Code along with us onCode Along
In the age of deep learning, data scientists and machine learning engineers seldom create and train neural networks from scratch. A big chunk of what goes into performing a machine learning task, however, is collecting, preparing, and loading data to feed into a model.
In this session, using DataLab we'll learn all about loading custom datasets into PyTorch and using transfer learning to perform an image processing task using a mostly-pretrained model, which we'll fine-tune.
We'll be using computer vision to answer the internet-popular question: is it a sloth or a pain au chocolat? This is a binary image classification task.
Key takeaways
- Machine learning engineers seldom train models from scratch
- Learning to load data is an important building block for deep learning in PyTorch
- Models trained on billions of images are available at our disposal for transfer learning
