Research activities

My researches are focused on developing a machine learning tools to analysis complex remote sensing images for environmental monitoring applications. In particular, I am interested to develop methods which facilitates the information transfer from one source domain to another (but related) domain, where there is no label information. I am also working on developing unsupervised feature learning for large scale data sets using kernel approximation methods (random Fourier features).

Kernel Approximation for large scale machine learning

The complexity of Kernel method grows with number of observations in the data, thus making it impossible to apply for large data sets. In this work, we are exploring the ways to scale the kernel methods for large scale machine learning classification. Recently, Random Fourier Features has shown the potential to scale kernel methods for large scale data sets in achieving comparable results with deep learning. In this work, we develop the methods to generate explicit feature maps which approximates the kernel matrix in fewer feature expansions than the traditional Random Fourier Features, and non-linear dimensionality reduction methods using random Fourier features.

Optimal transport for domain adaptation applications

I am interested in developing novel concepts, methodologies, and new tools for solving the domain adaption problem by leveraging on the theory of optimal transport. I will focus on both the theoretical aspects of the problem and its practical implementation in deep neural nets architecture. On the application side, a particular interest will be given to analysis of remote sensing images (RSI), for which transfer learning is a fundamental problem. Hence, in the context of RSI, the drifts observed in the probability density function (PDF) of different images acquired by different captors, at various locations and at different time are due to a variety of factors: different corrections from atmosphere scattering compensation, daylight conditions at the hour of acquisition or even slight changes in the chemical composition of the materials. This also leads to interesting and almost unexplored variants of the transfer learning problem, where the data can live in different spaces and at different resolutions if they are produced by different captors (heterogeneous domain adaptation)