Analogical Learning / Inference / Reasoning: A study based on Neural-Network Ideas and Cognitive Science

点击翻译按钮获取中文摘要

Earlier, S.Yasui (PI) presented a pruning algorithm called CSDF that is thought to be one form of the Principle of Redundancy Reduction which presumably underlies many of the real brain functions. This project has been devoted to refine/improve CSDF and also apply it mainly to the following two areas (A) and (B).(A) Abstraction-Based Connectionist Analogy Processor (AB-CAP):Analogy has been studied in various disciplines such as psychology, epistemology, pedagogy, science history, cognitive science and AI. Our AB-CAP is relatively simple. As a result of learning the training data, AB-CAP autonomously acquires an internal abstraction model as well as induces appropriate bindings between concrete and abstract entities. The internal model acts as an attractor of new relevant dataset, to allow AB-CAP to be able to deal with multiple analogy paradigms. These prospects have been successfully demonstrated with a number of examples.(B) Independent Component Analysis (ICA) or Blind Source Separation(BSS):ICA is a new useful IT innovation by which to extract otherwise unknown signals from their mixtures observed by sensors. Our method that came out from this project is fundamentally different from existing ones which are all based on information/probability theories. It uses the auto-encoder neural network which operates to minimize the error associated with the input-output identity mapping with the sensor signals as the input vector. CSDF is applied in the decoder part. The hidden nonlinear units that have survived the CSDF pruning will be the blind source extractors. Furthermore, the decoder matrix reconstructs the external mixing matrix, so that the entire decoder part is actually an internal model of the whole external situation. The method is characterized high adaptability and robustness, as has been shown by many simulation examples including real audio and visual data.