Wednesday, December 1, 2021
HomeNewsSTEMUCI Professor Dr. Pierre Baldi Discusses Artificial Intelligence Applications and Deep Learning...

UCI Professor Dr. Pierre Baldi Discusses Artificial Intelligence Applications and Deep Learning in New Book

- advertisement -

Dr. Pierre Baldi, a UCI distinguished professor of the School of Information and Computer Sciences, describes the evolving field of deep learning in his new book “Deep Learning in Science.” 

Baldi begins his book by establishing its main idea: intelligence. While many may have trouble understanding the meaning of intelligence, taking a broader approach towards the subject allows for a clearer distinction between how it is defined and applied to both brains and machines. Following this section, Baldi introduces carbon-based and silicon-based computing, systems in mechanical devices made up of carbon and silicon material. Deep learning, which is the ability to analyze data that mimics the functions of the human brain, has a long and developed history, which Baldi also briefly covers. He then goes on to explain the significant connection between deep learning and neuroscience. 

While focusing on these developmental factors of deep learning, Baldi encourages readers to reassess their conceptions of deep learning by asking scientific questions. Baldi provides a guide of exercises throughout his book so that readers can more easily absorb basic concepts. Through the rest of the book, Baldi applies such principles to physics, chemistry, biology and medicine. 

As a researcher at UCI’s Center for Machine Learning and Intelligent Systems, Baldi focuses his work on AI and machine learning. He is driven by his interest in understanding both natural and artificial. When asked about his research work in these fields, Baldi stated that much of his new book focuses on the work he has already done at UCI.

The New University had the chance to sit down with Baldi and discuss “Deep Learning in Science.” The discussion ranged from the definition of deep learning to his own research on the subject.

New University: At the beginning of this book, you pose the following research goal: “to understand intelligence in brains and machines.” Does this mean that a distinction of intelligence can be formed between both, in which they may be compared or contrasted? 

Baldi: “It is not a simple question because we don’t truly know what intelligence is. It is not something very well-defined. We are only at the beginning of understanding intelligence in many ways, so we don’t know how many different types of intelligence there are. It is very possible that there are different forms of achieving intelligence. The way in which it is achieved, like in computers, may have similarities and also differences from the way in which biology achieves intelligence in the human brain. As a researcher, one of the questions I ask myself every day is, ‘should I try to copy the brain and what we know about it, or should I go in a completely different direction?’ This is a recurrent theme for those of us who do research in this area.” 

New University: A fundamental part of this work is taking a dive into the concept of deep learning. What exactly is deep learning, for those who may not know?

Baldi: “Deep learning is a rebranding of what used to be called neurometrics. We can use either term. It is the same thing. Neurometrics are networks of very simple model neurons that we use when we do all kinds of experiments, simulations and learning. We want machines and computers to learn from data, so we use these neurometrics that are inspired by our brains. There are networks of small neurons that are connected by synaptic weights, which represent the strength of the connection between any two neurons. Learning, in this way, is the process of tweaking these weights. We change the weights so that the neurons communicate in different ways and are able to implement different functions.” 

New University: You discuss applications of deep learning in areas of the natural sciences. What were some of the most interesting findings in relation to these applications? 

Baldi: “I use it in the natural sciences. However, what is a little surprising for someone who doesn’t know the field is that deep learning can be applied so broadly. I work with physicists, chemists, biologists and medical doctors. I am not an expert of all of these disciplines, but how is it possible to work with all of these professionals? The best analogy I can give you is that deep learning is like linear regression: drawing lines with many points. If you can do linear regression, you can apply those techniques into any area since it does not matter. I can work with experts in a specific field and help understand and analyze the data they acquire. We have made many discoveries. One example [of] using these techniques is improving how we use particle colliders in high energy physics. Protons collide at a very high speed, resulting in a shower of the rise of particles. Complex detectors detect these signals, and you want to make sense of the signals to detect if a particular particle was reduced in the collision. Deep learning can be applied to these problems, showing that we can improve the accuracy of this detection by a very significant percentage. It is now used everywhere.”

New University: How do these findings connect to your own research for UCI’s Center for Machine Learning and Intelligent Systems? 

Baldi: “To be quite honest, most of the book comes from my own research. The first part of the book is about the mathematical aspect of deep learning, where I present some of the results I have contributed and the results others also have. I try to give an overview. The chapters of application, to a larger extent, showcase applications that were done in my research group. I had to make a choice and decided to focus primarily on the things I knew best.”

New University: Finally, what are some current or upcoming projects you can share?

Baldi: “My research is divided into theory and application. On the application side, I am currently working with chemists to use deep learning to predict chemical reactions. We are working our way towards being able to apply this to atmospheric chemistry, for example, in order to assess problems like global warming. On the theory side, I work on that myself, but do so with mathematicians at UCI to understand why deep learning works so well. I am working to understand how we move forward to build systems that are more intelligent than the ones we have today.”

To read more about Dr. Pierre Baldi and his work in deep learning, visit UCI’s Institute for Genomics and Bioinformatics website

Korintia Espinoza is a STEM Intern for the fall 2021 quarter. She can be reached at korintie@uci.edu.