Machine Learning and Go
summary
Someone ports MNIST logistic regression to Go, first using pure Go matrix code and one goroutine per row, sees single-threaded and naive parallel runs slower than Python+native C, then swaps to gonum with BLAS/GEMM via cgo and gets much faster performance. A toy demo hits about 98% accuracy with longer training. Also shows an immutable-matrix library, experiments with one goroutine per row, and points toward cgo + GPU and TensorFlow bindings as next steps.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:55History of Machine Learning
- 2:00Alphago
- 2:56AlphaGo Master
- 3:37GPU CPU
- 4:16Machine Artificial Intelligence
- 5:54Performance
- 6:10Input
- 6:24Logistic Regression
- 7:56What could go wrong
- 9:32Matrix
- 10:07Wheel
- 10:40Boyfriends
- 11:18ImageCut
- 13:13Matrix Multiplication
- 15:23References
related talks
31mMachine Learning for Gophers
50mDeep Learning with Go
29mLearn Neural Networks With Go, Not Math
06mLearn Neural Networks With Go—Not Math! (Interview)
33mLearn Neural Networks With Go - Not Math!
48mMachine learning by an imp-op-steur
29mThe Gopher Orchestra: Distributed Machine Learning with Go
26mDeep learning in Go - Present and future