Machine learning is a field in computer science that focuses on computational methods to recognize patterns and “learn” to perform tasks without specific programming. When algorithms are exposed to new data, they can autonomously adapt through a wide array of advanced techniques. The ability to perform complex mathematical calculations to extremely large quantities of data is becoming a driving force in the industry, especially in the realm of behavioral biometrics.
Machine learning is also an integral part of BioCatch’s technology. With a team of Ph.D.s and engineers, BioCatch’s machine learning team is responsible for developing algorithms to automate the collection of behavioral data from users, parsing incoming data and creating highly accurate results in authenticating users based on behavioral inputs. Meet Dr. Ariel Lubelski, Senior Algorithm Developer at BioCatch, who shares his thoughts on how machine learning drives BioCatch technological leadership in the marketplace.
Q1. Tell us a little about your academic background and your transition to machine learning in behavioral biometrics?
A: My academic background is from Tel Aviv University, where I completed my Ph.D. in Chemistry in 2008. My Ph.D. work consisted of mathematical calculation, software development, computational simulation and, of course, analyzing results. The research that started with a focused question was soon broadened to investigate general properties of anomaly diffusing systems which were never investigated before, reaching fundamental physical issues of ensemble properties. My academic research required the study of different fields and the acquisition of various skills. The transition to industry was pretty smooth and easy in that sense. I must admit, however, that it took me a while to understand and adjust myself to the fact that targets in the business world differ from the types of objectives one strives for in academia. I worked for three companies prior to joining BioCatch. In that sense, I feel that I came here with enough business experience to pursue my next challenge.
Q2. What important lessons or practices guide you in your role on the machine learning team at BioCatch?
A: The first good practice I carry from the previous companies I've worked for is to always understand where and how the current project that I work on fits into BioCatch needs and how it is going to improve our product. Having an overall picture and understanding of the company "ecosystem" is extremely important. Every new and innovative idea that arises, must first be considered in terms of how it would fit into the company and market needs. Another important lesson I've learned is that one should progress step by step when facing a new project, rather than attempting to conquer the entire project in one huge step. When one advances in a project, especially a research project, many new questions and issues can arise that were not known or were not considered earlier on. Particularly, in the field of machine learning, it is always a good practice to perform a "quick and dirty" study in order to figure out if new ideas can actually be of any use and should turn into a new project.
Q3. Research is an integral part of your position. Tell us a little about that.
A: I'm looking at many aspects of user behavior in front of a device. We obtain many features of that behavior and try to distinguish one user from another when entering a bank account or when using some purchasing application online. With all of our very large deployments, BioCatch currently obtains an immense amount of data, so we are able to conduct large scale examination of different user behaviors and continually improve our algorithm.
Q4. Could you tell us more about what you and the machine learning team do?
A: In general, machine learning algorithms’ main advantage is their ability to classify features into specified categories. In our case, the machine learning team tries to automatically learn user behavior features for profiling genuine users against fraudsters. We use various algorithms and methods to do this. The target is always to enhance detection while restricting ourselves to a very low rate of false positives. We incorporate business-orientated thinking even during the very early stages of examination of new algorithms and methods. We always keep our customers’ needs in mind and check our performance on both in "vitro" and in "vivo" scenarios. New ideas are always validated on data coming from the real world in order to make sure it will work flawlessly in production. We work on both web and mobile platforms and focus on learning features applied to our immense data sets.
Q5. Where do you think behavioral biometrics is going in the next five years?
A: The world of machine learning and deep learning is growing so fast these days. While newer and stronger algorithms continue to emerge and computation gets faster, it seems that there is always room to push the envelope and make even more progress. I would guess that behavioral biometrics and machine learning will become ubiquitous in the current use cases that we are pursuing today and will enable new use cases in the future. Online banking and e-commerce platforms are a haven for fraudsters to try new things and there is a constant need to follow their trends and to stay one step ahead. My goal is to apply machine learning to automatically recognize and respond to these emerging threats.
Dr. Ariel Lubelski is a Senior Algorithm Developer and member of the machine learning team at BioCatch. He holds a Ph.D. in Chemical Physics from the Tel Aviv University.