Key Considerations for User-Driven AI/ML Modeling, Development, and Management
By Will Astore
Vice President, Spatial Solutions
and Ken McCall
Senior Director of Business Development, Spatial Solutions
Within the defense community, the current approach to data analytics and modeling falls short of deriving meaningful insights. Many artificial intelligence and machine learning (AI/ML) solutions are rigid—they fail to integrate a wide range of data sources; they aren’t built for specific needs; and they produce more noise than signal. Effective AI/ML modeling is more than just applying a one-size-fits-all application: it requires looking at an organization holistically in order to design customized solutions tailored to an enterprise’s complex data, infrastructure, and operational needs.
ECS is the second-largest provider of AI/ML solutions to the defense community with 300+ AI models deployed. We help federal and commercial organizations create AI/ML and analytics solutions that generate predictive insights, reduce signal noise, streamline operational workflows, and improve model outcomes.
Three Key Considerations
for AI/ML Modeling
1. Adopt a Modular Architecture for Better Data Integration
The foundation of any good modeling system is the data sources that feed it. But data sources are often specific to an agency, business, or organization. An effective AI/ML solution requires systems to be built in a modular fashion, allowing development to operate in parallel between data sources.
2. Use Continuous Feedback to Build Custom-tailored Solutions
Creating a continuous feedback loop between users and application engineers ensures a solution is built to the user’s requirements and can easily integrate into the user’s workflow. Customization does not stop at application design–good systems must scale to specific user software, operations, hardware, and even business processes. Strong solutions providers are more than just developers, but operations experts who can work with clients to identify opportunities for workflow enhancement and improved modeling capabilities.
3. Model Relationships and Activities. Not Just Objects.
A good solution allows users to make valuable abstractions from complex data sets, but a great solution covers broad scopes of data while removing the complexities of the science itself. These solutions create ways of showing the relationships and activities between data points and don’t just treat the models as objects. By transforming unstructured data into uniform sets, great solutions pave the way for machine learning models to produce predictive analytics with high-fidelity and low signal noise.
The ECS Approach
Iterate, Iterate, Iterate…with Rapid Prototyping
At ECS, we build solutions that analyze and support the entire user workflow from start to finish. The process starts by identifying user groups and gathering requirements using a user-centric Design Thinking process. Then, following a phase-based approach, ECS experts create organizational processes that shorten the feedback loop between users and prototyping engineers, driving continuous application development.
A Modular System Design Owned by the Customer
Our analytics application solution follows a Modular Open Systems Approach (MOSA). MOSA is not a software design pattern, but a pattern for full system development. Using MOSA as a guide, ECS experts construct technical solutions that are adaptable and cost effective.
On a granular level, our processes for software engineering also follow a modular approach. They use a microservice architecture that allows our engineers to work different parts of the application development process in parallel. This leads to loosely coupled services that can scale independently of one another.
With modularity, systems build off one another. Our software solutions focus on exposed representational state transfer application programming interfaces (REST APIs), which let existing and new applications tap into the data, models, and insights our solution provides.
The ECS system design process is open in the truest sense of the word—ECS builds solutions with open-source frameworks and libraries. Our solution utilizes containerization for cloud-native, vendor-agnostic deployments. Most importantly, our solution is owned by the customer. Since the customer owns all source code, documentation, and training material, the transition from vendor to vendor is seamless.
Build a Foundation for AI/ML Success
Reaching a state where your organization can truly utilize AI/ML solutions doesn’t happen overnight. It takes time to integrate data, train models, and verify that the models are correct. The ECS approach builds a solid foundation that allows for AI/ML enhancements far into the future.
Our solutions allow users to model beyond surface-level abstractions and produce meaningful narratives from their models. With an intuitive modeling user interface, our solution uses defined models to help transform unstructured data into structured datasets and enable users to see data through multiple lenses. The data transformation performed by modeling is the backbone of analytics algorithms. Using this compliant data, our solutions can automate the creation of training sets for AI/ML algorithms that help provide dynamic predictive outcomes.
