Afroze Baqapuri

Working on algorithms for solving a series of synthetic computer vision problems which focus on visual reasoning, and finding global spatial relationships between components of the image. For more details on the problem: http://www.idiap.ch/~fleuret/svrt/

Worked on sparse modeling Restricted Boltzmann Machines (RBMs). Implemented model for enhancing their scalability to high dimensional input. Worked on multi-modal modeling of images and text using deep learning techniques. Designed an information retrieval based system for finding images relevant to a text query (image search) and finding descriptive sentences based on an image query (image annotation). Project counts towards master thesis at EPFL.

Worked on click-through rate (CTR) prediction modeling using neural networks. Technical report can be found here: http://arxiv.org/pdf/1412.6601v3.pdf Yandex is the largest search engine operating in Russian speaking countries.

Worked on applications of deep learning in automatic speech recognition (ASR) systems. Explored deep Convolutional Neural Network (CNN) architectures for context-dependent phoneme and grapheme recognition.

Machine Learning, Artificial Intelligence

Nano Technology, German language