Efficient Multi-Objective Neural Architecture Search via Tree Search with Training-Free Metrics

An Vo, Nhat Minh Le, and Ngoc Hoang Luong

Setup

• Clone this repository
• Install packages

$ pip install -r requirements.txt

• Download NATS-Bench, put it in the benchmark folder and follow instructions here

Usage

To run the code, use the command below with the required arguments

python search.py --method <method_name> --dataset <dataset_name> --n_runs <number_of_runs>

Refer to main.py for more details. Example commands:

# TF-MOTNAS-A
python main.py --method TF-MOTNAS-A --dataset cifar10 --n_runs 30

# TF-MOTNAS-B
python main.py --method TF-MOTNAS-B --dataset cifar10 --n_runs 30

Acknowledgement

Our source code is inspired by:

• pymoo: Multi-objective Optimization in Python
• Zero-Cost Proxies for Lightweight NAS
• NATS-Bench: Benchmarking NAS Algorithms for Architecture Topology and Size
• Enhancing Multi-Objective Evolutionary Neural Architecture Search with Training-Free Pareto Local Search