Towards Flatter Loss Surface via Nonmonotonic Learning Rate Scheduling.
Sihyeon SeongYegang LeeYoungwook KeeDongyoon HanJunmo KimPublished in: UAI (2018)
Keyphrases
- learning rate
- learning algorithm
- convergence rate
- error function
- hidden layer
- multilayer neural networks
- scheduling problem
- logic programming
- rapid convergence
- activation function
- adaptive learning rate
- convergence theorem
- scheduling algorithm
- convergence speed
- nonmonotonic reasoning
- surface reconstruction
- weight vector
- modal logic
- ant colony optimization
- high accuracy
- np hard
- evolutionary algorithm