TensorFlow.js : Bringing ML into the Browser?

Please log in to watch this conference skillscast.

The numerics of machine learning seem like the least likely thing you would expect to run in a browser. The recently-released TensorFlow.js will provide exactly that, however. This Javascript implementation of the TensorFlow APIs is backed by GPU-accelerated WebGL and the optimizations available to modern Javascript engines.

Now, as our production systems become more distributed geographically, we can push some ML capabilities to the edge.

Imagine:

• Reusing existing models in distributed and decentralized architectures
• Image and object detection from local video streams
• Event classification from IoT-based sensors
• Interact with WebAssembly-optimized code on all browsers and platforms
• Natural Language Processing of user text and audio

If models aren’t too big and can be cached, the Web architecture will happily support these new use cases.

YOU MAY ALSO LIKE:

• Keynote: AI, Natural Stupidity and Swimming Submarines (SkillsCast recorded in July 2019)
• F# for a Real World E-commerce Project (SkillsCast recorded in October 2020)
• Nagareyama: a new Fable, faster, more powerful and easier to use (SkillsCast recorded in October 2020)