7th CODH Seminar
Manifold Mixup: Encouraging Meaningful On-Manifold Interpolation
as a Regularizer

We organize a talk about deep networks by Alex Lamb, who is a PhD student at the University of Montreal with Yoshua Bengio and Aaron Courville. Anyone is welcome to join the seminar.

Basic Information

Title Manifold Mixup: Encouraging Meaningful On-Manifold Interpolation as a Regularizer
Speaker Alex Lamb
Date 15:00-16:00, July 31 (Tue), 2018
Venue 1901 Meeting Room, National Institute of Informatics
Participation Anyone is welome to join. Free of charge. Registration is not required, but if you are from the outside of NII, you can make a contact with kitamoto @ nii.ac.jp about how to enter the building. Access to NII.
Presentation Manifold Mixup (Google Slides)
Abstract Deep networks often perform well on the data manifold on which they are trained, yet give incorrect (and often very confident) answers when evaluated on points from off of the training distribution. This is exemplified by the adversarial examples phenomenon but can also be seen in terms of model generalization and domain shift. We propose Manifold Mixup which encourages the network to produce more reasonable and less confident predictions at points with combinations of attributes not seen in the training set. This is accomplished by training on convex combinations of the hidden state representations of data samples. Using this method, we demonstrate improved semi-supervised learning, learning with limited labeled data, and robustness to adversarial examples. Manifold Mixup requires no (significant) additional computation. Analytical experiments on both real data and synthetic data directly support our hypothesis for why the Manifold Mixup method improves results.
Bio Alex Lamb is a PhD student at the University of Montreal with Yoshua Bengio and Aaron Courville. He is interested in making deep learning more robust, general, and useful in realistic learning scenarios. Previously he worked on the demand forecasting group at Amazon and did his undergraduate at the Johns Hopkins University. He also creates machine learning tutorials and guides on his youtube channel: https://www.youtube.com/c/TheNuttyNetterAlexLamb

Past CODH Seminars

2018-07-31

7th CODH Seminar - Manifold Mixup: Encouraging Meaningful On-Manifold Interpolation as a Regularizer

2018-03-12

Sixth CODH Seminar: Historical Big Data - Challenges in Transforming Historical Documents to Structured Data for the Integrated Analysis of Records in the Past -

2017-12-04

Fifth CODH Seminar: Trustworthy Data Repositories - Forum for Sharing Practical Information about CoreTrustSeal Certification -

2017-07-27

Fourth CODH Seminar: A New Trend on Image Delivery in Digital Archives - IIIF's Potential for Standardization and Sophistication of Image Access -

2017-05-30

Third CODH Seminar: Usage of DOI for Humanities - Assignment of DOI for Scholarly Resources such as Research Data and Museum Collections -

2017-02-10

Second CODH Seminar: Old Japanese Character Challenge - Future of Machine Recognition and Human Transcription -

2017-01-23

First CODH Seminar: Big Data and Digital Humanities