Delta-Bar-Delta (Jacobs)

 

Since the cost surface for multi-layer networks can be complex, choosing a learning rate can be difficult. What works in one location of the cost surface may not work well in another location. Delta-Bar-Delta is a heuristic algorithm for modifying the learning rate as training progresses:

 

• Each weight has its own learning rate.
• For each weight: the gradient at the current timestep is compared with the gradient at the previous step (actually, previous gradients are averaged)
• If the gradient is in the same direction the learning rate is increased
• If the gradient is in the opposite direction the learning rate is decreased
• Should be used with batch only.

 

Let

gij(t) = gradient of E wrt wij at time t

then define

Then the learning rate μij for weight wij at time t+1 is given by

where β, κ , and γ are chosen by the hand.

 

Downsides:

• Knowing how to choose the parameters β, κ , and γ is not easy.
• Doesn't work for online.