In solving complex visual learning tasks, adopting multiple descriptors to more precisely characterize the data has been a feasible way for improving performance. These representations are typically high dimensional and assume diverse forms. Thus finding a way to transform them into a unified space of lower dimension generally facilitates the underlying tasks, such as object recognition or clus...

Speakers routinely gesture with their hands when they talk, and those gestures often convey information not found anywhere in their speech. This information is typically not consciously accessible, yet it provides an early sign that the speaker is ready to learn a particular task (S. Goldin-Meadow, 2003). In this sense, the unwitting gestures that speakers produce reveal their implicit knowledg...

The choice of the kernel is known to be a challenging and central problem of kernel based supervised learning. Recent applications and significant amount of literature have shown that using multiple kernels (the so-called Multiple Kernel Learning (MKL)) instead of a single one can enhance the interpretability of the learned function and improve performances. However, a comparison of existing MK...

One of the central issues in kernel methods [5] is the problem of kernel selection (learning). This problem has recently received considerable attention which can range from the width parameter selection of Gaussian kernels to obtaining an optimal linear combination from a set of finite candidate kernels, see [3, 4]. In the latter case, kernel learning problem is often termed multi-kernel learn...

In this paper, we propose a novel spatio-temporal feature which is useful for feature-fusion-based action recognition with Multiple Kernel Learning (MKL). The proposed spatio-temporal feature is based on moving SURF interest points grouped by Delaunay triangulation and on their motion over time. Since this local spatio-temporal feature has different characteristics from holistic appearance feat...

We present a fast algorithm for multiple kernel learning (MKL). Our matrix multiplicative weight update (MWUMKL) algorithm is based on a well-known QCQP formulation [5]. In addition, we propose a novel fast matrix exponentiation routine for QCQPs which might be of independent interest. Our method avoids the use of commercial nonlinear solvers and scales efficiently to large data sets. 1

In this paper we formulate multiple kernel learning (MKL) as a distance metric learning (DML) problem. More specifically, we learn a linear combination of a set of base kernels by optimising two objective functions that are commonly used in distance metric learning. We first propose a global version of such an MKL via DML scheme, then a localised version. We argue that the localised version not...

The asynchronous nature of the state-of-the-art reinforcement learning algorithms such as the Asynchronous Advantage ActorCritic algorithm, makes them exceptionally suitable for CPU computations. However, given the fact that deep reinforcement learning often deals with interpreting visual information, a large part of the train and inference time is spent performing convolutions. In this work we...

Multiple Kernel Learning, or MKL, extends (kernelized) SVM by attempting to learn not only a classifier/regressor but also the best kernel for the training task, usually from a combination of existing kernel functions. Most MKL methods seek the combined kernel that performs best over every training example, sacrificing performance in some areas to seek a global optimum. Localized kernel learnin...