Francisco Martin, Ph.D . CEO, BigML, Inc
Francisco is a co-founder and CEO at BigML, Inc, where he helps conceptualize, design, architect, and implement BigML’s distributed Machine Learning platform since 2011. Formerly, Francisco founded and led Strands, Inc, a company that pioneered Behavior-based Recommender Systems. Previously, he founded and led Intelligent Software Components, SA (iSOCO), the first spin-off of the Spanish National Research Council (CSIC).
He holds a 5-year degree in Computer Science from the Technical University of Valencia, a Ph.D. in Artificial Intelligence from the Technical University of Catalonia, and a post-doc in Machine Learning at Oregon State University. He is the holder of 20+ patents in the areas of Recommender Systems and Distributed Machine Learning.
Interview to Francisco Martín (CEO at BigML):
Q: What led you to dedicate professionally to this area of research?
A: When I was studying the fifth course of computer science at the UPV, Artificial Intelligence was one of the optional subjects and I was deeply captivated because it was the most advanced programming. I was “lucky” that the Higher Council for Scientific Research gave me an introductory scholarship to research and invited me to spend a few months at the Research Institute on Artificial Intelligence in Bellaterra.
I put luck in quotation marks because in fact I had worked a lot to have a good academic record, I finished with an average score of 9.3 after 5 years of degree. With honors in subjects that many students took several years to pass and of which I am sure many students today would have problems to overcome. All that effort and work that I devoted to strengthen in a deep way the concepts of computer science have served me for many years to solve many problems.
About my comment of current students, I am struck by how the recent graduates limped in subjects that were taught in such fundamental subjects as algorithmic, theoretical informatics, discrete mathematics, etc. and they are hardly prepared to solve problems of a certain technical complexity.
Q: You have a Postdoc in Machine Learning by the Oregon State University … Considering this training background, what technique of Machine Learning do you like more? Why?
A: In principle no specific technique, but the one that best solves the problem in question. After all, Machine Learning offers a toolbox and whoever uses it must know what tools to use and how to combine them to solve a particular problem. Anyway, I’m a fan of Leo Breiman and all his inventions CART, Bagging, or Random Forest.
Leo Breiman was a statistician who revolutionized statistics thanks to his passage of more than a decade by the company. It is something that should be promoted in Spain, where people usually follow a pure academic life. There should be a way for teachers to spend long periods of one or more years in the company, so that they could experience real-world problems first-hand, so that they could then focus their academic life in two different ways: focus on research new techniques to solve real world problems and teach students concepts and techniques closer to reality.
Q: And continuing with Machine Learning… Do you think Google will monopolize this technology?
A: I think he is trying at least. It accumulates much of the best talent worldwide and has the data and the ability to apply it internally and optimize its operations to be even more competitive. Any self-respecting company should try to imitate it. I am sorry that the asymmetry already existing between companies such as Google and large Spanish companies is even greater. Time passes and no Spanish company has made the slightest movement to take Machine Learning seriously and become a producer and not a mere consumer of foreign technology.
Q: You are currently CEO of BigML, could you tell us briefly what a typical day in your work is like?
A: Sometimes I introduce myself saying that the “E” of CEO is for “Everything”. And in these companies with such a flat structure and such an incipient state the CEO does absolutely everything. In the first four years of BigML I was much more involved in the development. I rescheduled continuously after many years and had a lot of fun. Later, as the company has grown, I have focused more on the general design of the product and more recently on everything related to its use.
On an ideal day, I usually get up at 5am. I spend the first hour of the day reading. After 6am to 8am I reserve it for meetings with Europe and from 8am to 1pm for meetings with clients in the US. I try to take a nap every day for at least 30 minutes. The afternoon I dedicate to review all the email and produce everything that touches me personally: presentations, proposals, etc. I try to go to bed before 11pm but it is not always possible. However, most days are unexpected with a multitude of tasks that we have to deliver now or to which we must respond immediately.
Many people are stressed in situations of this type since they can not parallel so much burden. I worked as a bartender at 18 and I think there is nothing like having to attend many tables and many clients to exercise the ability to do tasks in parallel. To change the topic many times I play 5 minutes of chess.
Q: What makes BigML special?
A: We have been able to combine a team that has designed and implemented a complex product in a very elegant and robust way. Many companies that have tried the same, either have been filled with scientists, or they did not have enough. We have combined profiles that understand very well the most basic Machine Learning with profiles with extensive experience in software engineering, other profiles that dominate the infrastructure and other experts in product design. All the combination is what has given magic to BigML to create a platform that not only lowers the barrier of entry to Machine Learning but it does so for customers to enjoy it.
Q: How do you think the irruption of chatbots and processing of natural language (NLP) is transforming the digital business?
A: I think there is still a lot to do. Everything I know in this regard is very basic. While the results in speech-to-text or text-to-speech of the new attendees such as Alexa, Siri, or Google Home are spectacular, greater progress is still needed for an early child not to discover that the technologies to understand our language leaves much to be desired. The level of conversation that chatbots currently allow is very poor. Commercially works well because who trains to use them is the person and not the machine. I think in about five years we will see inventions in this field that are much more advanced.
Q: Can you tell us a couple of names of people that you think have made (or are making) a significant contribution in the Machine Learning area?
A: Leo Breiman mentioned before and of course Tom Dietterich. I was lucky enough to meet him in 1998 when he had been leading the Machine Learning area for 18 years. I was able to work with Tom Dietterich in 2003 and we co-created two companies together. Tom is undoubtedly the brightest person I have ever met, also the most professional and the one who has most contributed to the Machine Learning community being serious, robust and open.