Deep Learning with TensorFlow
summary
Builds small TensorFlow pipelines that go from MNIST pixels to a deployed recognizer: defines X and Y placeholders, optimizes W and b via gradient descent/Adam, freezes a prediction graph and loads it into a WPF app. Trains on 60,000 digits with batch size ~1,000 per epoch, trained in ~60s on GPU, inspects runs in TensorBoard, and exports W matrix 784x10. Shows mean squared / cross-entropy losses and reports accuracy: linear 82% → MLP 97% → CNN higher.
AI-written summary · about
chapters (28)
- 0:00Intro
- 1:00AI and Machine Learning
- 12:11Accuracy vs Recall
- 14:47Programming
- 22:33Bias
- 25:06Linear Discriminator
- 25:35Model
- 28:46Output
- 32:24H function
- 33:33Mean squared error
- 33:47Parabola example
- 38:12Gradient Descent
- 39:28Tensor Flow
- 40:05TensorFlow Flow
- 41:50Accuracy
- 42:07Visual Studio
- 44:11Training the model
- 46:30Summary operation
- 48:09Logs
- 48:28Optimization Graph
- 48:43Freezing a Model
- 49:21Inference
- 50:57Linear Model
- 53:17Mathematical Induction
- 53:32Activation Function
- 53:59Derivatives
- 54:46Multilayer Perceptron
- 55:38Atom Optimizer
related talks
01h 06mDeep Learning with TensorFlow
34mML from a software engineer's perspective
01h 05mDeep Learning with PyTorch
01h 04mWhat you always wanted to know about Deep Learning, but were afraid to ask
49mDeep Learning With Python & Tensorflow
56mTensorFlow and Deep Learning without a PhD, Part 1
44mHello Deep Learning
57mDeep Learning with C++