Home/AI/ML Notes/Deep Learning and Machine Learning/Optimization & Learning-Rate Schedules
⌂ Main Page
Deep Learning and Machine Learning

Optimization & Learning-Rate Schedules

Optimizers from SGD to Adam, and learning-rate schedules.

On this page

OPTIMIZATION

Adjusting model’s parameters to minimize a loss function:

Deep neural networks involve millions or even billions of parameters, creating highly non-convex loss surfaces full of valleys and saddle points. Optimization tackles this by introducing adaptive LRs, momentum, curvature approximations, and regularization. Transformers, CNNs, or large-scale recommender systems would be impossible to train efficiently without optimizers.

TYPES OF OPTIMIZATION METHODS

🔹 First-Order Methods

🔹 Adaptive Gradient Methods - dynamically adjust LR per parameter, improving convergence.

🔹 Second-Order & Curvature-Based Methods

PRACTICAL ENHANCEMENTS IN OPTIMIZATION

🔹 Learning Rate Schedules

Control how aggressively updates are made across epochs

🔹 LR Warm-up

🔹 LR Re-Decay (for continual pretraining)

🔹 Training Stabilization Techniques

🔹 Hybrid & Advanced Optimizers