Python 初級：深度學習 + PyTorch 入門|Deep Learning|Neural Network|教學|廣東話|60fps

February 9, 2024

Patreon: https://www.patreon.com/kfsoft
Deep learning + Pytrorch
* Tensor computation
* Gradient
* Broadcasting

00:00 Introduction
00:33 Roadmap
02:17 Patreon
02:36 Background, supervised learning
07:21 Deep Learning & neural network
16:43 Training
26:14 Network types
29:13 Tensor
35:29 Computations
38:16 Device
40:37 Element-wise computation
44:00 Broadcasting
52:47 Vector matrix computation – matmul(x,y)
56:47 Vector-vector product
57:40 Matrix-matrix product
59:25 Vector-matrix product
01:01:03 Matrix-vector product
01:03:00 Gradient
01:06:56 Gradient-based optimization
01:12:35 Derivative
01:14:33 Partial-derivative & gradient
01:20:52 autograd.grad(y,x)
01:23:48 sum() to scalar
01:27:16 y.backward()
01:30:28 DEMO – MLP classifier
01:51:39 What tensor cannot do
01:52:20 Installation
01:53:38 Tensor basics
02:08:45 Scalar, vector, matrix
02:26:17 Tensor() VS tensor()
02:31:34 Create & initialize tensors
02:33:58 empty(), zeros(), ones(), full()
02:39:01 clone() + view() + reshape()
02:43:56 clone() & view()
02:53:18 create view with slicing
03:00:15 reshape()
03:07:24 arange(), flip()
03:15:45 transpose()
03:17:36 Random
03:20:21 rand(), randn(), randint()
03:24:10 linspace(), eye()
03:27:55 Computation
03:29:04 Addition (element-wise)
03:34:36 Multiplication (element-wise)
03:38:43 Broadcasting if shape not match
03:42:18 Vector / matrix product
03:44:32 vector-vector product
03:45:29 matrix-matrix product
03:50:23 vector-matrix product
03:52:19 matrix-vector product
03:53:08 boolean comparisons
03:56:10 any(), all()
04:01:44 logical operator
04:05:40 comparsion with boardcasting
04:08:06 boolean mask for item selection
04:10:43 Integer indexing + slicing
04:13:40 Boolean indexing (selection / update)
04:14:04 Fancy indexing (create new tensor using indices)
04:16:21 Norm
04:18:22 Unsequeeze()
04:21:59 Sequeeze()
04:25:34 sort()
04:29:22 topk()
04:32:13 Device
04:35:06 Cpu vs cuda
04:41:45 Tensor creation & move time
04:46:35 Gradient
04:47:18 autograd.grad()
04:52:51 backward()
05:01:03 clear gradients
05:02:15 Leaf node trap
05:05:30 detach() vs detach_()
05:07:28 detach_()
05:12:36 view cannot call detach_()
05:18:27 Summary