SGAS: Sequential Greedy Architecture Search (1912.00195v2)

Abstract: Architecture design has become a crucial component of successful deep learning. Recent progress in automatic neural architecture search (NAS) shows a lot of promise. However, discovered architectures often fail to generalize in the final evaluation. Architectures with a higher validation accuracy during the search phase may perform worse in the evaluation. Aiming to alleviate this common issue, we introduce sequential greedy architecture search (SGAS), an efficient method for neural architecture search. By dividing the search procedure into sub-problems, SGAS chooses and prunes candidate operations in a greedy fashion. We apply SGAS to search architectures for Convolutional Neural Networks (CNN) and Graph Convolutional Networks (GCN). Extensive experiments show that SGAS is able to find state-of-the-art architectures for tasks such as image classification, point cloud classification and node classification in protein-protein interaction graphs with minimal computational cost. Please visit https://www.deepgcns.org/auto/sgas for more information about SGAS.

• The paper introduces a novel sequential greedy approach for efficiently exploring deep neural network architectures.
• It demonstrates improved performance and reduced computational overhead compared to traditional exhaustive search methods.
• The method offers practical insights for scalable architecture design in resource-constrained research and industrial applications.

Overview of CVPR Proceedings Author Guidelines

The paper "Author Guidelines for CVPR Proceedings" provides a comprehensive outline for authors intending to submit their manuscripts for the CVPR conference. The guidelines are structured to ensure uniformity and adherence to the IEEE Computer Society Press standards. This document is crucial for authors to properly prepare their submissions in a manner that aligns with the conference's requirements for style, format, and content.

Key Content Summary

The guidelines encompass several critical areas authors must adhere to:

1. Language and Length: All manuscripts must be in English, and the submission should not exceed eight pages excluding references. It is explicitly stated that overlength papers will not be reviewed, highlighting the importance of compliance with the format specifications.
2. Reviewing Process: The paper emphasizes the blind review process. Authors are instructed to maintain anonymity through strategic phrasing and referencing, thereby facilitating impartial evaluations.
3. Formatting Details: Detailed instructions are provided on manuscript formatting, including margins, text justification, and section headings, ensuring consistency across submissions. The importance of aligning figures and captions with text for optimal readability is underscored.
4. Mathematical and Reference Protocols: Authors must number displayed equations and sections for ease of reference. A structured approach to citations is encouraged, with specific rules on the order and presentation of multi-author papers.
5. Illustrations and Figures: The paper guides authors on including graphics, advising on resizing and centering to maintain coherence with the main text. This ensures that visual aids contribute effectively to the scientific discourse.

Implications for Authors

The stringent guidelines are designed to streamline the reviewing process and ensure that the presentation of research is not hindered by formatting issues. By adhering to these standards, authors contribute to the collaborative spirit of scientific inquiry, facilitating a clear and focused exchange of ideas. This has practical implications for ensuring that authors' work is accessible, judge based on content rather than presentation, and efficiently integrated into the wider body of knowledge.

Prospects for Future Developments

While current guidelines are well-structured, the evolving landscape of scientific publishing may prompt revisions. Future developments could involve adaptive formats for digital consumption or more integrated tools for ensuring compliance during manuscript preparation. As AI continues to advance, tools that automate formatting checks or suggest improvements during the writing phase could prove beneficial, enhancing productivity and the quality of submissions.

In conclusion, the "Author Guidelines for CVPR Proceedings" offers a rigorous framework for manuscript preparation, serving as an essential reference for authors aiming to publish their work at one of the key conferences in computer vision. Compliance with these guidelines not only facilitates the review process but also underscores the professional presentation of scientific research.