Descriptive Complexity for Neural Networks via Boolean Networks

被引:0
作者
Ahvonen, Veeti [1 ]
Heiman, Damian [1 ]
Kuusisto, Antti [1 ]
机构
[1] Tampere Univ, Tampere, Finland
来源
32ND EACSL ANNUAL CONFERENCE ON COMPUTER SCIENCE LOGIC, CSL 2024 | 2024年 / 288卷
基金
芬兰科学院;
关键词
Descriptive complexity; neural networks; Boolean networks; floating-point arithmetic; logic;
D O I
10.4230/LIPIcs.CSL.2024.9
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
We investigate the descriptive complexity of a class of neural networks with unrestricted topologies and piecewise polynomial activation functions. We consider the general scenario where the running time is unlimited and floating-point numbers are used for simulating reals. We characterize these neural networks with a rule-based logic for Boolean networks. In particular, we show that the sizes of the neural networks and the corresponding Boolean rule formulae are polynomially related. In fact, in the direction from Boolean rules to neural networks, the blow-up is only linear. We also analyze the delays in running times due to the translations. In the translation from neural networks to Boolean rules, the time delay is polylogarithmic in the neural network size and linear in time. In the converse translation, the time delay is linear in both factors. We also obtain translations between the rule-based logic for Boolean networks, the diamond-free fragment of modal substitution calculus and a class of recursive Boolean circuits where the number of input and output gates match.
引用
收藏
页数:22
相关论文
共 16 条
  • [1] Ahvonen Veeti, 2023, LIPIcs, V272
  • [2] Ahvonen Veeti, 2023, arXiv, DOI [10.48550/ARXIV.2308.06277, DOI 10.48550/ARXIV.2308.06277]
  • [3] Barcelo P., 2020, 8 INT C LEARNING REP
  • [4] SIZE-DEPTH TRADEOFFS FOR BOOLEAN-FORMULAS
    BONET, ML
    BUSS, SR
    [J]. INFORMATION PROCESSING LETTERS, 1994, 49 (03) : 151 - 155
  • [5] A Linear Representation of Dynamics of Boolean Networks
    Cheng, Daizhan
    Qi, Hongsheng
    [J]. IEEE TRANSACTIONS ON AUTOMATIC CONTROL, 2010, 55 (10) : 2251 - 2258
  • [6] Grohe M, 2024, Arxiv, DOI [arXiv:2303.04613, DOI 10.48550/ARXIV.2303.04613, 10.48550/ARXIV.2303.04613]
  • [7] The Logic of Graph Neural Networks
    Grohe, Martin
    [J]. 2021 36TH ANNUAL ACM/IEEE SYMPOSIUM ON LOGIC IN COMPUTER SCIENCE (LICS), 2021,
  • [8] Hella L., 2012, ACM S PRINCIPLES DI, P185
  • [9] Weak models of distributed computing, with connections to modal logic
    Hella, Lauri
    Jarvisalo, Matti
    Kuusisto, Antti
    Laurinharju, Juhana
    Lempiainen, Tuomo
    Luosto, Kerkko
    Suomela, Jukka
    Virtema, Jonni
    [J]. DISTRIBUTED COMPUTING, 2015, 28 (01) : 31 - 53
  • [10] HOMEOSTASIS AND DIFFERENTIATION IN RANDOM GENETIC CONTROL NETWORKS
    KAUFFMAN, S
    [J]. NATURE, 1969, 224 (5215) : 177 - &