Running AI models on the client offers low latency and high privacy but is constrained by computing power and model size.
TensorFlow.js is a Google library that enables running, retraining, and developing machine learning models directly in JavaScript environments.
Machine learning models learn from datasets to perform tasks like classification, illustrated by training a model to distinguish between cats and dogs.
A step-by-step code walkthrough shows how to implement a linear regression using the neural network architecture in TensorFlow.js.
Properly training a model involves feeding it the right amount of data to avoid overfitting or underfitting, using a feedback loop of rewards and punishments.
The loss function measures the model's error, and its value should decrease over training epochs until it plateaus, indicating an optimal stopping point.
Apply your knowledge by building a classifier to predict customer purchasing behavior and explore other examples like gesture control or running GPT-2 in the browser.
Jobs that call for the skills explored in this talk.