Boostcamp AI Tech (Day 037)
My assignment: Hyperparameter search
Timespace of ML model
- Entropy
- initial state (low entropy)
- ML process (high entropy: many possible states)
- terminal state (low entropy)
- Parameter, Hyperparameter search
- Grid search: Stochastic
- SHA (Successive Having Algorithm)
- Random search(2012)): Non-stochastic
- Surrogate model
- Gaussian process
- Grid search: Stochastic
- Neural Architecture Search (NAS)
- automatic NAS with reinforcement learning (ICLR 2017)
- NAS for edge devices
- MnasNet
- Proxylessnas
- Once-for-all
- MobileNets