notesum.ai

Published at December 3

ILASH: A Predictive Neural Architecture Search Framework for Multi-Task Applications

cs.LG

cs.CV

Released Date: December 3, 2024

Authors: Md Hafizur Rahman¹, Md Mashfiq Rizvee², Sumaiya Shomaji², Prabuddha Chakraborty¹

Aff.: ¹University of Maine; ²University of Kansas

Arxiv: http://arxiv.org/pdf/2412.02116v1

Refer to caption

Method	Efficiency based on GPU usage	Search Strategy	Multi-task Support			Single task Support	AI Driven
Method	Efficiency based on GPU usage	Search Strategy	Regression	Classification	2D Semantic	Single task Support	AI Driven
NAS-RL[21]	Very Low	Cell-based	✗	✗	✗	✓	✗
EAS [22]	High	Transformation based	✗	✗	✗	✓	✗
Hierarchical-EAS [28]	Medium	Hierarchical	✗	✗	✗	✓	✗
AmoebaNet-A [29]	Low	Cell-based	✗	✗	✗	✓	✗
DARTS [30]	Medium	Cell-based	✗	✗	✗	✓	✗
PNAS[26]	Low	Cell-based	✗	✗	✗	✓	✗
AutoSNN [18]	Medium	Spiking Neural Network based	✗	✗	✗	✓	✗
Autokeras [31, 32]	Medium	Network Morphism	✓	✓	✗	✓	✗
CNN-GA [20]	Low	Mutation and Crossover	✗	✗	✗	✓	✗
DARTS-PT[24]	High	Gradient Stabilazation	✗	✗	✗	✓	✗
GENIUS [33]	Medium	NAS-Bench	✗	✗	✗	✓	✓
DCNN [19]	Low	Mutation and Crossover	✗	✗	✗	✓	✗
CGP-CNN[23]	Low	Mutation based	✗	✗	✗	✓	✗
LeMo-NADe[34]	Medium	Open Search	✗	✗	✗	✓	✓
ILASH-Heu (Proposed)	Medium	Branching based	✓	✓	✓	✓	✗
ILASH-Pred (Proposed)	Very High	Branching based	✓	✓	✓	✓	✓