【作者】 程献礼；

【导师】 任晓明；

【作者基本信息】 南开大学 ， 逻辑学， 2013， 博士

【摘要】 贝叶斯推理理论是归纳逻辑理论和应用研究的热点,本文在借鉴国内外上归纳逻辑最新研究成果的基础上,系统阐述了归纳疑难和归纳悖论、贝叶斯主义和贝叶斯概率逻辑的恰当性、贝叶斯统计推理的恰当性以及贝叶斯网络的哲学意义等问题,重点论证了归纳逻辑哲学的中心问题是逻辑形式系统与其现实原型的恰当相符性问题,分析了认知科学中著名的蒙提霍尔问题、从因明论式的角度对确证悖论做了全新的解读。从逻辑、哲学与认知三个方面总结了归纳逻辑发展的趋势和展望了现代归纳逻辑的未来。论文首先在比较解决休谟问题的不同方案的前提下,指出贝叶斯方案在解决休谟问题上的优越性,论证贝叶斯方法是如何实现客观性和主观性的统一,私人性和公共性的统一的。其次,阐明了贝叶斯推理的形式化发展对现代归纳逻辑发展的贡献。认为贝叶斯推理的形式化使得现代归纳推理获得并具备了与演绎推理大致相等的地位,按照贝叶斯主义的路径和方法所进行的推理研究,将是科学的、合理的推理方法。第三,指出贝叶斯推理可以更好地处理各种疑难问题。主张在经验与先验、主观与客观之间保持必要的张力,以避免和克服贝叶斯推理所面临的困难,预测了贝叶斯推理未来发展的方向和前景。第四,对三种主要的概率解释进行解读,认为概率解释是多元的,是一个从客观到主观的连续谱系。第五,从贝叶斯推理的视角分析了非经典统计推理的显著性检验和估计理论,认为经典“估计”理论和显著性检验理论都不具有归纳意义,不能用于裁定竞争理论。论文最后指出,不断追求恰当相符是贝叶斯主义归纳逻辑系统产生发展的原动力。在对贝叶斯推理和非贝叶斯推理比较研究的基础上,论文探讨了贝叶斯统计推理中的应用问题。首先,探讨了贝叶斯统计推理的可应用性问题,指出贝叶斯统计推理具有主观与客观相结合的特性。其次,探讨了贝叶斯推理的另一个重要的应用——贝叶斯网络,认为这种不确定推理可以广泛应用于机器学习、人工智能以及无人机等领域,具有较大的可应用性。最后,论文认为贝叶斯推理在认知科学领域及在因明理论中的分析表明,贝叶斯推理不仅要考虑外延因素,而且要探讨涉及背景信息之类的因素。更多还原

【Abstract】 Bayesianism has already gained a prominent. role as a theory for probability interpretation. Keynes’formal inductive system helped it to evolve into a new era: Bayesian reasoning is now remodeling statistics, economics, psychology and artificial intelligence, and its influence is still expanding.Such a trend can be traced back into the renewal of Bayesian method inside statistics. The revolutionary view Bayesianism took successfully eliminated the problem of ignore-to-prior, as well any doubt against subjective factor in induction. By solving the problem of deciding test statistic it averted the problem stopping rule has generated. When it comes to estimation, Bayesianism has replaced confidence interval with credible interval in order to obtain the prediction of the true value of a parameter. It also employs the a priori information through Bayes theorem on conditional probability.Despite its popularity, several challenges were raised against Bayesianism, specifically on its subjectivity, simplicity and on the problem of old-evidence. And as such, more space has been left for the further development of Bayesianism. The debate of subjectivity is focused on a priori-constraint, while the simplicity question is echoed by the inconsistency between probability axiom and the simplicity postulate. Forster and Sober remarked that the postulate itself is an ad hoc method; Howson argued that it should not be taken as an essential guideline. Additionally, an old-evidence in the Bayesian framework could not serve as affirming the current hypothesis, which contradicts with our instinct. Howson insightfully revealed that "evidence support" machanism is actually a ternary relation between data e, hypothesis h and background knowledge k, and the problem of old-evidence only become evident when e was decided as evidence and e was included in k. It shows that the further improvement and development were needed in Bayesian reasoning.Recent developments on the application of Bayesianism in cognitive psychology have invoked a possible cognitive turn in Bayesianism, and it reveals that the various possible approaches of Bayesian reasoning:the integration between Frequentism and Bayesianism; to conjoin the intentional factor with its logical indicative nature, it makes an attempt on the combination of extensionality and non-extensionality. Gigerenzer and Hoffrage has already offered a Bayesian model with frequency representation of information; Tversky, etc. proposed their non-extention logic for the support theory for subjective probability. This seems to be a prospective path for inductive reasoning and Bayesianism that worth further exploring.更多还原