Deep Learning with CNTK and F#
summary
Shows concrete use of CNTK .NET API from F# to train small models: a toy 64‑image classifier and a 5,000‑image handwriting model (500‑image test set yields 9 mistakes ≈ 4.5% error). Uses tensors, parameters, sigmoid loss and stochastic gradient descent, inspects learned 8x8 weight maps that match expected digit strokes, and builds a readable sequential model DSL (convolutions + pooling) because the raw C++/C# bindings are verbose. Runs on CPU, trains to low loss, and saves interoperable models.
AI-written summary · about
chapters (21)
- 0:00Intro
- 1:04Context
- 4:59Activation Gate
- 8:04Can we reverse engineer this?
- 10:08Learning by example
- 13:04Robustness
- 15:03Learning a Circuit / Board
- 15:40Composing more complex circuits
- 18:23Model: Composition of Layers
- 19:36What is CNTK?
- 20:55What does CNTK do?
- 21:34Core components
- 22:33What the hell is a Tensor?
- 23:35Tensors
- 24:27Functions
- 36:58Sequential Models
- 38:25Demo: MNIST dataset
- 39:20MNIST input
- 46:42Conclusion
- 47:36What about TensorFlow? PyTorch?
- 49:35Thank you :
related talks
57mDeep Learning with CNTK and F#
57mDeep Learning with CNTK and F#
26mIntroduction to the Microsoft Cognitive Toolkit (Deep Learning Framework)
01h 05mDeep Learning with PyTorch
01h 57mUsing CNTKs Python Interface for Deep Learning
44mAI for every developer
55mA Tour of Deep Learning With C++
57mWriting a Neural Net from Scratch