meziani1 Goto Github PK

-data-manipulation-

-feature-engineering

-machine-learning-explainability

autonomous-iot-drone

This repository describes the flight controller algorithm using kalman filter for accurate Gyro values

classification-of-anomalies-in-gastrointestinal-tract-through-endoscopic-imagery-with-deep-learning

We used pre-trained CNN for transfer learning to extract important features because this is relatively small dataset (only 8000 images). The pretrained models that we used in this approach are ResNet-50 and VGG-19. The proposed approach of anomalies classification is based on the steps as described below. First, we preprocessed the dataset due to some reasons as described in IV-B. Then we used pretrained CNNs and added global average pooling (GAP) layer to both and then combined the extracted features (2048 from ResNet50 and 512 from VGG19). After we added more dense layers and dropout regularization (dropout value as 0.5) to reduce overfitting and finally added SoftMax layer to identify the most matching category of an image. This model was trained with Adam optimization and we used mini batch size as 64 for this CNN. We used popular categorical cross-entropy loss function as the loss function of this mode. After 15 epochs, about 4 hours of training, our CNN achieved over 94% accuracy on train dataset and over 80% accuracy on test dataset.

data-cleaning

data-visualization

dataset

machine-learning-intro-and-intermediate

my_ml_service

My machine learning web service

my_spotify_app_heruko

porfolio

my portfolio designed by myself

project_spotify_prediction

pytorch-deep-learning-notebooks-in-french

Répertoire des traductions françaises des notebooks d'Alfredo Canziani du cours Deep Learning de la NYU

robot-mobile-

spotify_predict

stage1a

technocolabs-internship-mini-project-reducing-traffic-mortality

Mini-Project-Reducing-Traffic-Mortality

ymfc-mini-drone

mini-drone (hardware design and software modification), based on ymfc-32