Exploring the Complex Systems as a Base of Artificial Intelligence

我们利用我们在应用数学和计算方面的经验，从人工智能的角度研究了经济系统和大脑作为复杂系统。我们的主要主题是经济物理学，人类随机生成，神经规划在无创诊断中的应用，以及代理的合作，其中我们将大部分精力投入到经济物理学中。我们重点研究了金融高频数据的统计特性和人工市场的建模，这两个领域都是新兴的科学前沿领域。在研究过程中，我们发现模型参数空间中的关键区域对应于真实的高频金融数据，表明两者具有相同大小的指数Levy分布。这一结果分别发表在2001年的《信息》杂志和2000年召开的第一届经济物理学会议论文集《金融波动的实证科学》（b施普林格，2002）上。我们还通过快速计算机对HFDF进行了详细的研究，澄清了发达的湍流与外汇兑换之间有争议的类比，并发现了基于观察到的稳定模式的短期可预测性，以及三角套利机会及其松弛时间。我们曾多次在国内和国际会议上作报告，并两次出国旅行，包括2000年的信息科学联合会议和2002年在圣达菲研究所举行的非正式研讨会。在人类随机生成的研究中，我们通过分析从六个学生那里获得的新数据的时间序列，表明有可能将个体的特征纳入模型参数，以区分这些个体的差异。神经网络在无创诊断中的应用是我们与Satoshi Yoshida博士的合作成果，我们于2001年提交了专利申请。

We have investigated economic systems and the brain as complex systems from a viewpoint of artificial intelligence, by using our experience on applied mathematics and computation. Our major topics are Econophysics, Human Random Generation, Application of Neural Programming to Noninvasive Diagnosis, and Cooperation of Agents, out of which we devoted most of our effort to Econophysics.We focused on studying the statistical properties of high frequency data in finance (HFDF) an also modeling of artificial markets, both being emerging fields in frontier of science. In the course of study, we have discovered that the critical region in the parameter space of our model corresponds to the real high-frequency financial data, by showing that both have the Levy distribution of the same size of indices. This result was published in INFORMATION journal in 2001 and in Empirical Science of Financial Fluctuations (Springer, 2002) which is the proceedings of the first conference in Econophysics held in 2000. We have also studied HFDF by fast computer in detail to clarify the controversial analogy between well-developed turbulence and foreign currency exchange, and also found short-term predictability based on stable patterns observed, as well as the triangle arbitrage chances and its relaxation time. We have made various presentations in domestic meetings and international conferences and made two oversea travels including Joint Conference on Information Sciences in 2000 and an informal seminar at Santa Fe Institute in 2002.In the study of Human Random Generation, we have shown, by analyzing the time series of newly taken data from our six students, that it is possible to incorporate characteristics of individuals to model parameters that distinguish the differences of those individuals.Neural Network Application to Noninvasive diagnosis is a joint work with Dr. Satoshi Yoshida, on which we submitted an application for patent in 2001.