How can computer make sense of unstructured data, such as images, text or audio? Inspired by how humans perceive their environment, the special architecture of convolutional neural networks (CNNs) allows computers to perform tasks like face detection, machine translation, and transcribing text from audio. This is must-know knowledge for any data scientist who wants to extract insights from such non-classical data sources, which would be very hard to approach otherwise.
This training takes a hands-on approach to understand CNNs. You will become acquainted with the building blocks, such as convolutions and pooling layers, as well as non-linear activation functions, and develop an intuition on how these granular operations can contribute to solve more complex problems. During the training, you will implement your own CNN using the deep learning framework Keras, and train it using NVIDIA GPUs on the Google Cloud Platform. We will also introduce how transfer learning can be applied to CNNs, by re-using concepts that were previously learned from a larger dataset.