Characteristic Kernels on Structured Domains Excel in Robotics and Human Action Recognition

Abstract

Abstract. Embedding probability distributions into a sufficiently rich (characteristic) reproducing kernel Hilbert space enables us to take higher order statistics into account. Characterization also retains effective statistical relation between inputs and outputs in regression and classification. Recent works established conditions for characteristic kernels on groups and semigroups. Here we study characteristic kernels on periodic domains, rotation matrices, and histograms. Such structured domains are relevant for homogeneity testing, forward kinematics, forward dynamics, inverse dynamics, etc. Our kernel-based methods with tailored characteristic kernels outperform previous methods on robotics problems and also on a widely used benchmark for recognition of human actions in videos.

Slow feature action prototypes effect assessment in mechanism for recognition of biological movement ventral stream

“…The modified optimisation problem for the high-dimensional visual input utilises the information of biological movements and the human object through an ABM as a Gabor-based kernel. This pathway information is combined with features of optical flow in the motion pathway with respect to the original model (Giese and Poggio, 2003;Danafar et al, 2010aDanafar et al, , 2010b.…”

“…The proposed approach ABM (Wu et al, 2010) using spatiotemporal features intermediate complexity simulates the form pathway and optical flow (Liu, 2009) represents motion pathway for stimulating the area MST and MT in dorsal stream where neurons have significant location and unchangeable scale. The obtained information required to process applying two-fold synergetic neural network (SNN) (Riesenhuber and Poggio, 2000;Riesenhuber and Poggio, 2002) or slow feature analysis (SFA) Danafar et al, 2010b) regarding predefined patterns which represents the learning procedure. These patterns attain prototypes outcomes from ABM.…”

Higher-Order Regularized Kernel Canonical Correlation Analysis

“…9 By the proposed method, we extract a two-dimensional subspace using training and test set for all scenarios i.e. X 2 R 408Â384 000 .…”

Kinetic Morphogenesis of a Multilayer Perceptron

“…In comparison to the many new strategies considered to cope with difficult learning problems, such as kernel methods (Danafar et al, 2010), restricted Boltzmann machines (Hinton et al, 2006) and so on, we assume that most of the goals they pursue are implicitly resumed by the components of our extended Hamiltonian in terms of: i) emerging structures among data, ii) feature selection, iii) grouping of specialized branches of the network, and iv) wise shaking of their state to be unstuck from local minima. The technicalities of implementing backpropagation in our framework concern simply an accurate reckoning of the derivatives of the cost function along the layers.…”