CSCO: Connectivity Search of Convolutional Operators

2024-04-26

2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW)

AI Summary

To address the problem of local optima in exploring multi-scale feature connection patterns within convolutional neural networks, this paper proposes a dense connection architecture search method leveraging graph isomorphism augmentation and neural predictor guidance. Our approach introduces four key contributions: (1) a graph isomorphism-based data augmentation strategy to enhance structural representation robustness; (2) a lightweight neural performance predictor for rapid architecture evaluation; (3) a Metropolis-Hastings evolutionary search (MH-ES) algorithm to overcome local convergence bottlenecks in the connection space—common in conventional NAS methods; and (4) a reusable, lightweight connection motif mechanism to improve generalization across tasks. Evaluated on ImageNet, the discovered architecture achieves a 0.6% higher top-1 accuracy than state-of-the-art manually designed and NAS-based models. The source code is publicly available.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Learning & Optimization for CV

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Exploring dense connectivity of convolutional operators establishes critical "synapses" to communicate feature vectors from different levels and enriches the set of transformations on Computer Vision applications. Yet, even with heavy-machinery approaches such as Neural Architecture Search (NAS), discovering effective connectivity patterns requires tremendous efforts due to either constrained connectivity design space or a sub-optimal exploration process induced by an unconstrained search space. In this paper, we propose CSCO, a novel paradigm that fabricates effective connectivity of convolutional operators with minimal utilization of existing design motifs and further utilizes the discovered wiring to construct high-performing ConvNets. CSCO guides the exploration via a neural predictor as a surrogate of the ground-truth performance. We introduce Graph Isomorphism as data augmentation to improve sample efficiency and propose a Metropolis-Hastings Evolutionary Search (MH-ES) to evade locally optimal architectures and advance search quality. Results on ImageNet show ∼ 0.6% performance improvement over hand-crafted and NAS-crafted dense connectivity. Our code is publicly available here.

Problem

• Explores dense connectivity for better feature communication in ConvNets
• Addresses challenges in discovering optimal connectivity patterns efficiently
• Proposes CSCO to automate and enhance convolutional operator connectivity design

Innovation

• CSCO fabricates convolutional operator connectivity
• Uses neural predictor for exploration guidance
• Employs MH-ES to avoid local optima