Hans Peter Graf(PhD in Physics/Department Head, Machine Learning/NEC Laboratories America)

April 18, 2019

Machine learning that is going to leave significant impact on human society

I am the department head of the machine learning research department of NEC Laboratories America. Here at NEC, we focus on machine learning algorithms, as well as applications in a wide range of artificial intelligence, and data analytics.
I have been working in machine learning for over 25 years, and my expertise is in the implementation of machine learning algorithms and technologies to bring them to a stage where they can solve practical problems. In fact this technology has the potential for tremendous impact on human society.
One of the practical applications of our research is autonomous driving. You see many cars driving by themselves, and artificial intelligence and machine learning are the key technologies behind it.
In manufacturing, artificial intelligence helps automate the process. As a result we use less energy, less resources, and can even produce products of better quality. As the human population grows it is very important to produce goods more efficiently with less impact on the environment, and artificial intelligence helps achieve that goal.
The main challenge in research we face today is that machine learning is still a far cry from the way humans think. The human brain is actually extremely good at doing higher-level reasoning and intelligent combination of information, while the machines are still not good at these tasks. Our goal as researchers at NEC is first to understand better how the humans are doing it, and then to apply the knowledge to the machines we develop.

Emphasis on academic research and network within the community, as a pioneer of the machine learning field

NEC is a pioneer in the machine learning field, and has made tremendous contributions to the research in this field. Machine learning has been pursued for over 15 years at NEC, long before the latest boom in artificial intelligence. We actually have produced some of the first machine learning development environments, in particular Torch, the widely used machine learning platform, that has come out of our department. We released that into Open Source in 2006, and for several years our department was the main contributor to this system.
We can say that some of the most prolific researchers in this field have been part of our department. For example, let's take Yann LeCun . He is now a professor at New York University, and he is also heading the AI research at Facebook. And there is Vladimir Vapnik, the inventor of the support vector machine, who was a member of our department for many years.
We also keep in contact with the latest information and developments in research by having close relationships with top universities. We regularly have interns at NEC who are in their second or third year in their PhD program. That keeps us close with universities, and helps in producing publications that are accepted at the top conferences and top journals.
As researchers at NEC, solid technical knowledge, including the understanding of math and statistics is essential, but we also stress on an openness to learn about different application areas and to understand the problems in order to understand how technology or machine learning can contribute to the solution. We believe that an open mind and sharp observation are equally important for success in producing technology that is useful for society.

Machine learning that is coming close to human brain

There is no doubt that the human brain is the single most miraculous organ on the planet, yet our understanding of it is still rather rudimentary. In my 25 plus years of research in machine learning, we have tried to emulate some kinds of functions that the brain does with algorithms but our emulations are still primitive. There is a lot of research to be done to first understand the functioning of the brain better, and interpret how it could be implemented with a different technology.
The brain works with biological neurons that are really very slow elements, while our implementations work with transistors that are orders of magnitude faster. We do a very different type of processing, and the challenge is to translate the functionality of the brain into something that the machine can execute.
While the human brain is very difficult to emulate in machine learning, we have successfully imitated a few of the functions that the brain does, like recognizing objects. Nowadays, we can have a machine that can do face recognition. What it does is to take a picture of your face and then compares it with many, many other faces, and the machine identifies you against many other people.
The technology can be used to inspect manufactured parts. Let's say there are spark plugs coming down the production line. Currently we have people inspect all of the products and identify the faulty ones. By implementing our technology the machine learns how the product looks, and is able to differentiate between defective spark plugs and good ones.
Emulating the human brain functions in machine learning is a tremendously fascinating topic, and it will take more than ten years until we will have solved that problem. My passion is to bring machine learning to a practical level where it is a product that solves problems for people, and I believe that this will be a fascinating research topic for the foreseeable future.

Photo of Hans Peter Graf

Hans Peter Graf

Hans Peter Graf is head of the machine learning research department at NEC Laboratories America in Princeton. The department develops machine learning algorithms, as well as a range of applications in semantic text analysis, cognitive video interpretation and bio-medical analysis. His responsibilities include technology transfer to business units and the commercialization of research results. An example is the recently released e-Pathologist, a system assisting pathologists with the interpretation of histological samples, which contains key algorithms developed by the department.

Before joining NEC Laboratories, Hans Peter was a Distinguished Member of Technical Staff at Bell Laboratories and AT&T Laboratories where he developed neural net models and machine learning applications. Massively parallel neural net processors of his design were key parts in high-speed address readers and check processing systems.

Hans Peter received a Diploma and a PhD, both in physics, from the Swiss Federal Institute of Technology in Zurich, Switzerland. He is author or coauthor of over 100 reviewed articles and some 40 issued patents. He is a Fellow of the IEEE and a member of the American Physical Society.