Markov Chain Monte Carlo: Stochastic Simulation for Bayesian Inference
暫譯: 馬可夫鏈蒙地卡羅：貝葉斯推斷的隨機模擬
Gamerman, Dani, Lopes, Hedibert F., Bambirra Gonçalves, Flávio
商品描述
Marking a pivotal moment in the evolution of Bayesian inference, the third edition of this seminal textbook on Markov Chain Monte Carlo (MCMC) methods reflects the profound transformations in both the field of Statistics and the broader landscape of data science over the past two decades. Building on the foundations laid by its first two editions, this updated volume addresses the challenges posed by modern datasets, which now span millions or even billions of observations and high-dimensional parameter spaces. While faster, approximate methods have gained traction, MCMC remains the gold standard for rigorous and reliable Bayesian inference, and this book continues to champion its relevance in the face of evolving computational paradigms.
This edition introduces significant updates and expansions, including new material on infinite-dimensional MCMC, sequential Monte Carlo methods, and adaptive algorithms. It also revisits foundational topics with fresh insights, such as iterative dynamics, mixture distributions, and data augmentation, while incorporating cutting-edge developments like Hamiltonian Monte Carlo and Dirichlet process-based methods. With a focus on both theoretical rigor and practical application, the book equips readers to navigate the complexities of modern Bayesian modeling and computation.
Features:
- Expanded coverage of sequential Monte Carlo methods, complementing MCMC with probabilistic foundations
- A brand-new chapter on infinite-dimensional MCMC, addressing advanced stochastic simulation techniques for modern Bayesian modeling
- Enhanced theoretical treatment of Markov chains on continuous state spaces, including nonhomogeneous Markov chains and adaptive algorithms
- New sections on mixture distributions and data augmentation, showcasing their power in simplifying and improving MCMC algorithms
- Detailed exploration of Hamiltonian Monte Carlo and Dirichlet process-based methods, reflecting recent advances in high-dimensional and scalable MCMC techniques
- Completely revised software section, aligning with contemporary Bayesian computation practices and tools, with accompanying R and Python codes available on GitHub
This textbook is an essential resource for statisticians, data scientists, and researchers in fields such as machine learning, artificial intelligence, and computational biology who rely on Bayesian inference for analyzing complex, high-dimensional datasets. It is equally valuable for graduate students and academics seeking a comprehensive introduction to MCMC methods, as well as practitioners looking to deepen their understanding of modern Bayesian computation. With its blend of theoretical depth and practical guidance, this third edition serves as both a foundational text and a reference for advanced applications in the ever-expanding domain of Bayesian analysis.
商品描述(中文翻譯)
標誌著貝葉斯推斷演變中的關鍵時刻，這本關於馬可夫鏈蒙地卡羅（MCMC）方法的經典教科書第三版反映了過去二十年來統計學領域及數據科學更廣泛範疇的深刻變革。在前兩版奠定的基礎上，這本更新的書籍針對現代數據集所帶來的挑戰進行了探討，這些數據集現在涵蓋了數百萬甚至數十億的觀測值和高維參數空間。儘管更快的近似方法已獲得關注，但MCMC仍然是嚴謹且可靠的貝葉斯推斷的黃金標準，而本書繼續強調其在不斷演變的計算範式中的相關性。
本版引入了重要的更新和擴展，包括有關無限維MCMC、序列蒙地卡羅方法和自適應算法的新材料。它還以全新的見解重新探討了基礎主題，如迭代動力學、混合分佈和數據增強，同時納入了最新的發展，如哈密頓蒙地卡羅和基於狄利克雷過程的方法。該書專注於理論的嚴謹性和實際應用，幫助讀者駕馭現代貝葉斯建模和計算的複雜性。
特色：
- 擴展了序列蒙地卡羅方法的涵蓋範圍，補充了MCMC的概率基礎
- 全新章節介紹無限維MCMC，針對現代貝葉斯建模的高級隨機模擬技術
- 增強了對連續狀態空間上馬可夫鏈的理論處理，包括非齊次馬可夫鏈和自適應算法
- 新增有關混合分佈和數據增強的部分，展示其在簡化和改善MCMC算法中的強大能力
- 詳細探討哈密頓蒙地卡羅和基於狄利克雷過程的方法，反映高維和可擴展MCMC技術的最新進展
- 完全修訂的軟體部分，與當代貝葉斯計算實踐和工具相一致，並提供可在GitHub上獲得的R和Python代碼
這本教科書是統計學家、數據科學家以及在機器學習、人工智慧和計算生物學等領域依賴貝葉斯推斷來分析複雜高維數據集的研究人員的重要資源。對於尋求全面介紹MCMC方法的研究生和學者，以及希望加深對現代貝葉斯計算理解的實務工作者來說，它同樣具有價值。憑藉其理論深度和實用指導的結合，這第三版既是基礎文本，也是不斷擴展的貝葉斯分析領域中高級應用的參考資料。
作者簡介
Dani Gamerman:
Ph. D. in Statistics from University of Warwick in 1987. Professor of Statistics at UFRJ from 1996 to 2019. Professor Emeritus at UFRJ since 2021. Supervises graduate students and post-doctoral researchers.
Author of the books Statistical Inference: an Integrated Approach and Building a Platform for Data-Driven Pandemic Prediction: From Data Modelling to Visualisation - The CovidLP Project, both published by Chapman & Hall. Also published books and monographs in Portuguese.
Papers published in many statistical journals including Journal of the Royal Statistical Society, Series B & C, Biometrika, Bayesian Analysis, Annals of Applied Statistics, Statistics & Computing, and also in the multidisciplinary journal Science.
Foundational and Opening Lecturer at the 2024 and 2004 editions of the World Meeting of International Society for Bayesian Analysis (ISBA) and invited lecturer at BISP, International Meeting of the Psychometric Society, International Valencia Meeting on Bayesian Statistics and International Workshop on Statistical Modeling and at various other scientific meetings.
Visiting lecturer at UFMG, Carlos III-Madrid, Connecticut, University College London and Duke. Colaborador honorífico of Universidade Rey Juan Carlos in Madrid.
Invited seminars at many universities in Brazil and abroad.
Current Associate Editor of Statistical Modeling, International Statistical Review and Environmetrics. Former member of the Boards of Directors of ISBA and ABE (Brazilian Statistical Association). Elected member of ISI.
Organizer of many scientific meetings in Brazil.
Current research interests include: dynamic models, extreme value theory, item response thoery, spatial statistics, survival analysis, and Bayesian computation.
Hedibert Freitas Lopes: Full Professor, Insper Institute of Education and Research
He is a highly accomplished scholar in statistics and econometrics, currently serving as a Full Professor at the Insper Institute of Education and Research since 2013, where he leads the Data and Decision Sciences Unit. Previously, he held a decade-long tenure at the University of Chicago Booth School of Business. From 2021 to 2023, he served as the Charles Wexler Professor in Statistics in the School of Mathematical and Statistical Sciences at Arizona State University. Earlier in his teaching career, he held positions at the Fluminense Federal University and the Federal University of Rio de Janeiro.Holding a PhD from the Institute of Statistics and Decision Sciences at Duke University, Professor Lopes has established a prolific academic record with about one hundred publications in selective journals. His research impact is substantial, evidenced by more than ten thousand citations on Google Scholar. Additionally, he has demonstrated a strong commitment to mentorship, having successfully supervised more than 40 graduate students, with several more currently ongoing.
Beyond his publications, Lopes is a recognized leader in the global statistical community, highlighted by his election as an ISI Fellow in 2020 and becoming the first Brazilian to be named an ISBA Fellow in 2024. His influence is further quantified by his extensive dissemination of knowledge, having delivered over two hundred invited talks and more than thirty short courses in universities, central banks, and other institutions around the world.
He actively leads major research initiatives, currently managing multiple projects focused on time series and high-dimensional data, and serves the academic community as a peer reviewer for over 100 international journals. He is currently an Associate Editor (AE) at the Journal of Computational and Graphical Statistics and has served for several years as AE for Bayesian Analysis, Journal of Business and Economic Statistics, and the Brazilian Journal of Probability and Statistics.
Flávio Bambirra Gonçalves: Holds a PhD in Statistics from the University of Warwick (2011). He is currently an Associate Professor of Statistics at the Federal University of Minas Gerais (UFMG), Brazil. His research interests lie broadly in Probability and Statistics, with particular emphasis on Bayesian statistics, stochastic processes, stochastic simulation, and computational statistics. His work also spans geostatistics, item response theory, and mathematical statistics.
He has been invited as a speaker to several scientific events and has undertaken multiple research visits as a visiting scholar at the Department of Statistics, University of Warwick. He maintains active research collaborations with national and international researchers and has published in leading statistics journals, including Journal of the Royal Statistical Society (Series B and C), Biometrika, Journal of the American Statistical Association, Journal of Computational and Graphical Statistics, and Statistics and Computing.
He has supervised and co-supervised several PhD and MSc students, contributing to the training of researchers in Bayesian computation, Monte Carlo methods, and statistical theory. He has been the recipient of highly competitive research grants in Brazil and served as President of the Brazilian Chapter of the International Society for Bayesian Analysis (ISBrA). His current research focuses on modern Monte Carlo methods, with particular interest in infinite-dimensional MCMC, Gaussian process models, and scalable Bayesian computation.
作者簡介(中文翻譯)
**Dani Gamerman:**
1987年於華威大學獲得統計學博士學位。自1996年至2019年擔任聯邦里約熱內盧大學（UFRJ）統計學教授。自2021年起為UFRJ名譽教授。指導研究生和博士後研究人員。
著作包括《統計推斷：綜合方法》和《建立數據驅動的疫情預測平台：從數據建模到可視化 - CovidLP專案》，均由Chapman & Hall出版。還出版了葡萄牙語的書籍和專著。
在多本統計學期刊上發表論文，包括《皇家統計學會期刊》、《生物統計學》、《貝葉斯分析》、《應用統計年鑑》、《統計與計算》，以及多學科期刊《科學》。
在2024年和2004年國際貝葉斯分析學會（ISBA）世界會議上擔任基礎和開幕講者，並受邀在BISP、心理計量學會國際會議、瓦倫西亞國際貝葉斯統計會議及統計建模國際研討會等多個科學會議上演講。
曾擔任UFMG、卡洛斯三世大學-馬德里、康乃狄克大學、倫敦大學學院和杜克大學的訪問講師。是馬德里胡安卡洛斯國王大學的榮譽合作者。
在巴西及國外多所大學受邀舉辦研討會。
目前擔任《統計建模》、《國際統計評論》和《環境計量學》的副編輯。曾任ISBA和巴西統計協會（ABE）董事會成員。當選為國際統計學會（ISI）成員。
組織了多場巴西的科學會議。
目前的研究興趣包括：動態模型、極值理論、項目反應理論、空間統計、生存分析和貝葉斯計算。
**Hedibert Freitas Lopes:**
全職教授，Insper教育與研究院
他是一位在統計學和計量經濟學領域成就卓著的學者，自2013年以來擔任Insper教育與研究院的全職教授，負責數據與決策科學單位。此前，他在芝加哥大學布斯商學院任教十年。從2021年到2023年，他擔任亞利桑那州立大學數學與統計科學學院的查爾斯·韋克斯勒統計學教授。在教學生涯早期，他曾在弗魯米嫩塞聯邦大學和里約熱內盧聯邦大學任職。
持有杜克大學統計與決策科學研究所的博士學位，Lopes教授在選定期刊上發表了約一百篇論文，建立了豐富的學術記錄。他的研究影響力顯著，Google Scholar上的引用次數超過一萬次。此外，他對指導學生表現出強烈的承諾，成功指導了超過40名研究生，並且目前仍有多名學生在指導中。
除了出版物外，Lopes在全球統計社群中是一位公認的領導者，2020年當選為ISI Fellow，並於2024年成為第一位被任命為ISBA Fellow的巴西人。他的影響力還體現在他廣泛的知識傳播上，已在全球各大學、中央銀行和其他機構發表了超過兩百場受邀演講和三十多個短期課程。
他積極領導重大研究計畫，目前管理多個專注於時間序列和高維數據的項目，並作為超過100本國際期刊的同行評審。他目前是《計算與圖形統計期刊》的副編輯，並曾擔任《貝葉斯分析》、《商業與經濟統計期刊》和《巴西概率與統計期刊》的副編輯多年。
**Flávio Bambirra Gonçalves:**
持有華威大學的統計學博士學位（2011年）。目前是巴西米納斯吉拉斯聯邦大學（UFMG）的統計學副教授。他的研究興趣廣泛涵蓋概率與統計，特別強調貝葉斯統計、隨機過程、隨機模擬和計算統計。他的工作還涉及地統計學、項目反應理論和數學統計。
他曾受邀在多個科學活動中發表演講，並作為訪問學者在華威大學統計系進行多次研究訪問。他與國內外研究人員保持活躍的研究合作，並在領先的統計學期刊上發表文章，包括《皇家統計學會期刊》（B和C系列）、《生物統計學》、《美國統計協會期刊》、《計算與圖形統計期刊》和《統計與計算》。
他指導和共同指導了多名博士和碩士生，為貝葉斯計算、蒙特卡羅方法和統計理論的研究人員培訓做出了貢獻。他曾獲得巴西的高度競爭性研究資助，並擔任國際貝葉斯分析學會（ISBrA）巴西分會的會長。他目前的研究專注於現代蒙特卡羅方法，特別關注無限維MCMC、高斯過程模型和可擴展的貝葉斯計算。
