Nour

Nour

Machine Learning | Deep Learning

About

I am a Deep/Machine learning engineer, I am interested in building applications that have impact on our real-world using the knowledge I have in NLP, Computer Vision and self-driving cars. The more challenges i face, the more passionate and eager i get to solve them.

Currently, I am pursuing my master’s degree at Bielefeld university in intelligent systems. I am working now on two interesting projects:

  • Detecting Depression in social media.
  • Monocular visual odometry For Duckie Town competition.

Experience

Student Research Assistant at Fraunhofer institute (Germany) (june 2018 - september 2019)

Worked on different research and production tasks including:

  • Evaluation and implementation of different Fuzzy string Matching Procedures as well as Natural language processing procedures to compare between them.
  • Development of graphical java swing user interface for control and manipulating UML/SysML models.
  • Participating in research activities and presenting the research results in events as well being a teaching assistant in a company-university projects in context of Model based system engineering.

Intern at Microsoft AppFactory (Egypt) (October 2016 - June 2017)

  • Meeting Startups in events, introducing different Microsoft solutions and technologies to them like Azure, Xamarin, Hololens, etc.
  • Representing Microsoft in their Events and Booths.
  • Introducing Microsoft Hololens in one of the biggest conferences in egypt (Cairo ICT).

University’s projects

Depression Detection[github]

  • Dataset used from early risk competition in which they have collected data from users in social media especially Reddit in order to detect depression[link]
  • LSTM , BERT Models used in training.
  • Skorch used also to try cross validation.
  • Pytorch library used and Spacy used for tokanization.

Monocular visual odometry for Duckie Town competition(in progress)

  • Detecting features from an image and calculate the optical flow between two frames.
  • create a neural network architecture to calculate rotation and displacement instead of using openCV library (Research area still in progress).
  • Pytorch and opencv are used in this project.

Nod detection using SVM

  • SVM model to classify wether there is a nod or not(detect Nod),from annotated videos with nods.
  • Dlib machine learning library used and C++ used.

Manipulating robot arm (KUKA) using VR Controller (ROS C++)

  • Controlling KUKA arm to imitate the same motion as it is receiving from the VR Controller .
  • Using ROS (Robotic operating system) to receive the current pose of the KUKA arm and the VR controller to calculate the twist command.
  • Using OROCOS components to control KUKA in real time.

Udacity’s deep learning nanodegree projects

Face generation[github]

  • Generate images of new and realistic human faces using CelebA dataset.
  • DCGAN(Deep Convolutional Generative Adversarial Networks) used for training.
  • Pytorch library used.

Dermatologist-ai[github]

  • Create a model to predict and diagnose skin lesion images as one of three different skin diseases (melanoma, nevus, or seborrheic keratosis).
  • Created normal CNN architecture and transfer learning also used, using pretrained model(ResNet and Inception).
  • Pytorch used

TV script generator[github]

  • Generate a new fake TV script from Seinfeld dataset.
  • RNN model used for training.
  • Pytorch library used.