AutoRec: Autoencoders meet Neighborhood-based Collaborative Filtering

Share
Share to Google+Share to FacebookShare to TwitterShare to LinkedIn
Scientific Seminar - Los Altos, CA
October 13, 2014 - October 14, 2014
Find out more about my novel Autoencoder neighborhood-based CF (AutoRec) model trained with a standard back-propagation approach.
Innovation Seminar In Los Altos Technicolor Lab
 

Open to the public. 
 

Location: Technicolor, 175 S San Antonio Road, Los Altos, CA 94022

MapGoogle Maps (opens in new window)

Date: Monday October, 13th 2014 at 11:00 AM – 12:00 PM PT

 

SpeakerSuvash Sedhain, INRIA

 
Title: AutoRec: Autoencoders meet Neighborhood-based Collaborative Filtering
 
 
 
Abstract:
In this talk, I will give an introduction to Restricted Boltzmann Machine (RBM) based collaborative filtering (CF), which can be seen as a foundation for deep learning in recommendation.
 
Then I will discuss various shortcomings of this model including its failure to incorporate valuable user-user/item-item neighborhood information (similarities).
 
To alleviate these issues, I propose a novel Autoencoder neighborhood-based CF (AutoRec) model trained with a standard back-propagation approach.  
 
Experiments on the Movielens dataset shows significant improvement of AutoRec vs. a variety of state-of-the-art collaborative filtering techniques (RBM, PMF, SVD++, KNN), suggesting the importance of further investigating autoencoding approaches to recommendation. 
 
 
 
For more information about Technicolor Seminar Series: http://www.technicolorbayarea.com/#contactArea