Not Hotdog: A Machine Learning Masterclass in Disguise
The Not Hotdog app, featured in the popular HBO series Silicon Valley, started as a light-hearted joke. But under the hood, it was a serious exploration of machine learning, built by engineer Tim Anglade. Unlike using readily available APIs, Anglade dove deep into TensorFlow and Keras, building a model specifically for mobile devices.
He faced challenges, including dealing with memory limitations, overfitting, and finding the right balance between model size and accuracy. His journey led him to explore SqueezeNet and ultimately MobileNets, Google’s solution for efficient neural networks on mobile devices.
Anglade meticulously trained his model on a dataset of 150,000 images, with a majority being “not hotdogs.” This intentional imbalance reflected the reality of the world, where hotdogs are a minority of objects.
Beyond the technical details, Anglade highlights the importance of UX/UI and addressing biases during the training process.
The story of Not Hotdog is a testament to the creativity and dedication of machine learning developers, showcasing how a seemingly silly concept can be a powerful learning experience.
