Automation, artificial intelligence (AI), and machine learning are hot technologies across all industries—including the federal government—because of their ability to enable a robot or computer to learn and handle a task that would otherwise require a human resource to complete.
In fact, this is extremely useful for the federal government, which has been dealing with constrained budgets, hiring freezes, and vacancies across its workforce for what feels like decades. But if some tedious and repetitive tasks can be automated and taught to machines to handle, the limited human resources available to government agencies can pivot and work on more mission-critical and essential functions.
However, machine learning can take time. It can require multiple, highly-skilled and well-trained data scientists to do the systems engineering required to teach AI even relatively simple tasks. This means that the federal government would need to recruit well-compensated data scientists just to get around hiring freezes, recruitment challenges and tight budgets.
So, how do we teach the learning machines without needing data scientists? That’s exactly what DataRobot’s solutions are looking to enable.
This is cutting edge and exciting technology that could have enormous ramifications for the federal government, so we invited DataRobot to our recent CSRA Emerging Tech Day. But before that, we sat down with Brian Espinola, the company’s Director of Business Development and Technical Services for Public Sector, to learn more about DataRobot. Here is what he had to say:
Thinking Next (TN): DataRobot claims to be an automated machine learning platform. What is an automated machine learning platform and what does it do?
Brian Espinola: In a nutshell, an automated machine learning platform automates many of the tedious tasks that a data scientist would do to implement machine learning solutions. DataRobot’s automated machine learning platform is designed to be easier, faster, and more accurate than any other offering, thereby addressing the shortage of data scientists available, and the long time it takes it implement accurate solutions.
Machine learning is a type of Artificial Intelligence that allows computers to program themselves based on the data presented. Machine learning algorithms are often used in predictive models, whereby historical labelled data is presented to a predictive model containing one or many machine learning algorithms, to train the model. Essentially, the model learns patterns from historical data and is trained to recognize them.
Once trained, the predicative model can be presented new data to make a prediction—usually in terms of a probability—of how likely the new data fits into any of the patterns it has learned. This is essentially how humans learn, but what makes machine learning so powerful is its ability to see patterns in huge volumes of data. But there are challenges in implementing machine learning for predictive models.
The process of assembling and tuning a predictive model has historically been a very manual, iterative, error-prone process that can take weeks, months or even years until is accurate. So, while machine learning algorithms learn without being explicitly programmed based on the data presented, they do require a lot of fine tuning of the data and input parameters to work properly. And because it can take so long, even the best data scientists are limited in how many different algorithms and modeling approaches they can try.
There is also a huge demand for data scientists, but—unfortunately—a very low supply, as it can take many years to build up the skills and experience needed to manipulate and transform data, apply math and statistics, and gain knowledge of as many different types of machine learning algorithms as possible. Finally—and arguably most important—data scientists need domain knowledge of the problem they are trying to solve.
DataRobot’s automated machine learning platform addresses many of the challenges mentioned by automating as many of the tasks as possible, while doing it easier, faster, and more accurately than previously possible. DataRobot also addresses the critical shortage of data scientists by changing the speed and economics of predictive analytics.
TN: How can automated machine learning be implemented within federal agencies to help them operate more effectively and efficiently?
Espinola: There are a few ways automated machine learning can help. The most obvious is speed. When you automate the tedious tasks, you gain speed efficiencies that save time and money, which can impact multiple operations. So, it can be a force multiplier and efficiency tool for the few data scientists that a federal agency has, enabling them to solve many more problems in a shorter amount of time.
Secondly, when an automated machine learning solution automates tasks that normally require coding skills, math and statistics skills, and algorithms selection skills, you can put the automated machine learning solution into the hands of people who understand the problem and data, but might not necessarily have those other skills—so called citizen data scientists.
Most agencies have large workforces of management, analysts, and technologists that know their missions, operations, underlying processes and related data that make their organizations function. We call them “domain experts.” Automated machine learning platforms can be used by domain experts to easily, quickly and accurately create predictive models that can pull out the patterns and make predictions.
Most federal agencies have huge volumes of data, a lot of which has been collected for many years. Many of them have invested in “Big Data” platforms that collect, transform, and manage the data. Most of the domain experts have ideas about how to use the data they collect to make their operations more efficient. So, if you put a tool in their hands that enables them to take action, and you train them how to spot opportunities for machine learning solutions, you can end up with a wide ecosystem of valuable use cases.
TN: How can constituents benefit from agencies that are utilizing automated machine learning platforms?
Espinola: We’re all taxpayers and we all want our money spent wisely. So, minimizing the fraud that exists with all the federal social services programs would save a lot of money. Identifying fraud is very well-established use case for machine learning in the private sector.
Protecting its constituents by providing security is also a major function of the government, and so identifying risk—in people, behaviors, transactions, locations, and events—is another well-established use case for machine learning in the private sector.
DataRobot’s solutions enable the implementation of machine learning implementations that could allow for spending constituents’ money more wisely and providing better security.
TN: What trends and challenges are facing federal organizations that make automation, machine learning—and automated machine learning—essential today?
Espinola: It’s all about optimization and efficiencies gained through these types technologies. Think about all the machine learning already in our daily lives—from the fraud detection applied to all our credit card transactions, to location-based smart phone apps, to the recommendations that Netflix and Amazon provide. All of these have machine learning behind them.
Corporations are hiring data scientists and paying them huge salaries to implement AI and machine learning to get a competitive edge by optimizing the customer interaction, making their business processes more efficient to save money, and ultimately maximizing their profits. The result is a better experience for their customers. We should be able to expect the same from our government.
We think it’s time for the federal government to leverage the same technologies that the commercial sector is implementing to increase efficiency and deliver the same benefits to their employees and constituents. But they can’t compete with the salaries paid to data scientists in the private sector, so an automated approach that is easy to use that can be leveraged by the workforce they do have is the way to do it.
TN: Are any government organizations currently utilizing DataRobot's solutions? Do you have metrics or anecdotes that illustrate the benefits that they're receiving?
Espinola: DataRobot Public Sector team was started in November of 2016, just shy of a year ago. We have been primarily focused on the Intelligence Community and Department of Defense. That hard work has resulted in seven customers across the intelligence and defense sectors. A common result of DataRobot implementations across these seven organizations has been the ability to repeatedly produce models as accurate—and often more accurate—in minutes or hours as compared to the models manually created by the data scientists that took weeks or months.
TN: DataRobot is one of the IT solution providers that participated in our recent CSRA Emerging Tech Day. Why did DataRobot participate in this event, and what benefit does the company look to receive from participating?
Espinola: CSRA is a forward-leaning provider of technology solutions and professional services, and has deep domain expertise across a wide variety of mission areas. Combined with CSRA’s focus on improving effectiveness and efficiency for both our government and our citizens, CSRA established its reputation as trusted technical advisor in many of our government’s agencies long ago.
DataRobot shares these core values, of using the latest technologies to improve effectiveness and efficiency for our customers’ missions. We’d like to partner on joint opportunities where we can leverage CSRA’s domain expertise to identify the right agencies and processes that would benefit from machine learning solutions, where CSRA and DataRobot can help train the government decision makers and domain experts to spot use cases themselves and use DataRobot to execute on them.
DataRobot hopes that its participation in the Emerging Tech day results in identifying specific opportunities where we can jointly help the government achieve its mission more effectively and efficiently.
Learn more about CSRA’s most recent Emerging Technology Days and hear more DataRobot.
Please note: The content on this page was originally posted on CSRA.com prior to its acquisition by General Dynamics. This content was migrated to GDIT.com on July 9, 2018.