Artificial Intelligence

Principled Multi-Person Pose Estimation using Column Generation and Nested Benders Decomposition

When:
Friday, September 27, 2019, 11:00am PDTiCal
Where:
10th floor conference room (1016)
This event is open to the public.
Type:
AI Seminar
Speaker:
Dr. Julian Yarkony, Verisk
Video:
https://bluejeans.com/s/s3lL0/
Description:
We present a novel approach for multi-person pose estimation (MPPE) using column generation and nested benders decomposition.  We formulate MPPE as a weighted set packing problem over the set of person hypothesis (poses) in an image where the set of poses is the power set of detections of body parts in the image.  We model the quality of a pose as a function of its members as described by a tree structured deformable part model.
 
Since we cannot enumerate the set of poses we attack inference using column generation where the pricing problem is structured as a dynamic program and dual optimal inequalities are easily computed.  We exploit structure in the dynamic program to permit efficient inference using nested Benders decomposition.  We demonstrate the effectiveness of our approach on the MPII human pose annotation benchmark data set. 
 
Dr. Julian Yarkony is an AI/ML senior scientist with the Verisk  AI team. He leads Verisk’s research on the application of combinatorial optimization and operations research methodologies to computer vision/machine learning problems.
Before joining Verisk, Dr. Yarkony was a research scientist at Experian, where he introduced Benders decomposition to computer vision and developed column generation for application computer vision. Dr. Yarkony introduced the following column generation concepts to computer vision/machine learning:  dual-optimal inequalities, subset-row inequalities, and dynamic programming–based pricing problems. Dr. Yarkony has expanded the scope and application of dual optimal inequalities. 
Dr. Yarkony received his Ph.D. in computer science from the University of California, Irvine, where he developed planar decompositions for problems in computer vision. He also completed postdoctorates at the University of California, Santa Barbara,  and the Heidelberg Collaboratory for Image Processing (HCI
« Return to Events