- 【Updated on May 12, 2025】 Integration of CiNii Dissertations and CiNii Books into CiNii Research
- Trial version of CiNii Research Knowledge Graph Search feature is available on CiNii Labs
- 【Updated on June 30, 2025】Suspension and deletion of data provided by Nikkei BP
- Regarding the recording of “Research Data” and “Evidence Data”
概念学習における学習曲線の評価
-
- GU Hanzhong
- Department of Communications and Systems Engineering The University of Electro-Communications
-
- TAKAHASHI Haruhisa
- Department of Communications and Systems Engineering The University of Electro-Communications
Bibliographic Information
- Other Title
-
- Estimating Learning Curves of Concept Learning
Search this article
Description
本論文では、概念学習における学習曲線を評価するために仮説検定不等式を直接適用する近似手法を導入し、学習アルゴリズムの学習曲線の解析を行う。このため、最悪ではないが、Gibbs学習アルゴリズムよりも汎化性能において劣り、しかもoverfitting問題に関する評価に適するill-posed学習アルゴリズムを導入し、解析を行う。本文における学習曲線の上界の評価においてはVC次元ではなく、ネットワーク上でのパラメーター数以下の値を持つ係数(Regular Interpolation Dimension)が現れる。このためVC理論よりも良い上界となり、しかも、統計物理的な手法よりも一般的な結果が得られる。
In this paper, we describe an approximation method that enables us to study the average generalization performance of learning directly via hypothesis testing inequalities, and investigate learning curves of a so-called ill-posed learning algorithm that peroforms worse than the Gibbs learning algorithm and can provide useful implications regarding the problem of overfitting from a practical and yet scientific viewpoint. The resulting bounds are directly related to the number of system weights. The advantages of the theory are that it alleviates the practical pessimism frequently claimed for the results of the VC theory, and provides general insights.
Journal
-
- IEICE technical report. Neurocomputing
-
IEICE technical report. Neurocomputing 95 (346), 63-70, 1995-10-28
The Institute of Electronics, Information and Communication Engineers
- Tweet
Details 詳細情報について
-
- CRID
- 1573387452283705472
-
- NII Article ID
- 110003232999
-
- NII Book ID
- AN10091178
-
- Text Lang
- en
-
- Data Source
-
- CiNii Articles