본문

서브메뉴

Towards Structured Intelligence for Sequence Modeling
Towards Structured Intelligence for Sequence Modeling
Towards Structured Intelligence for Sequence Modeling

상세정보

자료유형  
 학위논문 서양
최종처리일시  
20260202105624
ISBN  
9798265429629
DDC  
500
저자명  
Wang, Ke Alexander.
서명/저자  
Towards Structured Intelligence for Sequence Modeling
발행사항  
[Sl] : Stanford University, 2025
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2025
형태사항  
163 p
주기사항  
Source: Dissertations Abstracts International, Volume: 87-05, Section: B.
주기사항  
Advisor: Fox, Emily.
학위논문주기  
Thesis (Ph.D.)--Stanford University, 2025.
초록/해제  
요약Learning, for both human and machine, is a search for explanations that fit our observations. This process is shaped by the priors, biases, and inductive assumptions embedded in the structure of our hypothesis space. In this thesis, we explore how introducing the right structure can guide machine learning in sequence modeling tasks. We begin with a real-world problem of modeling glucose control using free-living continuous glucose monitor data. By embedding mechanistic physiological models into the learning process, we constrain the hypothesis space to reflect our understanding of how the body regulates glucose, producing interpretable and clinically meaningful representations of glycemic control. We then turn to generic sequence modeling, where we study two ways to incorporate structure through memory. One approach introduces multiresolution memory as an inductive bias that captures hierarchical patterns common in sequence data, yielding a parameter-efficient neural network grounded in wavelet theory. Another offers a unifying framework for understanding and deriving modern neural sequence models, such as softmax attention and state-space models, viewing them as implicit algorithms for associative memory and retrieval. This thesis takes a step towards structured intelligence for sequence modeling, showing how principled architectural designs, whether drawn from domain knowledge or memory mechanisms, can produce models that are more interpretable and effective.
일반주제명  
Decomposition
일반주제명  
Glucose
일반주제명  
Diabetes
일반주제명  
Deep learning
일반주제명  
Wavelet transforms
일반주제명  
Insulin
일반주제명  
Neural networks
일반주제명  
Mathematics
일반주제명  
Medicine
일반주제명  
Pharmaceutical sciences
일반주제명  
Public health
기타저자  
Stanford University.
기본자료저록  
Dissertations Abstracts International. 87-05B.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008260126s2025        us                              c    eng  d
■001000017360826
■00520260202105624
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798265429629
■035    ▼a(MiAaPQ)AAI32316541
■035    ▼a(MiAaPQ)Stanfordng584zt1322
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a500
■1001  ▼aWang,  Ke  Alexander.
■24510▼aTowards  Structured  Intelligence  for  Sequence  Modeling
■260    ▼a[Sl]▼bStanford  University▼c2025
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2025
■300    ▼a163  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  87-05,  Section:  B.
■500    ▼aAdvisor:  Fox,  Emily.
■5021  ▼aThesis  (Ph.D.)--Stanford  University,  2025.
■520    ▼aLearning,  for  both  human  and  machine,  is  a  search  for  explanations  that  fit  our  observations.  This  process  is  shaped  by  the  priors,  biases,  and  inductive  assumptions  embedded  in  the  structure  of  our  hypothesis  space.  In  this  thesis,  we  explore  how  introducing  the  right  structure  can  guide  machine  learning  in  sequence  modeling  tasks.  We  begin  with  a  real-world  problem  of  modeling  glucose  control  using  free-living  continuous  glucose  monitor  data.  By  embedding  mechanistic  physiological  models  into  the  learning  process,  we  constrain  the  hypothesis  space  to  reflect  our  understanding  of  how  the  body  regulates  glucose,  producing  interpretable  and  clinically  meaningful  representations  of  glycemic  control.  We  then  turn  to  generic  sequence  modeling,  where  we  study  two  ways  to  incorporate  structure  through  memory.  One  approach  introduces  multiresolution  memory  as  an  inductive  bias  that  captures  hierarchical  patterns  common  in  sequence  data,  yielding  a  parameter-efficient  neural  network  grounded  in  wavelet  theory.  Another  offers  a  unifying  framework  for  understanding  and  deriving  modern  neural  sequence  models,  such  as  softmax  attention  and  state-space  models,  viewing  them  as  implicit  algorithms  for  associative  memory  and  retrieval.  This  thesis  takes  a  step  towards  structured  intelligence  for  sequence  modeling,  showing  how  principled  architectural  designs,  whether  drawn  from  domain  knowledge  or  memory  mechanisms,  can  produce  models  that  are  more  interpretable  and  effective.
■590    ▼aSchool  code:  0212.
■650  4▼aDecomposition
■650  4▼aGlucose
■650  4▼aDiabetes
■650  4▼aDeep  learning
■650  4▼aWavelet  transforms
■650  4▼aInsulin
■650  4▼aNeural  networks
■650  4▼aMathematics
■650  4▼aMedicine
■650  4▼aPharmaceutical  sciences
■650  4▼aPublic  health
■690    ▼a0800
■690    ▼a0405
■690    ▼a0564
■690    ▼a0572
■690    ▼a0573
■71020▼aStanford  University.
■7730  ▼tDissertations  Abstracts  International▼g87-05B.
■790    ▼a0212
■791    ▼aPh.D.
■792    ▼a2025
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17360826▼nKERIS▼z이  자료의  원문은  한국교육학술정보원에서  제공합니다.

미리보기

내보내기

chatGPT토론

Ai 추천 관련 도서


    신착도서 더보기
    최근 3년간 통계입니다.

    소장정보

    • 예약
    • 소재불명신고
    • 나의폴더
    • 우선정리요청
    • 비도서대출신청
    • 야간 도서대출신청
    소장자료
    등록번호 청구기호 소장처 대출가능여부 대출정보
    TF18281 전자도서 대출가능 마이폴더 부재도서신고 비도서대출신청 야간 도서대출신청

    * 대출중인 자료에 한하여 예약이 가능합니다. 예약을 원하시면 예약버튼을 클릭하십시오.

    해당 도서를 다른 이용자가 함께 대출한 도서

    관련 인기도서

    로그인 후 이용 가능합니다.