Suyash P. Awate Suyash P. Awate
Associate Professor
Computer Science and Engineering Department, Indian Institute of Technology (IIT) Bombay
Office: A-214, Kanwal Rekhi Building
Email: my-first-name @cse.iitb.ac.in
   
Research Publications Teaching Students CV Personal
 
Kernel Methods
 
Awate SP, Koushik NN
Robust dictionary learning on the Hilbert sphere in kernel feature space
Euro. Conf. on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) 2016 (1):731-748, Springer LNAI 9851
(podium presentation, acceptance rate 28%)

Dictionary Modeling on a Hilbert Sphere in RKHS.
(a) Points xn in input space get mapped implicitly, via several popular Mercer kernels, to yn := Φ(xn) on a Hilbert sphere in RKHS.
(b) Dictionary atoms dk, on the Hilbert sphere in RKHS, being used to fit to a point yn.
 
Awate SP, Dhar M, Kulkarni N
Robust kernel principal nested spheres (RKPNS)
IEEE Int. Conf. Pattern Recognition (ICPR) 2016, 23:402-7

Robust Kernel Principal Nested Spheres.
Input datum xi (orange) gets implicitly mapped to Φ(xi) (blue) on the unit Hilbert sphere in RKHS, centered at the origin.
A Hilbert subsphere O(v,r) (black) is parameterized by an axis v (magenta) orthogonal to itself and a geodesic distance r.
In each iteration, rkPNS
(i) fits a subsphere to the data and
(ii) projects the data (red) onto the subsphere using the minimal geodesic (green) between them.
 
Awate SP, Yu Y-Y, Whitaker RT
Kernel principal geodesic analysis (KPGA)
Euro. Conf. on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD) 2014 (1):82-98, Springer LNAI 8724
(podium presentation, acceptance rate 23.8%)

Kernel Principal Geodesic Analysis (KPGA).
(a) Points in input space get mapped, via several popular Mercer kernels, to a hypersphere or a Hilbert sphere in kernel feature space.
(b) Principal geodesic analysis on the Hilbert sphere in kernel feature space.
 
 
Related Works
 
P. Thomas Fletcher, Conglin Lu, Stephen M. Pizer, Sarang C. Joshi.
Principal geodesic analysis for the study of nonlinear statistics of shape.
IEEE Trans. Med. Imaging 23(8): 995-1005 (2004)