離散時間再帰的ニューラルネットワークの2目的最適化問題について

DOI 機関リポジトリ (HANDLE)

書誌事項

タイトル別名
  • ON MULTI-OBJECTIVE OPTIMIZATION PROBLEMS IN DISCRETE-TIME RECURRENT NEURAL NETWORKS

この論文をさがす

説明

This paper studies bi-objective optimization problems in binary neural networks: simple recurrent neural networks characterized by ternary cross connection parameters and the signum activation function. In order to synthesize efficient neural networks, we define bi-objective optimization problems based on two objectives. The first objective evaluates direct stability of desired memories, and the second objective evaluates sparsity of connection parameters. Performing precise numerical experiments for typical binary neural networks, we have obtained Pareto fronts that guarantees existence of trade-offs between memory stability and connection sparsity.

収録刊行物

キーワード

詳細情報 詳細情報について

  • CRID
    1390582462817879808
  • DOI
    10.15002/00030720
  • HANDLE
    10114/00030720
  • ISSN
    24368083
  • 本文言語コード
    ja
  • 資料種別
    departmental bulletin paper
  • データソース種別
    • JaLC
    • IRDB
  • 抄録ライセンスフラグ
    使用可

問題の指摘

ページトップへ