Celebrating DanNet!

May 15th marks the 12th anniversary of DanNet’s initial deployment (and win) in a computer vision contest. Similar to AlexNet and ResNet, implementations which came later, this so-called convolutional neural network (CNN), which employed deep learning to incredible effect, revolutionized AI. It accelerated the field and paved the way for LLMs like the GPT(s) which are all the rage as well as many other profound applications. DanNet also served to highlight the possibilities of implementations built around graphics processing units (GPUs), which streamline the tensor calculations that have transformed deep learning.

The fundamental innovation of deep learning techniques like CNN are “hidden layers”, called such because the layer’s output is not directly observable in the final model. In a CNN, a “convolutional layer” applies a set of filters to an input image, detecting features such as edges, corners, and shapes, in turn, facilitating image classification, object detection, and segmentation. The output of the convolutional layer is then passed through a non-linear activation function before being fed into the next layer of the network.

What does this all mean? These innovations allowed AI to become better at recognizing images than humans on very short order and led to the human-like interfaces of GPT-style transformer architectures, which allows ChatGPT and its kin to understand natural language and generate coherent and relevant responses based on simple (or complex) inputs.