본문

서브메뉴

Tree-Based Models for Learning Complex Distributions
Tree-Based Models for Learning Complex Distributions
Tree-Based Models for Learning Complex Distributions

Detailed Information

자료유형  
 학위논문 서양
최종처리일시  
20260202102953
ISBN  
9798315705444
DDC  
574
저자명  
de Marchi, Daniel.
서명/저자  
Tree-Based Models for Learning Complex Distributions
발행사항  
[Sl] : The University of North Carolina at Chapel Hill, 2024
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2024
형태사항  
83 p
주기사항  
Source: Dissertations Abstracts International, Volume: 86-11, Section: B.
주기사항  
Advisor: Kosorok, Michael.
학위논문주기  
Thesis (Ph.D.)--The University of North Carolina at Chapel Hill, 2024.
초록/해제  
요약Tree-based models are some of the most successful machine learning models. Their simplicity makes them considerably faster to tune and more robust than deep learning, leading to their dominant performance on tabular datasets. We investigate the use of tree-based models on two problem types. The first is feature importance in nonlinear settings using Shapley values and Shapley-inspired methods. While model-based Shapley values might be accurate explainers of model predictions, machine learning models themselves are often poor explainers of the DGP even if the model is highly accurate. We introduce a novel metric, Shapley Marginal Surplus for Strong Models, that samples the space of possible models to come up with a truly explanatory measure of feature importance. We compare this method to other popular feature importance methods, both Shapley-based and non-Shapley based, and demonstrate significant outperformance relative to other methods. We also provide potential extensions using Bayesian optimization and online reinforcement learning to improve performance. The second problem is using a tree-based representation of a bargaining game to predict coalition formation in parliamentary governments. Represented as extensive form bargaining games, single trials of these games have orders of magnitude more states than existing game theory challenge problems such as poker. Using a heavily coarsened game tree and a novel constrained fictitious play algorithm for most of the actors in the model, we reduce these incalculably large game trees to manageable dimensions and then find optimal coalitions using Monte Carlo counterfactual regret (MCCFR) tree search.
일반주제명  
Biostatistics
일반주제명  
Computer engineering
키워드  
Machine learning models
키워드  
Shapley values
키워드  
Online reinforcement learning
기타저자  
The University of North Carolina at Chapel Hill Biostatistics
기본자료저록  
Dissertations Abstracts International. 86-11B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2024        us                              c    eng  d
■001000017356563
■00520260202102953
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798315705444
■035    ▼a(MiAaPQ)AAI31767826
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a574
■1001  ▼ade  Marchi,  Daniel.
■24510▼aTree-Based  Models  for  Learning  Complex  Distributions
■260    ▼a[Sl]▼bThe  University  of  North  Carolina  at  Chapel  Hill▼c2024
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2024
■300    ▼a83  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  86-11,  Section:  B.
■500    ▼aAdvisor:  Kosorok,  Michael.
■5021  ▼aThesis  (Ph.D.)--The  University  of  North  Carolina  at  Chapel  Hill,  2024.
■520    ▼aTree-based  models  are  some  of  the  most  successful  machine  learning  models.  Their  simplicity  makes  them  considerably  faster  to  tune  and  more  robust  than  deep  learning,  leading  to  their  dominant  performance  on  tabular  datasets.  We  investigate  the  use  of  tree-based  models  on  two  problem  types.  The  first  is  feature  importance  in  nonlinear  settings  using  Shapley  values  and  Shapley-inspired  methods.  While  model-based  Shapley  values  might  be  accurate  explainers  of  model  predictions,  machine  learning  models  themselves  are  often  poor  explainers  of  the  DGP  even  if  the  model  is  highly  accurate.  We  introduce  a  novel  metric,  Shapley  Marginal  Surplus  for  Strong  Models,  that  samples  the  space  of  possible  models  to  come  up  with  a  truly  explanatory  measure  of  feature  importance.  We  compare  this  method  to  other  popular  feature  importance  methods,  both  Shapley-based  and  non-Shapley  based,  and  demonstrate  significant  outperformance  relative  to  other  methods.  We  also  provide  potential  extensions  using  Bayesian  optimization  and  online  reinforcement  learning  to  improve  performance.  The  second  problem  is  using  a  tree-based  representation  of  a  bargaining  game  to  predict  coalition  formation  in  parliamentary  governments.  Represented  as  extensive  form  bargaining  games,  single  trials  of  these  games  have  orders  of  magnitude  more  states  than  existing  game  theory  challenge  problems  such  as  poker.  Using  a  heavily  coarsened  game  tree  and  a  novel  constrained  fictitious  play  algorithm  for  most  of  the  actors  in  the  model,  we  reduce  these  incalculably  large  game  trees  to  manageable  dimensions  and  then  find  optimal  coalitions  using  Monte  Carlo  counterfactual  regret  (MCCFR)  tree  search.
■590    ▼aSchool  code:  0153.
■650  4▼aBiostatistics
■650  4▼aComputer  engineering
■653    ▼aMachine  learning  models
■653    ▼aShapley  values
■653    ▼aOnline  reinforcement  learning
■690    ▼a0308
■690    ▼a0464
■690    ▼a0800
■71020▼aThe  University  of  North  Carolina  at  Chapel  Hill▼bBiostatistics.
■7730  ▼tDissertations  Abstracts  International▼g86-11B.
■790    ▼a0153
■791    ▼aPh.D.
■792    ▼a2024
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17356563▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

Preview

Export

ChatGPT Discussion

AI Recommended Related Books


    New Books MORE
    Statistics for the past 3 years. Go to brief

    Buch Status

    • Reservierung
    • frei buchen
    • Meine Mappe
    • Erste Aufräumarbeiten Anfrage
    • Non-Book Loan Application
    • Nighttime Book Loan Application
    Sammlungen
    Registrierungsnummer callnumber Standort Verkehr Status Verkehr Info
    TF15131 전자도서 대출가능 My Folder 부재도서신고 비도서대출신청 야간 도서대출신청

    * Kredite nur für Ihre Daten gebucht werden. Wenn Sie buchen möchten Reservierungen, klicken Sie auf den Button.

    Books borrowed together with this book

    Related Popular Books

    Available after logging in.