In modern computer vision, there's a significant shift from geometry to machine learning, where algorithms learn to recognize and segment objects from vast amounts of image data.

My recent work involves starting with handcrafted models and parameterizing them in a simplified way, rather than using millions of parameters. I learn parameters from actual examples that I want my handcrafted model to produce, which is known as parameter estimation or optimization.

In 1993, the state of computer vision was primarily focused on geometry, lacking the machine learning techniques that are prevalent today. In the early 90s, Bayesian statistics began to influence computer vision, enabling better tracking of objects in images and identifying the most probable paths, especially in crowded environments.

The introduction of large neural networks has allowed us to solve complex problems in vision, speech, and language, shifting focus from traditional handcrafted approaches to machine learning.

I've been really interested in machine learning, particularly its creative applications, and how it can transform the economy and interact with words, pictures, and sounds.

I think the creative applications of machine learning are fascinating and present opportunities for artists to explore new forms of expression.

The principle 'bigger model, more data, better results' has held true for the last 12 to 15 years in AI development.

For those interested in image processing, I recommend starting with foundational books and exploring recent research from institutions like UCLA.

Machine learning emphasizes the task of making machines learn from little initial knowledge to improve over time, with deep learning techniques utilizing neural networks being particularly effective in recent years.