The Slingshot Effect: A Late-Stage Optimization Anomaly in Adaptive Gradient Methods.
Vimal ThilakEtai LittwinShuangfei ZhaiOmid SaremiRoni PaissJoshua M. SusskindPublished in: Trans. Mach. Learn. Res. (2024)
Keyphrases
- optimization methods
- machine learning methods
- benchmark datasets
- neural network
- computer vision
- data sets
- steepest ascent
- optimization approaches
- adaptive algorithms
- global optimization
- statistical methods
- optimization algorithm
- empirical studies
- computational cost
- feature space
- preprocessing
- multiscale
- face recognition