> ML_ALGORITHM // HYPERBAND-SUCCESSIVE-HALVING_v1.0

Hyperband & Successive Halving

Multi-fidelity bandit-based optimization algorithm that speeds up random and Bayesian search by dynamically allocating resources and early-stopping underperformers.

Multi-Fidelity Resource Allocationevolutionary-searchhigh-intrinsiclarge (>100k)

Back to All Algorithms

Computational Complexity

Training Complexity:O(B * log(B)) where B is total evaluation budget

Inference Complexity:O(1)

Hardware Profile

CPU Friendly:Yes

Requires GPU:Yes

Memory Footprint:moderate

Interpretability & Data

Interpretability Tier:high-intrinsic

Training Data Needs:large (>100k)

Interpretability Assessment

Successive halving brackets clearly discard non-promising models early to focus budget on top performers.

Suitable Tasks & Supported Modalities

Suitable Tasks:

hyperparameter tuningdeep learning architecture search

Supported Modalities:

tabular

Implementing Libraries

OptunaPreferred Networks · v3.6.1

View Spec

ray-tune

Foundational Literature

Common Pitfalls & Warnings

Slow-starting learning rate schedules or architectures with high initial warmups are prematurely killed at early rungs