The Pathway to Machine Learning in Federal
The need for machine learning within the Federal government and the Department of Defense (DoD) is loud and clear, as illustrated in the following comments.
Robert Work, Deputy Secretary of Defense stated, "Numerous studies have made clear that the DoD must integrate artificial intelligence and machine learning more effectively across operations to maintain advantages over increasingly capable adversaries and competitors. Although we have taken tentative steps to explore the potential of artificial intelligence, big data, and deep learning, I remain convinced that we need to do much more, and move much faster across DoD to take advantage of recent and future advances in these critical areas."
Lt. Gen. John N.T. "Jack" Shanahan, Director of Defense Intelligence, Warfighter Support, Office of the Under Secretary of Defense for Intelligence, commented: "The first and perhaps most important step [to solving our data problem] is to understand that it is not possible to solve these problems with brute force alone. Adding 1,000 more intelligence analysts is neither realistic nor feasible in today's fiscal environment. We must instead find creative ways to adapt to this new environment in which we are already deeply immersed. Artificial intelligence, machine learning, and deep learning [are] the critical base ingredients in the recipe for future success."
Focus on the Mission Outcome
At Sila we view machine learning from a business or mission-outcome perspective. That is, machine learning is all about finding real value in data for predictive analysis through discovery of trends and patterns. Discovering new insights from data and applying them to a business or mission challenge is key. Building upon big data capabilities that store, process, and access data, machine learning involves creating data models to discover patterns that can then be used to analyze future data sets and predict patterns.
The use cases for machine learning are limitless. Specific to the Federal government, use cases may include:
Pathway to Machine Learning
Many federal agencies are now on the path to understanding how machine learning can apply to their specific missions. For agencies exploring the transition from understanding to implementation, Sila offers the following items to consider when embarking on a machine learning journey.
Start with security
Like any systems integration project, it is critical to include security considerations into the initial requirements and design process.
Questions specific to data are important. For example, who can see the data? How do I change data access rights? Can I integrate with Active Directory/Lightweight Directory Access Protocol (AD/LDAP) solutions? Can I anonymize data at the row or cell level? Can I share my data, algorithms, and project results with other data scientists and then change access controls when needed?
Answers to questions around data management, data security, governance, and lineage are fundamental before starting any machine learning project.
Focus on the mission
Understand exactly what problem you are trying to solve. Choose a technology to fit the problem, not the other way around. Frame the question to maximize value from the technology.
Need data, will travel
One of the biggest challenges to any machine learning project is getting access to data sets. This often requires getting access from multiple data owners, as well as different data types (SQL, HDFS, etc.). Data owners must be willing to share data and participate in machine learning projects.
In its simplest form, machine learning is based on algorithms to identify trends based on historical data and then make predictions. Better data, better algorithms, better insights.
I finally have the data, now what?
After obtaining the data set, it is time to fully understand the data. Engage domain experts, data scientists, and developers. Understand your data so that you can engineer it to maximize its usefulness. This might require adding new data elements, merging multiple data sources, conducting data analysis, and beginning "feature" engineering. In machine learning, a feature is an individual attribute or "explanatory variable." It takes time and domain expertise to identify specific, independent features in your data. Knowledge of the data is key to selecting appropriate features to make algorithms successful. After features are selected, start training and refining the model.
Manage the data
Gone are the days when everyone moves data to a single data warehouse or data lake or Hadoop ecosystem. Having a control layer in place makes it easier to pull data from multiple sources and make changes, especially as it relates to data access and data sharing.
Leverage the legacy data stores that you have, and then manage both data and interactions (connectors) to accelerate access to the data.
Eliminate manual checkpoints to optimize the feedback loop between the model outputs and the overall enterprise.
Ensure policies are in place and enforced for compliance and security.
Remember the use case. Avoid "the science project syndrome" and stay focused on answering the initial question and identifying actual insights that can be gained from the model. Find ways to communicate these insights. Many tools provide visualization methods to make this easier.
Make models "production ready" and create long-term sustainability
Operationalize by moving from proof of concept to production quickly. Once environments and control layers are in place, continue to add use cases and more data sets.
Enabling an organization to make the most out of machine learning and data science requires a long-term commitment to build talent and grow skills over time. Entering the machine learning space might require a shift in skill set from operational analytics to predictive analytics.
Recognize that a culture shift might be required as leadership needs to recognize the importance of making decisions based on data insights as opposed to gut emotions. Encourage data sharing and welcome collaboration.
Maintain your models. Data changes over time. Trends change over time. Building accurate, predictive models is an ongoing effort. Develop a plan to track your model's performance and a cycle to update it.