Trading Convexity for Scalability

Channel:
Microsoft Research
Subscribers:
344,000
Published on ● Video Link: https://www.youtube.com/watch?v=qxC5Qh-AHi8


Duration: 48:26
126 views
0

Joint work with Ronan Collobert, Fabian Sinz, and Jason Weston. Convex learning algorithms, such as Support Vector Machines (SVMs), are often seen as highly desirable because they offer strong practical properties and are amenable to theoretical analysis.  However, in this work we show how non-convexity can provide scalability advantages over convexity.  We show how concave-convex programming can be applied to produce (1)~faster SVMs where training errors are no longer support vectors, and (2)~much faster Transductive SVMs.



Other Videos By Microsoft Research


2016-09-06Brain Computer Interface Systems: Progress and Opportunities
2016-09-06Need: How PowerPoint Adversely Mediates Thought and Possible Remedies [1/7]
2016-09-06Edgenet 2006 - Managing Corporate WiFi Networks Using DAIR
2016-09-06Edgenet 2006 - The Case for Comprehensive Diagnostics
2016-09-06Edgenet 2006 - The Protection Problem in Enterprise Networks
2016-09-06Edgenet 2006 - Wireless Network Measurement Challenges
2016-09-06Graph Cuts without Eigenvectors
2016-09-06Searching for Privacy [1/2]
2016-09-06Algorithms for Data Management and Migration
2016-09-06Edgenet 2006 - Customer Insight: CSO's Perspective on Network Policies
2016-09-06Trading Convexity for Scalability
2016-09-06Congressional Panel and Dialogue
2016-09-06Edgenet 2006 - Measuring and Monitoring MicrosoftΓÇÖs Enterprise Network
2016-09-06Memex Summit (Digital Memories Workshop) - Personal Audio Life Logs
2016-09-06Off the Beaten Path in Pervasive Computing
2016-09-06Memex Summit (Digital Memories Workshop)
2016-09-06CMPLE - Melting the Ice Between Cores
2016-09-06Capturing People
2016-09-06New Market Models and Algorithms
2016-09-06Introduction and Welcome
2016-09-06Data Abstraction without Control Abstraction in Software Model Checking


Tags:
microsoft research