Research Assistant. Advised by Prof. Bernard Ghanem
King Abdullah University of Science and Technology, Saudi Arabia
Aug 2017 — Present
- Implementing actor-supervision based algorithm for spatiotemporal actional localization in video data. The system achieves state-of-the-art performance on weakly-supervised action localization tasks on a number of standard benchmarks.
- Developing benchmark dataset which contributes to a novel framework for robustness analysis of Visual Question Answering (VQA) models.
- Developing algorithms for fine-grained fish detection in underwater camera footage.
Research Assistant. Advised by Prof. Thi-Oanh Nguyen
Hanoi University of Science and Technology, Vietnam
Sep 2015 — Jun 2016
- Implemented image classification systems using convolutional neural network.
- Evaluated several neural image captioning models and propose alternatives for the vision module and optimization algorithm of an image captioning model which is composed of a convolutional and a long short-term memory network. Our proposal achivend the same perfomance with well-known models while being less complex in terms of size and faster to train on commodity hardware.
Research Assistant. Advised by Prof. Stéphane Bressan
National University of Singapore, Singapore
Jun 2015 — Sep 2015
- Designed and implemented a data crawling and wrangling system to collect maritime data (particularly vessel images and meta data) from various sources in the Internet.
- Proposed, implemented and evaluated a deep convolutional neural network for vessel images classiﬁcation. Our model is half-size compared to well-known convolutional networks and achieves a top-1 accuracy of 95.31% when classifying natural vessel images into 35 classes.
Jun 2016 — Jan 2017
- Designed and developed a content management system for Airpoli language learning app using Python Django, PostgreSQL, Redis, ReactJS.
- Designed and developed natural language processing modules for analyzing and annotating the content of the app using NTLK.
Jan 2014 — Sep 2014
- Developed tax invoices processi ng modules for National Income Tax System using plSQL on Oracle database.
- Deployed the modules to more than 200 tax offices across the country.