State-Regularized Recurrent Neural Networks to Extract Automata and Explain Predictions

被引:1
|
作者
Wang, Cheng [1 ]
Lawrence, Carolin [2 ]
Niepert, Mathias [2 ,3 ]
机构
[1] Amazon, D-10117 Berlin, Germany
[2] NEC Labs Europe, D-69115 Heidelberg, Germany
[3] Univ Stuttgart, D-70174 Stuttgart, Germany
关键词
Stochastic processes; Logic gates; Learning automata; Behavioral sciences; Probabilistic logic; Symbols; Memory management; Automata extraction; explainability; interpretability; memorization; recurrent neural networks; state machine;
D O I
10.1109/TPAMI.2022.3225334
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recurrent neural networks are a widely used class of neural architectures. They have, however, two shortcomings. First, they are often treated as black-box models and as such it is difficult to understand what exactly they learn as well as how they arrive at a particular prediction. Second, they tend to work poorly on sequences requiring long-term memorization, despite having this capacity in principle. We aim to address both shortcomings with a class of recurrent networks that use a stochastic state transition mechanism between cell applications. This mechanism, which we term state-regularization, makes RNNs transition between a finite set of learnable states. We evaluate state-regularized RNNs on (1) regular languages for the purpose of automata extraction; (2) non-regular languages such as balanced parentheses and palindromes where external memory is required; and (3) real-word sequence learning tasks for sentiment analysis, visual object recognition and text categorisation. We show that state-regularization (a) simplifies the extraction of finite state automata that display an RNN's state transition dynamic; (b) forces RNNs to operate more like automata with external memory and less like finite state machines, which potentiality leads to a more structural memory; (c) leads to better interpretability and explainability of RNNs by leveraging the probabilistic finite state transition mechanism over time steps.
引用
收藏
页码:7739 / 7750
页数:12
相关论文
共 50 条
  • [1] State-Regularized Recurrent Neural Networks
    Wang, Cheng
    Niepert, Mathias
    INTERNATIONAL CONFERENCE ON MACHINE LEARNING, VOL 97, 2019, 97
  • [2] Representation and identification of finite state automata by recurrent neural networks
    Kuroe, Y
    NEURAL INFORMATION PROCESSING, 2004, 3316 : 261 - 268
  • [3] Identification of Finite State Automata With a Class of Recurrent Neural Networks
    Won, Sung Hwan
    Song, Iickho
    Lee, Sun Young
    Park, Cheol Hoon
    IEEE TRANSACTIONS ON NEURAL NETWORKS, 2010, 21 (09): : 1408 - 1421
  • [4] Constructing deterministic finite-state automata in recurrent neural networks
    Omlin, CW
    Giles, CL
    JOURNAL OF THE ACM, 1996, 43 (06) : 937 - 972
  • [5] Training and extraction of fuzzy finite state automata in recurrent neural networks
    Chandra, Rohitash
    Omlin, Christian W.
    PROCEEDINGS OF THE SECOND IASTED INTERNATIONAL CONFERENCE ON COMPUTATIONAL INTELLIGENCE, 2006, : 271 - 275
  • [6] Representation of fuzzy finite state automata in continuous recurrent neural networks
    Omlin, CW
    Thornber, KK
    Giles, CL
    ICNN - 1996 IEEE INTERNATIONAL CONFERENCE ON NEURAL NETWORKS, VOLS. 1-4, 1996, : 1023 - 1027
  • [7] Recurrent neural networks and finite automata
    Siegelmann, HT
    COMPUTATIONAL INTELLIGENCE, 1996, 12 (04) : 567 - 574
  • [8] Adaptive β scheduling learning method of finite state automata by recurrent neural networks
    Arai, K
    Nakano, R
    PROGRESS IN CONNECTIONIST-BASED INFORMATION SYSTEMS, VOLS 1 AND 2, 1998, : 351 - 354
  • [9] Learning minimal automata with recurrent neural networks
    Aichernig, Bernhard K.
    Koenig, Sandra
    Mateis, Cristinel
    Pferscher, Andrea
    Tappler, Martin
    SOFTWARE AND SYSTEMS MODELING, 2024, 23 (03): : 625 - 655
  • [10] Finite State Automata and Simple Recurrent Networks
    Cleeremans, Axel
    Servan-Schreiber, David
    McClelland, James L.
    NEURAL COMPUTATION, 1989, 1 (03) : 372 - 381