Armin Shafiee

To take a look at the full record of my works, you can have my CV.

RESEARCH EXPERIENCES

Multimedia and Signal Processing Lab | Department of Electrical Engineering
Research Assistant Supervised by Prof. Arash Amini | Feb 2020 - Mar 2021
By utilizing Video Processing and Image Processing methods to extract features and deploying Deep Neural Networks, We worked on :
- Distinguishing upscaled videos from original videos for every possible resolution
- Detecting the rate of upscaling
- Detecting the software used to upscale the video

Secure Communications and Machine Learning Lab | University of Tehran
Research Intern Supervised by Prof. Hamid Behroozi and Prof. Mohammad Ali Akhaee | Jul 2019 - Sep 2019
After learning basic concepts of Deep Learning, I worked on:
- Developing a Convolutional Neural Network for classification task on Google ”Quick, Draw!” huge data-set
- Real-time painting classification by designing a software application like Google ”Quick, Draw!”[GitHub]

Design and implementation of Deep Learning models for Image Captioning task on COCO data-set, using Transfer Learning and Data Augmentation for training CNN model to extract features (while reaching 97.3% accuracy on test data-set), using pre-trained CBOW model for Word Embeddings, using Attention Layer and Bidirectional LSTM for training RNN model to generate captions.[GitHub]

Supervised by

| Fall 2019

Implementation of LSTM-Based Seq2Seq Encoder-Decoder and Skip-Gram models on a persian text data-set called ’Shahname Ferdowsi’.[GitHub]

Supervised by

| Fall 2019

Analysing ”Titanic” data-set and extract useful features by Data Analysis techniques as well as implementation and optimization of Logistic Regression, K-NN, SVM, Naive Bayes Classifier, Random Forest and Neural Network methods on the analysed data-set to achieve highest possible accuracy on classification task, reaching 92% test accuracy, graded as the best work of the course.[GitHub]

Supervised by

| Fall 2019

Implementation of K-NN classification algorithm using CUDA on a GPU, which was capable of classifying huge data-sets of up to 2³⁰ samples efficiently and fast.[GitHub]