I'm starting a newsletter. It's called "Sparks in the Wind": mostly ephemeral random thoughts — but with a small chance of starting a fire. Posts are going to be similar to my Twitter threads, but longer and more polished. First post is on education: https://
fchollet.substack.com/p/education-as
-civilization-building
…
@fchollet
-
François Chollet Launches Sparks in the Wind Newsletter on Education
By
–
-
Keras Preprocessing Layers: Seeking Company Collaboration for Large-Scale adapt()
By
–
If you're a company that uses Keras and you face this use case (large scale adapt() of Keras preprocessing layers), would you consider working with us to implement it? We're a small team and we don't have the resources for this at this time…
-
Beam-style computation support designed but not yet implemented
By
–
Correct, the underlying API / infra is designed to potentially allow Beam-style computation. We have not implemented it and it's currently deprioritized (the current adapt() is serial and single-threaded). But we could if there's demand in the future — the design is there.
-
TensorFlow Graph Performance Eliminates Python Slowness Issues
By
–
The fact that we're able to do everything as part of the TF graph is really nice — Python slowness is never an issue. There's no need for us to rewrite anything in, like, Cython or Rust.
-
Keras Preprocessing Layers: High-Performance In-Graph Implementation
By
–
All the work is done in Keras preprocessing layers, which are implemented in TF ops (everything is 100% in-graph!) so it's highly performant. During training (presumably on GPU/TPU) you'd use async preprocessing in TF data to avoid CPU preprocessing being a bottleneck.