Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Experiment Videos

Learning to generate combinatorial action sequences utilizing the initial sensitivity of deterministic dynamical

Ryu Nishimoto1, Jun Tani

  • 1Brain Science Institute (RIKEN), 2-1 Hirosawa, Wako-shi, Saitama 351-0198, Japan. ryu@brain.riken.go.jp

Neural Networks : the Official Journal of the International Neural Network Society
|August 18, 2004
PubMed
Summary
This summary is machine-generated.

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Working Memory and Self-Directed Inner Speech Enhance Multitask Generalization in Active Inference.

Neural computation·2025
Same author

Active Inference with Dynamic Planning and Information Gain in Continuous Space by Inferring Low-Dimensional Latent States.

Entropy (Basel, Switzerland)·2025
Same author

Modeling autonomous shifts between focus state and mind-wandering using a predictive-coding-inspired variational recurrent neural network.

Frontiers in computational neuroscience·2025
Same author

How deep will you go? Hierarchy in predictive coding and transformers.

Cognitive neuroscience·2025
Same author

The Epistemic Uncertainty Gradient in Spaces of Random Projections.

Entropy (Basel, Switzerland)·2025
Same author

Digital twin brain simulator for real-time consciousness monitoring and virtual intervention using primate electrocorticogram data.

NPJ digital medicine·2025

Recurrent neural networks (RNNs) learn sensory-action sequences by imitating finite state machines (FSMs). Fractal patterns emerge in initial state mappings, with sequences encoded in transient dynamics, not invariant sets.

Area of Science:

  • Computational neuroscience
  • Machine learning
  • Dynamical systems

Background:

  • Recurrent neural networks (RNNs) are powerful tools for modeling sequential data.
  • Finite state machines (FSMs) provide a framework for understanding discrete sequential processes.
  • Bridging the gap between continuous RNN dynamics and discrete FSM behavior is an ongoing challenge.

Purpose of the Study:

  • To investigate how recurrent neural networks (RNNs) can learn to imitate finite state machine (FSM) behaviors.
  • To analyze the internal representations and dynamics within RNNs that enable FSM imitation.
  • To explore the emergent properties of RNNs when trained on FSM-like sensory-action sequences.

Main Methods:

  • Training RNNs to perform sensory-action sequences mirroring FSMs.

Related Experiment Videos

  • Analyzing the learned initial state mappings and their relationship to FSM states.
  • Investigating the role of transient dynamics versus invariant sets in RNNs for sequence encoding.

    • Main Results:

    • RNNs successfully learned to imitate FSM sensory-action sequences.
    • A fractal structure was observed in the initial state mapping after learning convergence.
    • FSM sequence imitation was found to be encoded in the transient dynamics of the RNNs' internal states, not their invariant sets.

    Conclusions:

    • RNNs can effectively learn and represent discrete sequential behaviors of FSMs.
    • The emergent fractal structures in initial state mappings offer insights into RNN learning.
    • Transient dynamics play a crucial role in RNNs' ability to encode and recall complex sequences.