Pytorch common parameter initialization methods: [uniform distribution, normal (Gaussian) distribution, Xavier, Kaiming, orthogonal matrix, sparse matrix, constant, identity matrix, zero filling]

One 、“ Uniform distribution ” initialization

torch.nn.init.uniform_(tensor, a=0, b=1)

From uniform distribution U(a, b) In the sample , Initialization tensor .

Two 、“ normal ( gaussian ) Distribution ” initialization

3、 ... and 、Xavier initialization

Four 、kaiming initialization

5、 ... and 、 Orthogonal matrix initialization

6、 ... and 、 Sparse matrix initialization

7、 ... and 、 Constant initialization

8、 ... and 、 Unit matrix initialization

Nine 、 Zero fill initialization

---

---

---

Reference material ：
PyTorch Common parameter initialization methods are explained in detail
PyTorch Common parameter initialization methods are explained in detail