Return to search

認知能力與一般化強化學習–凱因斯選美賽局實驗之實證分析 / Effects of cognitive on the Generalized Reinforcement Learning(EWA Learning):an empirical study of Keynes's beauty contest experiment

本文研究的主要目的為研究學習行為與智能的相關性,採經驗加權吸引模型(Experience-Weighted Attraction)來描述受測者在選美競賽賽局(Beauty Contest Game)實驗下的決策及行為,不單只選擇強化學習或信念學習模型,其理由為經驗加權吸引模型綜合了以上兩個學習的特點;在智能的部分,本文以實驗所得的工作記憶能力(Working Memory Capacity)分數的高低,來代表智能高低。
  從研究結果發現,智能高與低兩類受測者的初始吸引分配,皆與其在第一期的選擇類似,而低智能的受測者在初始吸引部分,所估計的分配結果會與第一期之後期間的分配較不相似,這可能代表著低智能的受測者對於類似賽局實驗以及此實驗的分析或想像較差;在ψ的參數估計上,由於低智能的初始吸引與最後一期猜測的分配較不相似,以理論來說會遞減較快,也就是說會小於高智能所估計的結果,而真實的估計結果也顯示如此,此外,ψ的估計結果,也表示低智能受測者,對於之前吸引遞減較快,遺忘地較快。本文在δ的部分,高智能的受測者明顯地大於低智能的受測者,這表示高智能的受測者,對於失去的報酬比較敏感,會較關心沒有選擇到的數字所能得到的報酬,本文認為此結果可能隱含高智能受測者的認知階層較高。在受測者對於吸引敏感度λ的參數部分,本研究發現,此兩類受測者並無太大差異,也就是此兩類受測者對於吸引的變動,敏感度差距不大,也就代表影響人們對於吸引敏感度的原因,可能不是來自於智商高低。由本文的實證結果,可以發現學習行為與智能的相關性,因此本文建議在往後與學習行為有關的研究上,也許可以納入智能為主要研究探討的核心。 / The purpose of this paper is to explore possible relationships between individual differences in working memory capacity (WMC) and behavioral heterogeneity
revealed in a repeated beauty contest experiment. We use ‘experience-weighted attraction’ (EWA) learning, to describe the decision-making and learning behavior of
subjects in the beauty contest game (BCG) experiment. In the intelligence section, the level of the experiment from the working memory capacity score, to represent the
intelligence high or low.
We found high and low intelligent of the subject's initial attract allocated all its similar to the first period choice. In parameter estimation of ψ, due to the initial
attraction of low intelligence subject is less similar to the choice of the following periods. According to theory,ψ of the low intelligence subjects will decrease faster,
and the estimate of the results also show that. In addition, part of δ, high intelligence subjects was larger than the low intelligent subjects, which means that subjects of high intelligence, more sensitive to lost revenue. We think that may be implied the higher intelligent subjects are higher cognitive hierarchy. The sensitivity of players to attractions λ, we found that these two types of intelligent is not much difference. This
means that the impact of the sensitivity of players to attractions, may not be from the intelligent level. By the empirical results, we can find the relationship between
behavioral heterogeneity and intelligence, so we suggests that it may be incorporated into the intelligent as the main research.

Identiferoai:union.ndltd.org:CHENGCHI/G0099258022
Creators蔡明翰
Publisher國立政治大學
Source SetsNational Chengchi University Libraries
Language中文
Detected LanguageEnglish
Typetext
RightsCopyright © nccu library on behalf of the copyright holders

Page generated in 0.0025 seconds