TensorFlow

TensorFlow is an open-source machine learning framework released by Google Brain in November 2015. It provides a comprehensive ecosystem for building, training, and deploying machine learning models at every scale, from mobile phones to large distributed systems.

The name comes from its core abstraction: tensors (multi-dimensional arrays) that flow through a computational graph. In TensorFlow 1.x, users defined a static computational graph first, then executed it in a session. TensorFlow 2.0, released in 2019, switched to eager execution by default, making it behave more like standard Python and significantly lowering the learning curve.

TensorFlow's high-level API is Keras, which became the official high-level interface in TensorFlow 2.0. Keras provides a simple, consistent API for defining layers, models, and training loops. Most TensorFlow users interact with Keras rather than the lower-level TensorFlow operations directly.

The framework includes TensorFlow Serving for production model deployment, TensorFlow Lite for mobile and embedded devices, TensorFlow.js for browser-based ML, and TensorFlow Extended (TFX) for end-to-end ML pipelines. This breadth of deployment targets is TensorFlow's primary advantage over competitors.

TensorFlow was the dominant ML framework from 2015 to roughly 2019, when PyTorch began overtaking it in research usage. By 2024, PyTorch dominates academic research and most new AI projects, while TensorFlow retains significant presence in production systems, particularly at Google and in mobile/edge deployment via TensorFlow Lite.

Google's own AI infrastructure runs heavily on TensorFlow, and it remains the framework behind many Google products including Search, Translate, and Google Photos. JAX, also from Google, has increasingly replaced TensorFlow for Google's internal research.