大岡山情報統計力学セミナー 10/11(金) (Speaker: Xiangming Meng)

[小渕智之]

IBISMLメーリングリストメンバーの皆様

東京工業大学 情報理工学院 数理・計算科学系の小渕智之です。
下記セミナーを企画しました。奮って御参加下さい。
お近くにご興味のありそうな方がおられましたら転送して頂ければ幸いです。
宜しくお願い致します。
The talk will be given in English.
--------------------------------------------------------------------------------------------
Speaker: Xiangming Meng (理研AIP)
Schedule: 10/11(金)  15:05~(16:35)
場所：東工大大岡山キャンパス 西８号館-W1008 (Ookayama-campus, W8-W1008, Tokyo Tech.)

Title: A High-bias Low-variance Introduction to Approximate Bayesian Inference

Abstract:
A variety of fundamental problems in information theory, computer
science and statistical physics could be formulated as Bayesian
inference. However, exact Bayesian inference is usually intractable in
practical applications due to the curse of dimensionality. This talk
is a brief introduction to various approximate inference methods with
a particular focus on the expectation propagation (EP) algorithm.
Specifically, we first introduce the variational inference framework
and draw an analogy between the fields of computer science and
statistical physics. Then, a tutorial introduction to expectation
propagation is given using one toy example, along with some
comparisons with belief propagation. Finally, a unified EP perspective
on approximate message passing(AMP) as well as its extensions such as
vector AMP (VAMP) and generalized AMP (GAMP) is briefly illustrated.
--------------------------------------------------------------------------------------------