- 學位論文
演化法則下的資訊評估法於人工智慧發展
Evolutionary Perspective for Approaching AI by Relative Entropy
摘要
AI是人在自我的理解下所做的發展。研究AI使得我們得到更多自我思考模式和評估的知識，在追尋自我的思考脈絡時，能更理解其自身在環境中的定位角色。現在的人工智慧的發展是基於在新的科學領域不斷出現下和對生物有更深入和廣大的研究成果而開始有比較可能性的創作，從對一些簡單的動物社群研究到人類的社群行為研究，從這些研究卻都可以找出人和動物社群有相似的地方。在比較過人和動物的社會性結構後，可以理解到什麼是必然的優式現象會發生在一個較為先進的群體。早期的人工智慧有評估人性行為、數學決策系統、潛意識、行為心理學和粒子群最佳化，這些知識背景幫助我們整理出我們的思考邏輯並做為程式的設計，然而這些背後的哲學都引領出一個關鍵的主題”什麼是合理的反應” 。 我們本身會觀察環境資料並且在一個混亂的環境中找出什麼資料和行為是相關的。簡而言之，當我們遇到特定事件時,我們會認得該資料和此事件的相關性，最後我們會建立起因果關係。然而如何分辦出環境中的大量資料之間的相關性,我們使用Information entropy來量測事件和資料的關係，這樣的關係是一種邏輯思考層，引導我們如何做出合理的反應。本研究發展出一個程式來表達這樣的關係，該程式是由粒子群最佳化(PSO)結合ENTROPY，我們利用PSO中的群體修正方式做為我們學習或建模的資料，當中的為了達到這個效果，我們使用MAX ENTROPY的概念做為PSO中的群體修正方式，並利用 Kullback–Leibler divergence數學式定義了事件和環境資料的相關性。
並列摘要
Artificial intelligence (AI) is a developing odyssey of human self-understanding. Researching the designing AI allows us to evaluate our thinking model and knowledge, we have better recognized the character of humans’ position in nature during this developing odyssey. How do we design a self-thinking AI that in current sciences must have a lot of ways to achieve new sciences that are continuously discovering. This research provides a general direction for design AI, according to views of creature evolution develops the vital necessity and features that those have better performed AI achievement. In all science relative to AI, the earlier concepts of AI have to talk about estimating the behaviors of humanity, mathematical decision systems, subconscious affecting behaviors of psychology and particle swarm optimal of computer engineering, those accumulated background knowledge lead us a direction to design a reasonable AI. Those fields help us to figure out the steps of our thinking and design the program. All of those philosophical views are considered about proper interaction, what leads to the creatures properly interacting and learning response with the environment. The truth is that those proper interactions with environmental data can be observed by our sensory organs. In a short, when we have engaged one event, we recognized what data is connecting with this event. After we constructed the relation of cause and result data or models, we can understand the cause and result relation. However, there is a lot of data, and we need to identify the relationship of different data. We use information entropy to measure happened times of data, this kind of formulation is helping us to find out the relationship between different data and events. Once we can identify the connection of events and environmental data that connection becomes a logic for judgement and guiding our interactions more properly. We have coded a program to demonstrate this dynamic environment and interacting behaviors by combined particle swarm optimization (PSO) and information entropy. Our program is presenting that the learning way of a creature is constructed by the identifying mechanism to form our logic or knowledge. The idea of what we propose is basic on Maximum entropy principle for designing the global searching ability of PSO. And Kullback–Leibler divergence function identifies the connection between events and relative data. On the other hand, we found a mathematical rule to solve the problem of what is reasonable response for AI.
參考文獻
延伸閱讀
- 鄭明淵、張于漢、Doddy Prayogo、吳建燁（2018）。演化式人工智慧建立專案實獲完工工期推論模式之研究。建築學報，(104)，73-87。https://doi.org/10.3966/101632122018060104005
- 黃開義、林旭昇（1994）。人工智慧搜尋策略在裝配程序規劃之應用。大葉學報，3(1)，21-35。https://doi.org/10.7119/JDYU.199412.0021
- 黃開義、吳泰熙、劉愷信（1997）。裝配程序之系統化評估與人工智慧規劃模式之建立。大葉學報，6(1)，73-85。https://doi.org/10.7119/JDYU.199712.0073
- 楊承祐（2021）。Explainable Artificial Intelligence: Research Framework and A Bibliometric Analysis〔碩士論文，國立政治大學〕。華藝線上圖書館。https://doi.org/10.6814/NCCU202100316
- 李崇僖（2022）。Reflections on AI Patent Issues from the Perspective of the Industrial Revolution。萬國法律，(241)，2-11。https://www.airitilibrary.com/Article/Detail?DocID=15606473-202202-202202220012-202202220012-2-11