Madiou Diene

As a deep learning engineer, I am currently working on computer vision. I have just finish a four months internship at Baamtu, a senegalese company which works very deeply for solving artificial intelligence problems. Some of the most exciting thing on which I was working at Baamtu was to use both machine learning and deep learning technics in order to extract personal information in some cards like national identity card, permit, passport etc. There, at Baamtu, I have also set up a classifier model which is able to recognize images. The based model we use is sam(sharpness award minimization) which was the stat-of-the-art model in images recognition in December 2020. The model allows us to recognize an input image with its classes among our four classes: passport, visas, permit, national identity card. We use Microsoft azure to train our deep learning models. Before I went to Baamtu, I have completed the prestigious master ammi(African Master in Machine Intelligence) which is sponsored by Facebook and Google. At ammi, to complete the program I had to work in a final project and write a short paper based on that project. For me, to complete that program I have chosen to work on nlp(Natural Language Processing) and especially on chatbot. My goal was to build a conversational agent(chatbot) in open domain chatbot capable of speaking french. We were excited by the fact that almost all the open domain chatbot which are done, are in english language and we wanted to use some technics to show that we can set up an open domain chatbot which is able to speak french.

During our master program, I have also work on many projects in basic machine learning as well as in deep learning. For those projects I want to cite two of them. The is when we were classify DNA sequence data on what we did binary sequence classification task by using soft margin svm. The other is an assignment on which our teacher gave us a Kaggle competition problem. It was about Cassava Disease Classification on what we classified pictures of cassava leaves into 1 of 5 disease categories (or healthy). I lead the problem by using Resnet101 and the concept of transfer learning knowing that there are high similarity between images from two different classes. That requires a very deep neural network that is the reason why Resnet101 was a very good choicep learning model and I finished as one of the best for that competition regarding my classmates.