Training Large Neural Networks with Constant Memory using a New Execution Algorithm.
Bharadwaj PudipeddiMaral MesmakhosroshahiJinwen XiSujeeth BharadwajPublished in: CoRR (2020)
Keyphrases
- neural network
- memory requirements
- neural network training
- experimental evaluation
- learning algorithm
- dynamic programming
- memory space
- computational cost
- simulated annealing
- times faster
- detection algorithm
- cost function
- recognition algorithm
- training algorithm
- preprocessing
- segmentation algorithm
- associative memory
- training phase
- search space
- artificial neural networks
- multi layer perceptron
- backpropagation algorithm
- convergence rate
- computational complexity
- data sets
- optimization algorithm
- expectation maximization
- high accuracy
- worst case
- supervised learning
- probabilistic model
- feature selection
- np hard
- training set