Generating gradients in the energy landscape using rectified linear type cost functions for efficiently solving 0/1 matrix factorization in Simulated Annealing.
Makiko KonoshimaHirotaka TamuraYoshiyuki KabashimaPublished in: CoRR (2023)
Keyphrases
- matrix factorization
- energy landscape
- simulated annealing
- cost function
- collaborative filtering
- recommender systems
- factorization methods
- negative matrix factorization
- low rank
- low rank matrix
- genetic algorithm
- missing data
- nonnegative matrix factorization
- factor analysis
- data sparsity
- evolutionary algorithm
- item recommendation
- stochastic gradient descent
- implicit feedback
- probabilistic matrix factorization
- binary matrix
- closed form
- energy function
- computer vision