I'm a machine learning and computer vision researcher as well as an expert Java developer. I have been actively working in the machine learning, computer vision and related areas since 2008 and I have mostly worked with massive amounts of data since then. I worked with diverse modalities of data including text information in multiple languages (poetry detection in Persian text, statistical machine translation, news search engine, website categorization), speech signals and visual information (video data, light-field images, color photographs). In all these cases, what I did was essentially organizing masses of data and summarizing them into meaningful and desirable information. This is the task I enjoy most: seeing meaningful information appearing out of masses of seemingly incomprehensible data.
The imaging technology is rapidly changing. The so-called light-field or plenoptic cameras can dramatically change the way we capture photos, both for entertainment and commercial purposes. This is mostly because of the vast amount of visual information available in the plenoptic function, compared to traditional photos.
Currently, I'm working on plenoptic image processing and computer vision. I will develop novel and efficient methods and algorithms to analyze and extract features from plenoptic images by making best use of the information inherent in them. Utilizing plenoptic information can significantly improve the accuracy and efficiency of current traditional computer vision algorithms.
For my M. Sc. studies, I worked on applications of machine learning and pattern recognition to image retrieval problems (Link in Persian). Specifically, I studied semi-supervised and active learning methods and their applicability to image retrieval and categorization. I worked on different measures for sample selection in active learning and trying to utilize active learning in improving relevance feedback efficiency in image retrieval. I proposed and implemented a novel one-class learning method which utilizes unlabeled data and can be used as a an active learning approach.
I also studied one-class learning algorithms, support estimation methods and methods for learning with few or imbalanced data because these limitations directly apply to image retrieval problem.
I’ve had experiences in different classification tasks, from document classification to mood classification and poetry detection in text data. I have done my Bachelor’s thesis on website classification which is a special classification task because of presence of links between pages and different page categories. I’ve had also experiments with different term weighting schemes for information retrieval and text classification. In addition to these I’ve worked on prediction of people’s psychological characteristics from their personal blogs’ posts.