Deep Learning with C++
summary
Repository with code and slides plus live demos show real C++ kernels and deployments: a working custom TensorFlow op implemented with Eigen, a CUDA/cuDNN convolution that runs an edge detector, and a live MNIST trainer using MXNet and DLib. Discussion covers lowering convolution to matrix multiply via m2col lowering, quantization to 8‑bit with speed/size tradeoffs, and hardware numbers (Titan X ~3040 CUDA cores, 12 GB, 540 GB/s; 8× Tesla P100 in a Big Basin server; TPU specs ~92 TOPs for 8‑bit).
AI-written summary · about
chapters (17)
- 0:00Intro
- 1:12Strategy
- 2:07Task Layer
- 3:21Model Layer
- 5:51Layer Layer
- 7:09Layers: Convolution
- 12:18Graph Layer Parallelism
- 14:34Static Graphs
- 16:22Dynamic Graphs
- 17:20Op Layer
- 19:43Kernel Layer
- 22:53Quantization
- 25:19Hardware Layer
- 26:36Hardware: GPU
- 28:32Hardware: Big Basin
- 29:50Hardware: IPU
- 42:56Graph Layer. TensorFlow
related talks
55mA Tour of Deep Learning With C++
47mDeep Learning: a practical introduction from an Android dev perspective
37mDeep Learning Neural Network Acceleration at the Edge
26mDeep Learning in OpenCV
34mFaster deep learning solutions...
49mDeep Learning With Python & Tensorflow
19mWhy Machine Learning on Microcontrollers?
01h 07mDeep Learning with TensorFlow