Younger: The First Dataset for Artificial Intelligence-Generated Neural Network Architecture
Keywords: Artificial Intelligence-Generated Neural Network Architecture, Neural Architecture Design, Graph Neural Network, Benchmark, Dataset
Abstract: Designing and optimizing neural network architectures typically require extensive expertise, starting from handcrafted designs followed by manual or automated refinement, which significantly hinders rapid innovation. To address these challenges, Younger is introduced as a comprehensive dataset derived from over 174K real-world models across more than 30 tasks from various public model hubs. After extensive processing and filtering, Younger includes 7,629 unique architectures, each represented as a directed acyclic graph with detailed operator-level information based on ONNX operator definitions, enabling compatibility across different deep learning frameworks. The dataset is designed to support the emerging research area of Artificial Intelligence-Generated Neural Network Architecture (AIGNNA), which aims to automate their generation and refinement. Comprehensive statistical analysis, including architecture component analyses, highlights the diversity and complexity of architectures in Younger, revealing the potential for future research in this domain. Initial experiments, including operator and dataflow predictions, demonstrate the dataset's utility for architecture exploration and evaluation, and highlight its potential as a benchmark for graph neural networks. Furthermore, an online platform ensures continuous maintenance and expansion of the dataset, supporting global researchers in their endeavors. The dataset and source code are publicly available to encourage further research and lower entry barriers in this challenging domain.
Primary Area: datasets and benchmarks
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 13877
Loading