Olga Petrova

Olga Petrova is a machine learning engineer at Scaleway, a French cloud provider, where her focus lies on deep learning R&D. Previously, she has worked as a researcher in theoretical physics, looking into the applications of artificial intelligence to quantum systems. Olga has a PhD from Johns Hopkins University, and a BS from Worcester Polytechnic Institute. She enjoys blogging about the latest advancements in AI.

projects by Olga Petrova

Semi-Supervised Deep Learning with GANs for Melanoma Detection

3 weeks · 8-10 hours per week average · INTERMEDIATE

In this liveProject series, you’ll take on the role of a computer vision engineer creating a proof of concept for an image recognition mobile app—one with world-changing potential. You’ll build a machine learning model that can identify cancerous moles in low-resolution photos from a phone’s camera. To produce the model, you’ll work through different supervised, unsupervised, and semi-supervised learning techniques, and use data augmentation to improve your training dataset. Each liveProject in this series covers a different deep learning approach for you to build a toolbox of skills that are most relevant to your career.

Semi-Supervised GANs for Melanoma Detection

1 week · 8-10 hours per week · INTERMEDIATE

In this liveProject, you’ll utilize PyTorch and powerful semi-supervised learning techniques to construct an advanced image classifier that can tell whether a 32x32 pixel photo of a mole is melanoma-positive—despite working with a very small labelled dataset. You’ll set up your image preprocessing pipeline, feed data into your PyTorch model, and then train a semi-supervised GAN model on both labeled and unlabeled datasets.

Generative Modeling with VAEs and GANs

1 week · 8-10 hours per week · INTERMEDIATE

In this liveProject, you’ll utilize unlabeled data and unsupervised machine learning techniques to build and train data generative models. You’ll employ generative modeling, such as a variational autoencoder (VAE) that can generate new images by sampling from the latent distribution. You’ll then use an unsupervised generative adversarial network to generate new images.

Train a Supervised Learning Image Classifier

1 week · 8-10 hours per week · INTERMEDIATE

In this liveProject, you’ll use the popular deep learning framework PyTorch to train a supervised learning model on a dataset of melanoma images. Your final product will be a basic image classifier that can spot the difference between cancerous and non-cancerous moles. You’ll create a custom dataset class and data loaders that can handle image preprocessing and data augmentation, and even improve the accuracy of your model with transfer learning.