Mittagsseminar (in cooperation with J. Lengler, A. Steger, and D. Steurer)

Date and Time: Tuesday, October 02, 2012, 12:15 pm

Duration: 30 minutes

Location: OAT S15/S16/S17

Speaker: Jan Hazla

Concentration Bounds using Bayes' Rule

Concentration bounds limit the probability of a random variable deviating too much from a certain value (usually its expectation).

Let us have a look at a simple version of the so called Chernoff bound:
Let A1, A2, ... be independent events with Pr[A_i] = 1/2. Let k be a number between 0 and n/2 and let S(n, k) be the event that at least n/2 + k of the events A_1, ..., A_n occur. We would like to show that Pr[S(n,k)] \leq exp(-Omega(k^2/n)).

The standard proof optimizes certain exponential function of the number of events occurring and applies Markov's inequality.

We propose a different approach: one can use Bayes' rule to bound Pr[S(n,k)] with exp(-k/n) * Pr[S(n-1,k-1)] and apply the induction to finish the proof.

We will present some generalisations and applications of this technique. In particular, we will show how the independence assumption can be relaxed while still obtaining a good bound. We will also sketch how to apply the technique to the problem of counting occurences of a certain subgraph (e.g., a triangle) in a random graph.

Upcoming talks     |     All previous talks     |     Talks by speaker     |     Upcoming talks in iCal format (beta version!)

Previous talks by year:   2025  2024  2023  2022  2021  2020  2019  2018  2017  2016  2015  2014  2013  2012  2011  2010  2009  2008  2007  2006  2005  2004  2003  2002  2001  2000  1999  1998  1997  1996  

Information for students and suggested topics for student talks

Automatic MiSe System Software Version 1.4803M   |   admin login