Studies on a Novel Neuro-dynamic Model for Prediction Learning of Fluctuated Data Streams: Beyond Dichotomy between Probabilistic and Deterministic Models
Abstract
The proposed study investigates a novel neuro-dynamic model which can learn to predictor regenerate fluctuated sequence patterns by extracting latent statistical structures in the patterns. The novelty of the model is that the fluctuated sequences are learned by adequately incorporating stochastic dynamics and deterministic chaos self-organized in the network. The model is expected to bring the following advantages \2011\202 adequate mixtures of stochastic dynamics and deterministic one can gain representation power of the model, \2012\202 no needs for arbitrary manipulation of data as well as interpretation of them by human, \2013\202 possibility for scaling of the model by incorporating with the scheme of multiple timescales dynamics for extracting temporal hierarchy from the data. The potential impacts by applying the model to sensory-motor sequence learning by robots as well as video image understanding by accumulated learning of the exemplars are discussed.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 04, 2014
- Accession Number
- ADA614658
Entities
People
- Jun Tani
Organizations
- KAIST