Active Learning: Is your AI-model working hard enough?

Ellie Mae, now a part of ICE Mortgage Technology, is at the forefront of applying artificial intelligence (AI) and robotic process automation (RPA) technology to the unique workflows and use cases specific to mortgage industry loan originators, investors, and servicers. In this post, we further explore our new eBook, entitled Achieving Automation with AI & RPA, to discuss how to incorporate active learning to unlock a more scalable and responsive system. Here are some key components required to achieve this.  

Neural networks 

Like the human brain, neural networks move through numerous layers to process data and make a decision. They are non-deterministic, meaning they learn based on past patterns and corrections, rather than explicit rule sets. While deterministic rules may be helpful for a sub-set of narrowly defined information, they may be less flexible in the mortgage industry, where documents and data are varied and ever evolving. Given the scale of big data and machines,  neural networks can achieve in seconds what would take weeks or even months for humans—all with greater accuracy and reduced risk.

Active learning

Active learning is a machine learning methodology that enables more rapid training of a system. In active learning, an algorithm interactively queries the user to obtain the desired outputs at new data points. For example, when a new state-specific document is released that was not previously processed, it can be routed to a validation queue where humans then label the document. That newly labeled document subsequently becomes an input to the AI model, and future versions of this same document can be more readily recognized. 

Explore vs. exploit 

AI systems are intended to strike the balance between “explore” versus “exploit” activities. Explore activities enable the system to learn, and are often referred to as “training the model.” Exploit activities take those learnings and use them to drive decisions and actions. Exploring for an extended period of time may lead to greater intelligence in the system, but the benefits may not be realized if it is not deployed in a timely manner. Likewise, if an AI system spends too little time exploring and moves too rapidly into exploit mode, the time to market is compelling but the results can be sub-optimal. Therefore, cycling between periods of exploration and exploitation are central to identifying new possibilities, refining the prediction model, and putting the improved model to use. Fortunately, ICE Mortgage Technology’s footprint in the mortgage industry and insights gleaned from its big data repository accelerates the explore mode so that customers can more quickly realize value from the exploit mode. The end results include improved accuracy, greater cost savings, and increased capacity. 

Tuning and trusting the AI model

The sheer size and ever-evolving nature of mortgage industry documents and data underscore the importance of diligent, ongoing tuning and calibration of algorithms in an AI model. While machines can process enormous volumes of information more rapidly than humans, a common hesitation for organizations is whether or not they can trust the accuracy of an AI solution over a process solely managed by humans. The degree of accuracy of human review in an AI solution is another common concern. Ellie Mae AIQ accounts for each of these valid points, with a clear, data-driven approach: Every iteration of data flowing through an AI model is verified against a regression data set to confirm that no errors are made when labeling documents or data. Thanks to AIQ’s discipline, experience, and scale in the tuning and calibration of its AI models, mortgage lenders using the solution can significantly reduce errors and lower compliance risk. 

Download the full eBook now, to learn even more about the advanced AI capabilities of AIQ and how to experience true mortgage process automation.  

Keep your AI momentum going...

Listen to our “Get peace of mind in the unknown with RPA and AI” podcast to hear Eric Kujala, Product Marketing Director at ICE Mortgage Technology, and Nolan Johnson, Sales Engineer, HPA at A Cognizant Company, discuss the key things to consider when evaluating RPA for your business. 

Remember to register for Experience 2021 to access exclusive content to help you accelerate AI adoption, ask questions at live Q&A sessions with subject matter experts and connect with peers facing similar challenges on their AI journey.

Get eBook

Related posts