NEAR: A Training-Free Pre-Estimator of Machine Learning Model Performance

AI-generated keywords: NEAR

AI-generated Key Points

The license of the paper does not allow us to build upon its content and the key points are generated using the paper metadata rather than the full article.

  • Authors Raphael T. Husistein, Markus Reiher, and Marco Eckhoff focus on artificial neural networks in their paper "NEAR: A Training-Free Pre-Estimator of Machine Learning Model Performance."
  • Challenges in constructing high-performing neural networks are highlighted due to the laborious nature and substantial computing power required.
  • Introduction of "Near," a training-free pre-estimator that automatically selects optimal network architectures using zero-cost proxies.
  • Traditional Neural Architecture Search (NAS) methods involve training multiple networks, but "Near" estimates network expressivity scores without training data by utilizing activation rank within matrices.
  • Effectiveness of "Near" demonstrated through strong correlation with model accuracy on benchmark datasets like NAS-Bench-101 and NATS-Bench-SSS/TSS.
  • Proposal of estimating optimal layer sizes in multi-layer perceptrons using "Near" scores and informing decisions on hyperparameters like activation functions and weight initialization schemes.
  • Study emphasizes the potential of "Near" as a valuable tool for automated model selection processes in machine learning research and application domains.
Also access our AI generated: Comprehensive summary, Lay summary, Blog-like article; or ask questions about this paper to our AI assistant.

Authors: Raphael T. Husistein, Markus Reiher, Marco Eckhoff

13th International Conference on Learning Representations, ICLR 2025, Singapore
21 pages, 9 figures, 13 tables

Abstract: Artificial neural networks have been shown to be state-of-the-art machine learning models in a wide variety of applications, including natural language processing and image recognition. However, building a performant neural network is a laborious task and requires substantial computing power. Neural Architecture Search (NAS) addresses this issue by an automatic selection of the optimal network from a set of potential candidates. While many NAS methods still require training of (some) neural networks, zero-cost proxies promise to identify the optimal network without training. In this work, we propose the zero-cost proxy \textit{Network Expressivity by Activation Rank} (NEAR). It is based on the effective rank of the pre- and post-activation matrix, i.e., the values of a neural network layer before and after applying its activation function. We demonstrate the cutting-edge correlation between this network score and the model accuracy on NAS-Bench-101 and NATS-Bench-SSS/TSS. In addition, we present a simple approach to estimate the optimal layer sizes in multi-layer perceptrons. Furthermore, we show that this score can be utilized to select hyperparameters such as the activation function and the neural network weight initialization scheme.

Submitted to arXiv on 16 Aug. 2024

Ask questions about this paper to our AI assistant

You can also chat with multiple papers at once here.

The license of the paper does not allow us to build upon its content and the AI assistant only knows about the paper metadata rather than the full article.

AI assistant instructions?

Results of the summarizing process for the arXiv paper: 2408.08776v2

This paper's license doesn't allow us to build upon its content and the summarizing process is here made with the paper's metadata rather than the article.

In their paper titled "NEAR: A Training-Free Pre-Estimator of Machine Learning Model Performance," authors Raphael T. Husistein, Markus Reiher, and Marco Eckhoff delve into the realm of artificial neural networks and their significance as cutting-edge machine learning models across various applications such as natural language processing and image recognition. The authors highlight the challenges associated with constructing high-performing neural networks, emphasizing the laborious nature of this task and the substantial computing power required. To address these challenges, the authors introduce <keyword>Near</keyword>, a training-free pre-estimator that leverages zero-cost proxies to automatically select optimal network architectures from a pool of potential candidates. Traditional methods for <keyword>Neural Architecture Search (NAS)</keyword> involve training multiple neural networks to identify the best architecture, which can be time-consuming and computationally expensive. However, <keyword>Near</keyword> offers a promising alternative by utilizing activation rank within pre- and post-activation matrices to estimate network expressivity scores without requiring any training data. The effectiveness of <keyword>Near</keyword> is demonstrated through its strong correlation with model accuracy on benchmark datasets such as NAS-Bench-101 and NATS-Bench-SSS/TSS. Additionally, the authors propose a straightforward approach for estimating optimal layer sizes in multi-layer perceptrons using <keyword>Near</keyword> scores. They also showcase how <keyword>Near</keyword> can inform decisions regarding hyperparameters like activation functions and weight initialization schemes in neural networks. Overall, this study sheds light on the potential of <keyword>Near</keyword> as a training-free pre-estimator for machine learning model performance. Its findings contribute to advancing automated model selection processes in machine learning research and application domains. With <keyword>Near</keyword>, researchers and practitioners can optimize neural network architectures with minimal computational overhead, making it a valuable tool in the field of artificial intelligence.
Created on 04 Mar. 2025

Assess the quality of the AI-generated content by voting

Score: 0

Why do we need votes?

Votes are used to determine whether we need to re-run our summarizing tools. If the count reaches -10, our tools can be restarted.

Similar papers summarized with our AI tools

Navigate through even more similar papers through a

tree representation

Look for similar papers (in beta version)

By clicking on the button above, our algorithm will scan all papers in our database to find the closest based on the contents of the full papers and not just on metadata. Please note that it only works for papers that we have generated summaries for and you can rerun it from time to time to get a more accurate result while our database grows.

Disclaimer: The AI-based summarization tool and virtual assistant provided on this website may not always provide accurate and complete summaries or responses. We encourage you to carefully review and evaluate the generated content to ensure its quality and relevance to your needs.