Keras is a high-level Python library for neural networks. Its user friendliness and modularity make it well-suited for the rapid prototyping and iteration demanded of a hackathon.
We’ll begin with logistic regression, the simplest of neural networks, to classify handwritten digits in images. Then, we’ll tune a pre-trained network to classify dog breeds from image data. With time, we’ll discuss methods for audio recognition.
This talk will emphasize certain best practices for building and interpreting the performance of models.
No familiarity with deep learning is required. This workshop will be conducted using Kaggle’s online kernels environment, so no installation is required, either.