ESP Biography


Major: N/A

College/Employer: Software Engineer at Two Sigma

Year of Graduation: 2016

Picture of Arthur Ozga

Brief Biographical Sketch:

Not Available.

Past Classes

  (Clicking a class title will bring you to the course's section of the corresponding course catalog)

M13994: Generalization Bounds in Machine Learning in HSSP Summer 2020 (Jul. 11, 2020)
This course will ask the question, "why do we think that the sun will rise tomorrow?". After all, we've only seen the sun rise finitely many times. Why **do** we expect the pattern that "the sun rises each day" to continue? Why **don't** we expect the pattern that "the sun rises on and only on days when the human population is less than 8 billion" to continue? More generally, as humans and scientists, we find patterns that match our meager observations, then we guess that those patterns will continue to hold. When we interpret the word "pattern" with sufficient generality, there are lots of mutually inconsistent patterns that explain any given data. Addressing this ambiguity, we typically restrict to small classes of "simple patterns". This preference for simplicity is called Occam's Razor. We'll investigate Occam's Razor from a statistical point of view. Topics we'll discuss may include: VC dimension, information-theoretic bounds, Dudley chaining, concentration inequalities.