Thesis Project
- Image Captioning for Translated MSCoCo Dataset (March 2019 - July 2019), Supervised by Dr. Nasser Mozayani and Dr. Sauleh eetemadi
- I worked with another B.Sc student on my thesis project. We use MSCoCo dataset for image captioning and translated captions into Persion, as there is no dataset for this problem in Persian (We will publish the dataset soon). Then we use Show, Attend, Model: Neural Image Caption Generation with Visual Attention model, then we used a Tensorflow implementation with some modifications. We trained the model on Google Colaboratory GPU. You can view the Jupyter nootebook file here. (Google Colab view only link). Here is a good and a bad result:
The upper one means “an elephant in an captured area in a zoo” and the bottom one means “a little boat is flying in a cloudy day”.
Research Projects
- Research on using Transfer Learning for hasten the converge time of Reinforcement Learning for ATARI games (June 2018 - October 2018), under Dr. Nasser Mozayani’s supervision
- My first project as I have join the Machine Learning Laboratory was on Transfer Learning supervised by Dr Nasser Mozayani. I worked in a group three B.Sc students and one M.S student. The whole idea was to build a model as a state representation for ATARI games. I continued the implementation of the M.S student which was Deep Q-Learning for Breakout and Pong games and enhanced the converge time with multi-agent. Then I build an Autoencoder and Variational Autoencoder model which they were feed with images of these two games as a skilled agent were playing the games. The results were fine for state representation for new ATARI games, but the group are still working on using Video Encoding for increasing the accuracy.
- Research on Twitter, for Hashtag Recommendation(March 2019 - May 2019), under Dr. Hossein Rahmani’s supervision
- I was working with a M.S student on her Master thesis topic which is Hashtag recommendation. After the studying related works phase, I scrapped twitter data using Twint module whith Python. Unfortunately, I was moving to Edomonton, Canada so I couldn’t continue working with her but she is going to implement a word embedding and build a graph for words and hashtags in tweets and then use graph mining methods to find clusters of tweets. These clusters will be used for Hashtag recommending.
- Analysing of Persian Word Association using Cognitive Science (January 2018 - June 2018), under Dr. Hossein Rahmani’s supervision
- My second project in the Data Mining Laboratory was Persian Word Association supervised by Dr Hossein Rahmani. I worked with two B.Sc students and one M.S student on this project. We gathered a dataset with developing a survey website and analysing the results and eventually published the dataset. Here is a sample graph of Persian word associations:
- My second project in the Data Mining Laboratory was Persian Word Association supervised by Dr Hossein Rahmani. I worked with two B.Sc students and one M.S student on this project. We gathered a dataset with developing a survey website and analysing the results and eventually published the dataset. Here is a sample graph of Persian word associations:
- Fraud Detection (January 2016 - April 2016), under Dr. Hossein Rahmani’s supervision
- My first project as I have join the Data Mining Laboratory was on Fraud Detection for German Bank Dataset supervised by Dr Hossein Rahmani. The goal was to fit the best model in order to classify people into two classes of trustworthy and untrurstworthy. The best model was SVM with more than 91% accuracy and I use decision tree for visualization the model. Here is the result of SVM model:
- My first project as I have join the Data Mining Laboratory was on Fraud Detection for German Bank Dataset supervised by Dr Hossein Rahmani. The goal was to fit the best model in order to classify people into two classes of trustworthy and untrurstworthy. The best model was SVM with more than 91% accuracy and I use decision tree for visualization the model. Here is the result of SVM model:
Course Projects
- Computational Intelligence instructed by Dr. Nasser Mozayani:
- Solving Inverted Pendulum problem with Fuzzy Logic and Reinforcement Learning
- Multi Layer Perceptron for MNIST Dataset with Keras, Numpy
- Function Approximation with MLP and RBF Networks
- Genetic Algorithm for N-Queen problem
- Natural Language Processing instructed by Dr. Sauleh eetemadi:
- Machine Translation for Poetry to Prose and vice versa using OpenNMT
- Sentiment Analysis (for tweets that have #worldcup) using AFINN
- Classifying speeches of two famous persian politicians (Naive Bayes and Maximum Entropy)
- Artificial Intelligence and Expert Systems instructed by Dr. Behrouz Minaei Bidgoli:
- Face Recognition for LFW (Labeled Face in the Wild) dataset using OpenCV
- Advanced Computer Programming
- Prototype of Social Network Website with comment, like, and post abilities qith Python
- Web Scraper and Search Engine using Scrapy
- File Manager with Python