# Author Archives: Sebastien Bubeck

## COLT 2013/ICML 2013 videos

The videos for COLT 2013 were just published on videolectures.net. The videos for ICML 2013 are also available on TechTalks.tv.

## Two summer readings on Big Data and Deep Learning

This is the first (short) post dedicated to the Big Data program of the Simons Institute. We received from the program organizer Mike Jordan our first reading assignment which is a report published by the National Academy of Sciences on the “Frontiers in Massive … Continue reading

## ICML and isotropic position of a convex body

ICML 2013 just finished a few days ago. The presentation that inspired me the most was the invited talk by Santosh Vempala. He talked about the relations between sampling, optimization, integration, learning, and rounding. I strongly recommend Vempala’s short survey … Continue reading

## COLT (with bombs, deep learning and other exciting stuff)

The 2013 edition of COLT at Princeton just finished a few days ago and I’m happy to report that everything went smoothly! We had a sort of stressful start (at least for the organizers..) since on the day before the … Continue reading

## Embeddings of finite metric spaces in Hilbert space

In this post we discuss the following notion: Let and be two metric spaces, one says that embeds into with distortion if there exists and such that for any , We write this as . Note that if … Continue reading

## The aftermath of ORF523 and the final

It has been two weeks since my last post on the blog, and I have to admit that it felt really good to take a break from my 2-posts-per-week regime of the previous months. Now that ORF523 is over the … Continue reading

## ORF523: Optimization with bandit feedback

In this last lecture we consider the case where one can only access the function with a noisy -order oracle (see this lecture for a definition). This assumption models the ‘physical reality’ of many practical problems (on the contrary to … Continue reading

## ORF523: Acceleration by randomization for a sum of smooth and strongly convex functions

In this lecture we are interested in the following problem: where each is -smooth and strongly-convex. We denote by the condition number of these functions (which is also the condition number of the average). Using a gradient descent … Continue reading

## ORF523: Noisy oracles

Today we start the third (and last) part of our class titled: ‘Optimization and Randomness’. An important setting that we will explore is the one of optimization with noisy oracles: Noisy -order oracle: given it outputs such that . Noisy … Continue reading

## ORF523: Mirror Prox

Today I’m going to talk about a variant of Mirror Descent invented by Nemirovski in 2004 (see this paper). This new algorithm is called Mirror Prox and it is designed to minimize a smooth convex function over a compact convex … Continue reading