When we started TensorFlow.js (TF.js), formerly called deeplearn.js, machine learning (ML) was done mostly in Python. As both JavaScript developers and ML practitioners on the Google Brain team, we quickly realized that there was an opportunity to bridge the two worlds. Today, TF.js has empowered a new set of developers from the extensive JavaScript community to build and deploy ML models and enabled new classes of on-device computation.
TF.js would not exist in its form today without Shanqing, Stan, and Eric. Their contributions to TensorFlow Python, including the TensorFlow Debugger, eager execution, and build and test infrastructure, uniquely positioned them to tie the Python and JavaScript worlds together. Early on in the development, their team realized the need for a library on top of deeplearn.js that would provide high-level building blocks to develop ML models. Shanqing, Stan, and Eric, among others, built TF.js Layers, allowing conversion of Keras models to JavaScript, which dramatically increased the wealth of available models in the TF.js ecosystem. When TF.js Layers was ready, we released TF.js to the world.