Sign up for our newsletter and get the latest big data news and analysis.

Interview: Dr. Roger Brooks, Chief Scientist at Guavus

I recently caught up with Dr. Roger Brooks, Chief Scientist at Guavus, to get some insights into how his company is positioned in the marketplace in terms of solutions based on AI and machine learning. Guavus is a big data analytics company developing apps that allow companies to embed data driven-decisions into agile businesses processes. Dr. Roger Brooks is 20-year veteran in the software industry and a leading expert in information analytics based on large-scale machine learning and advancing the use of data mining to help businesses strengthen IT operations, engineer new products, optimize marketing and deliver real-time personalized content. Dr. Brooks has held various executive positions including an eight-year tenure at Hewlett-Packard, most recently as Chief Architect in the new HP Software. Dr. Brooks spent the first half of his career in academia as a theoretical particle physicist at Stanford University and M.I.T., where his last position was as Associate Professor of Physics. He received his B.S. and Ph.D. degrees in physics from M.I.T.

insideBIGDATA: Our audience is finely tuned to AI and machine learning. Can you give us an overview for where Guavus is positioned in the marketplace in terms of solutions based on these technologies?

Dr. Roger Brooks: The Guavus Reflex® AI solution enables customers to gain a competitive edge using machine learning and AI to turn raw data into actionable insights, facilitating real-time decision making and improving the customer’s time to value. In our particular market segment, service providers, there is the added challenge of the massive volumes of data to be analyzed.  For example, in one of our customers we analyze petabytes of streaming data per day in a highly available manner.  It is this ability to perform novel machine learning on these high volume and high velocity data sets while delivering domain-specific and actionable results that distinguishes Guavus.

In Operational Intelligence, our applications lower costs through operational efficiencies, predict network issues and prescribe remedial actions. Our Customer Intelligence applications create new revenue and marketing opportunities by using natural language processing (NLP) to understand and anticipate end user preferences.

insideBIGDATA: You have a provocative quote on your website: “At Guavus, AI is more than just deep learning. It’s about understanding events that machine learning cannot interpret from the patterns on which it was trained.” Can you explain this perspective in more detail?

Dr. Roger Brooks: Suppose you are given a failure indicator that you have not seen before and are asked to interpret the type of failure it will lead to and what action you will take in response.  How do you handle that?  You would reason to interpret that new mode in terms of all the failures you have seen in the past to diagnose the issue and conclude the nature of this new failure.  This requires intelligent processing as opposed to pattern-based mapping. Machine learning excels at the latter but not the former.  In short, you need to create new information.  The systems with which we work aren’t kind enough to always fail in known ways, therefore our predictive maintenance solutions must exhibit the intelligence necessary to analyze and deal with new phenomena.

insideBIGDATA: “Explainability” is a hot topic in AI circles today. Can you give me a sense for how Guavus is addressing the need for an added layer of transparency where AI actions can be easily understood by humans?

Dr. Roger Brooks: When our solutions make a prediction or detect an issue, we need to annotate that insight with the supporting rationale in terms of the original data that was analyzed.  We need to be able to justify the conclusions we reach based on supporting data. As a result, we go to great pains in designing features with traceability while yielding optimal accuracy of the insights generated.

We recognized that “explainability” is a challenge for deep learning due to the fact that the rationale of how results were reached cannot easily be explained.  While the neural networks approach is optimal for cases such as speech recognition, the classes of problems Guavus is solving require supporting criteria for the conclusions and thus require that non-neural network algorithms be applied. That said, neural networks are able to handle some types of non-linearity in the data and so to address that issue, we have developed a more geometric approach which does not suffer from the aforementioned traceability issue.

insideBIGDATA: Please describe a compelling use case where Guavus technology has made a big difference.

Dr. Roger Brooks: One of Guavus’ products is focused on the optimization of the Network Operations Center (NOC) through prioritization and reduction of alarms.  As you may know, of the thousands of alarms that a NOC receives per day, only 2-3% of the alarms actually lead to true incidents or problems.  The rest are simply noise that can and should be ignored.  The problem is distinguishing the noise from the signal. Using machine learning, our solution automatically takes in streams of alarms, classifies them and predicts which alarms will lead to incidents. Now the NOC operators only need to focus on a much smaller subset of alarms and confidently ignore the rest.  In our customer results, we have seen that our solution accurately classifies and predicts alarms that will be lead to incidents and, as a result, reduces the number of alarms that NOC operators needs to pay attention to by over 90%!

insideBIGDATA: What’s out on the horizon for Guavus; what’s the next step for taking advantage of the rise in interest in AI?

Dr. Roger Brooks: This year at Guavus we are deploying predictive maintenance: predicting things that will fail, how they will fail, and identifying the root issue leading the failure. Our research organization is creating new artificial intelligence solutions for prescribing actions that can be automated or used by technicians to prevent or fix failures.  It is through this automation that the ROI for AI becomes dramatically compelling.


Sign up for the free insideBIGDATA newsletter.




Leave a Comment


Resource Links: