage of native cloud scaling technologies, the customer required a way to predict peak usage days of the week and implement manual scaling solutions as well as staff augmentation to provide required support.

Solution

In support of our cloud support solutions, we determined the best indication to utilization of the customer application was the peak CPU utilization on each server that composed the application stack. We collected the CPU utilization metrics from CloudWatch and exported that data to S3 serialized in CSV format through a scheduled AWS Lambda function and began training an AI/ML model. We utilized the AWS Machine Learning service to generate a linear regression model that could predict peak CPU utilization for each instance based on the day of the week. Initially, this began as an empirical evaluation process of the model’s accuracy to help improve the customer’s confidence in the predictions before business processes were augmented to support the new insights. After several months of evaluation and testing, the AI model was generating information that could be relied upon to support manually scaling the application to meet expected usage.

We delivered this solution using the following Amazon Web Services: AWS ML, Amazon EC2, Amazon ELB, AWS Lambda, Amazon S3.

Benefits

By not having to run the application with enough capacity to meet peak requirements 24/7/365, the customer was able to reduce operational costs while still supporting the usage requirements of their user base. These measures provided a stop-gap method that provided more time to conduct re-engineering planning for the components of the system that did not support automated scaling actions and allowed the system to continue operating without needing to change its infrastructure or compliance documentation.

ECS deploys AWS ML solutions for Independent Nonpartisan Investigative Government Agency

Challenge

In order to help increase the security posture of a public-facing website, our client required the ability to intelligently filter thousands of access logs generated per hour to help them focus on patterns that posed a greater risk to the system. There was limited ability to develop complex training environments and manage AI/ML infrastructure, and the customer required a short turnaround from training to deployment.

Solution

In support of our MSP and cloud support solutions, we utilized the AWS Machine Learning service to add AI/ML capabilities to our managed services. We chose this supervised learning service in order to meet the customer’s requirements to reduce complexity and improve turnaround from training to detection. All access to the website is performed through an AWS Elastic Load Balancer (ELB), thus the ELB access logs provided a robust source of information to detect access patterns that posed risk to the customer’s website. Our resulting solution was to evaluate each ELB access log and provide insight as to whether the access was typical or atypical for the website. Using the AWS Machine Learning service, we created a binary classification model to classify each ELB access log in a batch each hour through an AWS Lambda function that collected all of the pertinent access logs, added the source country code, serialized the log data into CSV, stored them back to S3, and invoked the ML evaluation. We utilized the resulting evaluated logs to add to the training data set and conduct weekly retraining of the model to help improve the accuracy of each classification.

We delivered this solution using the following Amazon Web Services: AWS ML, Amazon EC2, Amazon ELB, AWS Lambda, Amazon S3.

Benefits

By providing additional insight into the access patterns of the website, we were able to filter out over 90% of the generated log data, which allowed a more granular focus on logs that demonstrated some type of potential risk to the system. This allowed the incident response team and security analysts to focus their time on more complex and harder problems by removing a majority of the manual log review process and provided much more extensive coverage of the log analysis since formerly only a sampling of the logs were reviewed.