Efficient Automatic Machine Learning via Design Graphs

TL;DR

FALCON introduces a graph-based AutoML approach that efficiently predicts optimal model designs using minimal exploration, combining task-agnostic and task-specific modules to outperform baselines with less computation.

Contribution

The paper presents FALCON, a novel AutoML method that models the design space as a graph and uses GNNs and label propagation to efficiently find high-performing models.

Findings

FALCON achieves comparable time cost to one-shot methods.

FALCON improves performance by an average of 3.3% over baselines.

FALCON requires only 30 explored nodes to find good designs.

Abstract

Despite the success of automated machine learning (AutoML), which aims to find the best design, including the architecture of deep networks and hyper-parameters, conventional AutoML methods are computationally expensive and hardly provide insights into the relations of different model design choices. To tackle the challenges, we propose FALCON, an efficient sample-based method to search for the optimal model design. Our key insight is to model the design space of possible model designs as a design graph, where the nodes represent design choices, and the edges denote design similarities. FALCON features 1) a task-agnostic module, which performs message passing on the design graph via a Graph Neural Network (GNN), and 2) a task-specific module, which conducts label propagation of the known model performance information on the design graph. Both modules are combined to predict the design performances in the design space, navigating the search direction. We conduct extensive experiments on 27 node and graph classification tasks from various application domains, and an image classification task on the CIFAR-10 dataset. We empirically show that FALCON can efficiently obtain the well-performing designs for each task using only 30 explored nodes. Specifically, FALCON has a comparable time cost with the one-shot approaches while achieving an average improvement of 3.3% compared with the best baselines.