Learning to perceive the world as articulated: an approach for hierarchical learning in sensory-motor systems

Tani, Jun; Nolfi, Stefano

doi:10.1016/s0893-6080(99)00060-x

Cited by 194 publications

(131 citation statements)

References 21 publications

Supporting

Mentioning

131

Contrasting

Order By: Relevance

“…Figure 6 (b) indicates that the adaptive optimization of σ allows a set of reusable chunks corresponding to the Lissajous curve patterns to be successfully extracted, and self-organized chunks to be allocated in each corresponding module. The segmentation mechanism caused by the indeterminacy has also been discussed in a study by Tani and Nolfi [26]. In their study, involving a robot actively exploring two rooms connected by a door, the sensory-motor flow was segmented by means of the uncertainty of the door opening.…”

Section: Segmentation Of Temporal Time Series Caused By Indeterminacymentioning

confidence: 95%

“…An important difference between the proposed method and the conventional method used in [26,6] is the use of an optimized variance σ of the normal distribution. Indeed, if σ is a constant, e.g., σ = 1, then the proposed method is equivalent to the conventional one.…”

Section: Learning Methodsmentioning

confidence: 99%

“…The proposed learning method allows a sequence to be segmented into reusable blocks without loss of stability with increasing number of learning modules. A maximum likelihood estimation based on the gradient descent algorithm is employed, similar to the conventional learning method for mixture of RNN experts [26,6], although the present likelihood function differs from those presented previously. In the proposed method, the weighting factor for the allocation of blocks to modules is changed adaptively via a variance parameter for each module.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

A model for learning to segment temporal sequences, utilizing a mixture of RNN experts together with adaptive variance

Namikawa

Tani

2008

Neural Networks

View full text Add to dashboard Cite

This paper proposes a novel learning method for a mixture of recurrent neural network (RNN) experts model, which can acquire the ability to generate desired sequences by dynamically switching between experts. Our method is based on maximum likelihood estimation, using a gradient descent algorithm. This approach is similar to that used in conventional methods; however, we modify the likelihood function by adding a mechanism to alter the variance for each expert. The proposed method is demonstrated to successfully learn Markov chain switching among a set of 9 Lissajous curves, for which the conventional method fails. The learning performance, analyzed in terms of the generalization capability, of the proposed method is also shown to be superior to that of the conventional method. With the addition of a gating network, the proposed method is successfully applied to the learning of sensory-motor flows for a small humanoid robot as a realistic problem of time series prediction and generation.

show abstract

Section: Segmentation Of Temporal Time Series Caused By Indeterminacymentioning

confidence: 95%

Section: Learning Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

A model for learning to segment temporal sequences, utilizing a mixture of RNN experts together with adaptive variance

Namikawa

Tani

2008

Neural Networks

View full text Add to dashboard Cite

show abstract

“…This becomes possible by equipping the robot with a sensor to detect the position of the reward-zone used for fitness evaluation. 1 A different line of research has studied how agents in a self-organized ways can learn internal models of the environment [9]. The authors successfully trained a hierarchy of recurrent neural networks to predict increasingly complex information about the environment.…”

Section: Introductionmentioning

confidence: 99%

Exploring the T-Maze: Evolving Learning-Like Robot Behaviors Using CTRNNs

Blynel

Floreano

2003

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. This paper explores the capabilities of continuous time recurrent neural networks (CTRNNs) to display reinforcement learning-like abilities on a set of T-Maze and double T-Maze navigation tasks, where the robot has to locate and "remember" the position of a reward-zone. The "learning" comes about without modifications of synapse strengths, but simply from internal network dynamics, as proposed by [12]. Neural controllers are evolved in simulation and in the simple case evaluated on a real robot. The evolved controllers are analyzed and the results obtained are discussed.

show abstract

“…Neural networks provide a sensed input to output mapping function which may be refined by a variety of well known learning methods ( [5], [14]). Genetic Learning algorithms rank and refine sets of condition-action rules according to some "fitness function" (past payoff in the case of [4]).…”

Section: Introductionmentioning

confidence: 99%

Integrating Unsupervised Learning, Motivation and Action Selection in an A-life Agent

Witkowski

1999

Advances in Artificial Life

View full text Add to dashboard Cite

Abstract.How can we expect an A-life Agent to learn how to perform tasks when it is not told what those tasks are, and it is not provided any indication or feedback as to its performance? This is at the heart of the unsupervised learning problem. If the Agent were able to learn in this manner, how could specific tasks be communicated to it? This is the Goal setting problem. Having been set a task, how would the Agent go about choosing things to do that will lead it to perform those tasks in an orderly manner? This is at the heart of the action selection problem.

show abstract

Learning to perceive the world as articulated: an approach for hierarchical learning in sensory-motor systems

Cited by 194 publications

References 21 publications

A model for learning to segment temporal sequences, utilizing a mixture of RNN experts together with adaptive variance

A model for learning to segment temporal sequences, utilizing a mixture of RNN experts together with adaptive variance

Exploring the T-Maze: Evolving Learning-Like Robot Behaviors Using CTRNNs

Integrating Unsupervised Learning, Motivation and Action Selection in an A-life Agent

Contact Info

Product

Resources

About