Michael Teti

Michael Teti

Degree Program

Center for Complex Systems and Brain Sciences

Tell us about yourself:

My name is Michael Teti. I’m a third year Ph.D. student in the Center for Complex Systems and Brain Sciences. I received my bachelor’s degree from FAU in Biological Science, and during this time, I helped start the Machine Perception and Cognitive Robotics (MPCR) Lab. I like to take inspiration from my background in biology and neuroscience when developing new machine learning algorithms and applying these algorithms to different applications. After I graduate, I have plans to do a postdoc at Los Alamos National Lab in New Mexico and later become a staff scientist there. Outside of the lab, I am a big baseball fan and you will probably see me at a lot of the FAU home games.

What are you currently working on?

Detecting Deepfake Speech. A “deepfake” is fake data, such as a picture, video, or audio file, made by a deep learning algorithm that appear real to people. These are made by having an algorithm look at a bunch of real images, video, etc. and then learn to create new ones that have never existed before but appear real. For example, these algorithms are really good at making completely new pictures of people’s faces that have never existed before by learning what people look like in general and changing different characteristics. Audio on the other hand is less studied on deepfakes, but there have been recent advancements in audio and speech synthesis (i.e. making fake voices and audio snippets that sound real), many of which are able to trick smart devices into believing its owner said something they never said. These attacks are getting more advanced, and they are very dangerous in the wrong hands. I am working with the Advanced Research in Cyber Systems group at Los Alamos National Lab to try and find new ways to defend against these attacks.

What interested you choose this area of research?

This project is interesting because it is obviously an important problem, but it also allows us more insight into how deep neural networks work. For example, if we can see what is able to trick a neural network, we can learn more about how they operate and what they are learning.

What do you hope to accomplish with this project?

I hope to help further the understanding of neural networks and how to defend against these deepfakes in the future.

What advice would you give to a new student interested in working in the Gruber Sandbox?

Try to stay focused and disciplined. There are a huge amount of applications neural networks can be applied to, and many of them will seem interesting to you. This means you will need to exercise a lot of discipline and focus on what you really want to do, because you can’t do everything.