To avoid overshooting as the model nears the optimum, the learning rate can be decayed over time using exponential or inverse functions.

To avoid overshooting as the model nears the optimum, the learning rate can be decayed over time using exponential or inverse functions.

• A. Dynamic Learning Rate
• B. Mini-Batch Gradient Descent
• C. Stochastic Gradient Descent
• D. Batch Gradient Descent

Answer: (A)

Dynamic learning rate is the idea of adjusting the step size of gradient updates as training progresses. By decaying the learning rate over time with exponential or inverse schedules, the model takes larger steps early to explore the landscape and smaller steps later to fine-tune near the optimum. This helps prevent overshooting as you approach the minimum and promotes smoother, more reliable convergence. The other terms describe how data is fed into updates (full batch, mini-batches, or single-sample updates) rather than how the learning rate itself changes over time, so they don’t capture the described technique.

Dynamic learning rate is the idea of adjusting the step size of gradient updates as training progresses. By decaying the learning rate over time with exponential or inverse schedules, the model takes larger steps early to explore the landscape and smaller steps later to fine-tune near the optimum. This helps prevent overshooting as you approach the minimum and promotes smoother, more reliable convergence. The other terms describe how data is fed into updates (full batch, mini-batches, or single-sample updates) rather than how the learning rate itself changes over time, so they don’t capture the described technique.