Today Gil dives into the life of Datagen’s head researcher Dr. Jonathan Laserson. Dr. Jonathan Laserson is the head of research at Datagen and an early adopter of deep learning algorithms. He has a great list of credentials. He has a bachelor’s and master’s at Technion- the Israel Institute of Technology- and earned his Ph.D. from the computer science AI lab at Stanford University. After working for a couple of years with Google, he ventured into the startup world and became deeply involved in the practical applications of ML and deep learning. Before Datagen, he worked at Zebra Medical where he led the development of two FDA-approved clinical products, applying neural networks to millions of medical images and unstructured textual reports.
Here are some excerpts from the podcast:
- “Throughout my career, the things that I was working on, they chose me. In the beginning, working on my Ph.D.- I hated computer vision. I started a computer vision project with a senior student of my advisor Daphne Koller. The project didn’t go well. But it did relate to another project of one of Daphne’s postdocs in computational biology. So he snatched me from the computer vision project and I was very happy to try something else. The next time I got into computer vision was only after Google, at PointGrab. I didn’t see myself as a computer vision person before deep learning came along.”
- “I joined Datagen because I think this is the right time and the right place to be during the revolution that is happening right now in the Deep Learning and Graphics worlds. That revolution is the revolution of photorealism and implicit representation (NeRF, SDFs, and all that jazz). This is a completely radical difference from the classic graphics approach that has all these weird data structures that represent objects (like meshes, and triangles, surfaces and normals). I wanted to be in a place where I can be part of this new movement, this revolution. I looked around and saw Datagen generating this amazing photorealistic synthetic data but doing it using these classic tools. I thought that this makes Datagen ready to make this technological jump towards the new Deep Learning breakthrough that will make our data even more photorealistic.”
- “One of the breakthroughs that make this jump to photorealism is the new representation that kind of goes into the physical level. It’s like you suddenly stop looking at surfaces and objects and instead look at the atoms. There is no notion of a surface. There is no notion of an object. There is just a point in space (x, y, z), and whatever there is in that particular (x, y, z). If you think about it, this is how the real world works. When you look at the scene in the real world, there’s nothing that tells you when one project starts and where one ends. And this is also what makes it more photorealistic.”
Photo credit: David Garb