Machine Learning Meetings and Events
Group Meetings: Group meetings are held Mondays from 11am to Noon (talk starts 11:10am) in D.L. Pratt 290C unless otherwise noted. Meetings are coordinated by Hugo Larochelle.
Tea Talks: Tea talks are held every Wednesday at 4:00pm in D.L. Pratt 290C. Talks should be simple, accessible, and not exceed 15 minutes. Speakers bring snacks, make tea, and provide a copy of the presented paper.
Group Meeting Oct 1, 2008: Deep Belief Nets for Visual Object Recognition.
- Speaker: Vinod Nair
- Abstract:
I will present some results of applying Deep Belief Nets (DBNs) to a 3D visual object recognition task. The inputs are stereo-pair images from the NORB database, which contains five classes of objects imaged under different lighting conditions and viewpoints. Learning to recognize the object class is difficult because the DBN must learn a high-level image representation that is invariant to changes in lighting and viewpoint within a class while still accurately discriminating among the different classes.
Highlights of this work are:
- Preliminary classification error rates that are already close to the current state-of-the-art for NORB.
- A new kind of top-level model that learns the joint density of inputs and labels as a third-order Boltzmann machine.
- A way to regularize Restricted Boltzmann Machines during pre-training that encourages sparse hidden representations.
- Much higher dimensional inputs than previous applications of DBNs.
- Roughly two orders of magnitude more parameters than the original DBN (applied to the MNIST handwritten digit images), which forces the use of parallelization to keep the training times practical.
This is work done jointly with Geoff Hinton.
- Notes: Wednesday 4pm