A5000727773- Neural Networks and Applications
- Robot Manipulation and Learning
- Neural dynamics and brain function
- Action Observation and Synchronization
- Reinforcement Learning in Robotics
- Embodied and Extended Cognition
- EEG and Brain-Computer Interfaces
- Motor Control and Adaptation
- Functional Brain Connectivity Studies
- Cognitive Science and Education Research
- Language and cultural evolution
- Anomaly Detection Techniques and Applications
- Evolutionary Algorithms and Applications
- Human Pose and Action Recognition
- Visual perception and processing mechanisms
- Robotics and Automated Systems
- Child and Animal Learning Development
- Time Series Analysis and Forecasting
- Cognitive Science and Mapping
- Social Robot Interaction and HRI
- Robotic Path Planning Algorithms
- Speech and dialogue systems
- Modular Robots and Swarm Intelligence
- Tactile and Sensory Interactions
- Natural Language Processing Techniques
Okinawa Institute of Science and Technology Graduate University
2016-2025
Korea Advanced Institute of Science and Technology
2012-2018
Korea Institute of Science & Technology Information
2017
RIKEN Center for Brain Science
2004-2013
Kootenay Association for Science & Technology
2013
The University of Tokyo
1986-2011
Tokyo University of the Arts
2011
Dynamic Imaging (United Kingdom)
2010
Indiana University
2010
RIKEN
2003-2009
It is generally thought that skilled behavior in human beings results from a functional hierarchy of the motor control system, within which reusable primitives are flexibly integrated into various sensori-motor sequence patterns. The underlying neural mechanisms governing way continuous flows segmented and series sequences have, however, not yet been clarified. In earlier studies, this has realized through use explicit hierarchical structure, with local modules representing lower level...
This paper discusses how a behavior-based robot can construct "symbolic process" that accounts for its deliberative thinking processes using models of the environment. The focuses on two essential problems; one is symbol grounding problem and other internal symbolic be situated with respect to behavioral contexts. We investigate these problems by applying dynamical system's approach navigation learning problem. Our formulation, based forward modeling scheme recurrent neural learning, shows...
This paper investigates how behavior primitives are self-organized in a neural network model utilizing distributed representation scheme. The is characterized by so-called parametric biases which adaptively modulate the encoding of different patterns single recurrent net (RNN). Our experiments, using real robot arm, showed that set end-point and oscillatory learned self-organizing fixed points limit cycle dynamics form primitives. It was also found diverse novel can be generated modulating...
We present a novel connectionist model for acquiring the semantics of simple language through behavioral experiences real robot. focus on “compositionality” and examine how it can be generated experiments. Our experimental results showed that essential structures situated self-organize themselves dense interactions between linguistic processes whereby certain generalization in learning is achieved. analysis acquired dynamical indicates an equivalence compositionality appears combinatorial...
This position paper proposes that the study of embodied cognitive agents, such as humanoid robots, can advance our understanding development complex sensorimotor, linguistic, and social learning skills. in turn will benefit design robots capable to handle manipulate objects tools autonomously, cooperate communicate with other humans, adapt their abilities changing internal, environmental, conditions. Four key areas research challenges are discussed, specifically for issues related of: 1) how...
Artificial autonomous agents and robots interacting in complex environments are required to continually acquire fine-tune knowledge over sustained periods of time. The ability learn from continuous streams information is referred as lifelong learning represents a long-standing challenge for neural network models due catastrophic forgetting which novel sensory experience interferes with existing representations leads abrupt decreases the performance on previously acquired knowledge....
Lifelong learning is fundamental in autonomous robotics for the acquisition and fine-tuning of knowledge through experience. However, conventional deep neural models action recognition from videos do not account lifelong but rather learn a batch training data with predefined number classes samples. Thus, there need to develop systems ability incrementally process available perceptual cues adapt their responses over time. We propose self-organizing architecture classify human actions video...
This study proposes a novel type of dynamic neural network model that can learn to extract stochastic or fluctuating structures hidden in time series data. The learns predict not only the mean next input state, but also its time-dependent variance. training method is based on maximum likelihood estimation by using gradient descent and function expressed as estimated Regarding evaluation, we present numerical experiments which data were generated different ways utilizing Gaussian noise. Our...
This study introduces PV-RNN, a novel variational RNN inspired by predictive-coding ideas. The model learns to extract the probabilistic structures hidden in fluctuating temporal patterns dynamically changing stochasticity of its latent states. Its architecture attempts address two major concerns Bayes RNNs: how variables can learn meaningful representations and inference transfer future observations variables. PV-RNN does both introducing adaptive vectors mirroring training data, whose...
This study presents experiments on the imitative interactions between a small humanoid robot and user. A dynamic neural network model of mirror system was implemented in robot, based recurrent with parametric bias (RNNPB). The showed that after learns multiple cyclic movement patterns as embedded RNNPB, it can regenerate each pattern synchronously movements human who is demonstrating corresponding front robot. Further, exhibits diverse interactive responses when user demonstrates novel...
Diverse, complex, and adaptive animal behaviors are achieved by organizing hierarchically structured controllers in motor systems. The levels of control progress from simple spinal reflexes central pattern generators through to executive cognitive the frontal cortex. Various types hierarchical structures have been introduced shown be effective past artificial agent models, but few studies how such can self-organize. This study describes may evolve a recurrent neural network model implemented...
We suggest that different behavior generation schemes, such as sensory reflex and intentional proactive behavior, can be developed by a newly proposed dynamic neural network model, named stochastic multiple timescale recurrent (S-MTRNN). The model learns to predict subsequent inputs, generating both their means uncertainty levels in terms of variance (or inverse precision) utilizing its property. This was employed robotics learning experiments which one robot controlled the S-MTRNN required...
Active inference is a mathematical framework which originated in computational neuroscience as theory of how the brain implements action, perception and learning. Recently, it has been shown to be promising approach problems state-estimation control under uncertainty, well foundation for construction goal-driven behaviours robotics artificial agents general. Here, we review state-of-the-art implementations active state-estimation, control, planning learning; describing current achievements...
Goal-directed human behavior is enabled by hierarchically-organized neural systems that process executive commands associated with higher brain areas in response to sensory and motor signals from lower areas. Psychiatric diseases psychotic conditions are postulated involve disturbances these hierarchical network interactions, but the mechanism for how aberrant disease generated networks, a systems-level framework linking specific psychiatric symptoms remains undetermined. In this study, we...
This paper aims to investigate how adequate cognitive functions for recognizing, predicting, and generating a variety of actions can be developed through iterative learning action-caused dynamic perceptual patterns. Particularly, we examined the capabilities mental simulation one's own as well inference others' intention because they play crucial role, especially in social cognition. We propose neural network model based on predictive coding which generate recognize visuo-proprioceptive The...
The integration of multisensory information plays a crucial role in autonomous robotics to forming robust and meaningful representations the environment. In this work, we investigate how multimodal can naturally develop self-organizing manner from co-occurring inputs. We propose hierarchical architecture with growing neural networks for learning human actions audiovisual processing visual inputs allows obtain progressively specialized neurons encoding latent spatiotemporal dynamics input,...
We present a connectionist model that combines motions and language based on the behavioral experiences of real robot. Two models recurrent neural network with parametric bias (RNNPB) were trained using motion sequences linguistic sequences. These combined their respective parameters so robot could handle many-to-many relationships between Motion articulated into some primitives corresponding to given prediction error RNNPB model. The experimental task in which humanoid moved its arm table...