ASGD

Average Stochastic Gradient Descent, abbreviated as ASGD, averages the weights that are calculated in every iteration.

$\$\$w\_\{t+1\}=w\_t-\backslash eta\; \backslash nabla\; Q(w\_t)\$\$$

where $\$w\_t\$$ being the weight tensor , $\$\backslash eta\$$ being the base learning rate and $\$\backslash nabla\; Q(w\_t)\$$ being the gradient of the objective function evaluated at $\$w\_t\$$.

With the given update rule SGD assigns calculated weight to the model. But with ASGD assigns the following averaged weight $\$\backslash overline\{w\}\$$,

$\$\$\backslash overline\{w\}=\backslash frac\{1\}\{N\}\; \backslash sum\_\{t=1\}^Nw\_t\$\$$

where $\$w\_t\$$ is the weight tensor calculated in iteration 't'.

Major Parameters

• Base Learning Rate
• Weight Decay
• Lambda
• Alpha
• TO

Lambda

It is the decay term for the past weights used in the average.

Alpha

It is the power value that is used to update the learning rate.

TO

It is the optimization step at which the averaging is started. If the required number of iteration is lower than the TO value, then the averaging will not happen.

Code Implementation

      # importing the library
import torch
import torch.nn as nn

x = torch.randn(10, 3)
y = torch.randn(10, 2)

# Build a fully connected layer.
linear = nn.Linear(3, 2)

# Build MSE loss function and optimizer.
criterion = nn.MSELoss()

# Optimization method using ASGD
optimizer = torch.optim.ASGD(linear.parameters(), lr=0.01, lambd=0.0001, 
alpha=0.75, t0=1000000.0, weight_decay=0)

# Forward pass.
pred = linear(x)

# Compute loss.
loss = criterion(pred, y)
print('loss:', loss.item())

optimizer.step()