项目名称: 基于视觉动力神经场的机器学习方法研究
项目编号: No.11301096
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 金德泉
作者单位: 广西大学
项目金额: 24万元
中文摘要: 机器学习在人工智能领域具有重要的地位,其基本目标之一,是利用计算机来模拟人类的学习行为,使得机器具有类似于人类的学习能力,从而能够自主的获取新知识,完成一些特定工作。因此,一直以来,通过利用神经生理学和认知科学研究成果,模拟神经系统的结构来建立神经系统和认知活动的模型,并进一步建立学习算法,实现自主学习的功能,成为了机器学习方法的一个重要来源。本项目基于视觉认知活动的特点,依据神经生理学和神经动力学的理论和方法,通过将基本的视觉认知活动分解为不同的功能层次,有效降低视觉认知系统模型的复杂性,并给出新的基于视觉动力神经场理论的视觉认知模型,并在该模型基础上,通过解决一些关键技术和理论问题,给出基于视觉分类与视觉聚类活动的机器学习算法,并将其应用于解决数据的分类和聚类分析等实际问题中,为视觉认知和神经动力学理论在建立机器学习算法,解决数据挖掘等实际问题中的应用进行了有益探索和尝试。
中文关键词: 动力神经场;工作记忆;短期记忆;聚类;分类
英文摘要: Machine learning plays an important role in artificial intelligent. One of its basic aims is simulating man's learning activity by computer, which give machine some learning ability to obtain new knowledge and fulfill some tasks. As a result, simulating the structure of nerve system by using research achievements of neurophysiology and cognition science to build new models of nerve system and recognition activity and new learning algorithms is always an important source of machine learning. Based on the features of visual cognitive activities, this project divides the basic visual cognitive activity into different functional levels according to theory and methods of neurophysiology and neurodynamics, which effectively reduces the complexity of modeling visual cognitive system. A new visual cognition model is given based on visual dynamical neural field theory, with which machine learning algorithms based on visual classification and visual clustering activities are also given. These algorithms are then used to solve practical issues like data classification and clustering analysis. Therefore, this project is a beneficial exploration and attempt in introducing neurophysiologic and neurodynamical achievements to building machine learning algorithm and solving practical problems like data mining.
英文关键词: Dynamical Neural Field;Working Memory;Short-term Memory;Cluster;Classification