본문

서브메뉴

Resource-Efficient Machine Learning Systems: From Natural Behavior to Natural Language
Resource-Efficient Machine Learning Systems: From Natural Behavior to Natural Language
Resource-Efficient Machine Learning Systems: From Natural Behavior to Natural Language

상세정보

자료유형  
 학위논문 서양
최종처리일시  
20250211152034
ISBN  
9798383198735
DDC  
616
저자명  
Biderman, Dan.
서명/저자  
Resource-Efficient Machine Learning Systems: From Natural Behavior to Natural Language
발행사항  
[Sl] : Columbia University, 2024
발행사항  
Ann Arbor : ProQuest Dissertations & Theses, 2024
형태사항  
233 p
주기사항  
Source: Dissertations Abstracts International, Volume: 86-01, Section: A.
주기사항  
Advisor: Cunningham, John P.
학위논문주기  
Thesis (Ph.D.)--Columbia University, 2024.
초록/해제  
요약Contemporary machine learning models exhibit unprecedented performance in the text, vision, and time-series domains, but at the cost of significant computational and human resources. Applying these technologies for science requires balancing accuracy and resource allocation, which I investigate here via three unique case studies.In Chapter 1, I present a deep learning system for animal pose estimation from video. Existing approaches rely on frame-by-frame supervised deep learning, which requires extensive manual labeling, fails to generalize to data far outside of its training set, and occasionally produces scientifically-critical errors that are hard to detect. The solution proposed here includes semi-supervised learning on unlabeled videos, video-centric network architectures, and a post-processing step that combines network ensembling and state-space modeling. These methods improve performance both with scarce and abundant labels, and are implemented in an easy-to-use software package and cloud application. In Chapter 2, I turn to the Gaussian process, a canonical nonparametric model, known for its poor scaling with dataset size. Existing methods accelerate Gaussian processes at the cost of modeling biases. I analyze two common techniques -- early truncated conjugate gradients and random Fourier features -- showing that they find hyperparameters that underfit and overfit the data, respectively. I then propose to eliminate these biases in exchange of increased variance, via randomized truncation estimators. In In Chapter 3, I investigate continual learning, or "finetuning", in large language models (LLMs) with billions of weights. Training these models requires more memory than typically available in academic clusters. Low-Rank Adaptation (LoRA) is a widely-used technique that saves memory by training only low rank perturbations to selected weight matrices in a so-called "base model'". I compare the performance of LoRA and full finetuning on two target domains, programming and mathematics, across different data regimes. I find that in most common settings, LoRA underperforms full finetuning, but it nevertheless exhibits a desirable form of regularization: it better maintains the base model's performance on tasks outside the target domain. I then propose best practices for finetuning with LoRA.In summary, applying state-of-the-art models to large scientific datasets necessitates taking computational shortcuts. This thesis highlights the implications of these shortcuts and emphasizes the need for careful empirical and theoretical investigation to find favorable trade-offs between accuracy and resource allocation.
일반주제명  
Neurosciences
일반주제명  
Statistics
일반주제명  
Computer science
일반주제명  
Information science
키워드  
Computer vision
키워드  
Gaussian processes
키워드  
Large language models
키워드  
Neuroethology
키워드  
Pose estimation
키워드  
Software systems
기타저자  
Columbia University Neurobiology and Behavior
기본자료저록  
Dissertations Abstracts International. 86-01A.
전자적 위치 및 접속  
로그인 후 원문을 볼 수 있습니다.

MARC

 008250123s2024        us                              c    eng  d
■001000017162625
■00520250211152034
■006m          o    d                
■007cr#unu||||||||
■020    ▼a9798383198735
■035    ▼a(MiAaPQ)AAI31334959
■040    ▼aMiAaPQ▼cMiAaPQ
■0820  ▼a616
■1001  ▼aBiderman,  Dan.
■24510▼aResource-Efficient  Machine  Learning  Systems:  From  Natural  Behavior  to  Natural  Language
■260    ▼a[Sl]▼bColumbia  University▼c2024
■260  1▼aAnn  Arbor▼bProQuest  Dissertations  &  Theses▼c2024
■300    ▼a233  p
■500    ▼aSource:  Dissertations  Abstracts  International,  Volume:  86-01,  Section:  A.
■500    ▼aAdvisor:  Cunningham,  John  P.
■5021  ▼aThesis  (Ph.D.)--Columbia  University,  2024.
■520    ▼aContemporary  machine  learning  models  exhibit  unprecedented  performance  in  the  text,  vision,  and  time-series  domains,  but  at  the  cost  of  significant  computational  and  human  resources.  Applying  these  technologies  for  science  requires  balancing  accuracy  and  resource  allocation,  which  I  investigate  here  via  three  unique  case  studies.In  Chapter  1,  I  present  a  deep  learning  system  for  animal  pose  estimation  from  video.  Existing  approaches  rely  on  frame-by-frame  supervised  deep  learning,  which  requires  extensive  manual  labeling,  fails  to  generalize  to  data  far  outside  of  its  training  set,  and  occasionally  produces  scientifically-critical  errors  that  are  hard  to  detect.  The  solution  proposed  here  includes  semi-supervised  learning  on  unlabeled  videos,  video-centric  network  architectures,  and  a  post-processing  step  that  combines  network  ensembling  and  state-space  modeling.  These  methods  improve  performance  both  with  scarce  and  abundant  labels,  and  are  implemented  in  an  easy-to-use  software  package  and  cloud  application.  In  Chapter  2,  I  turn  to  the  Gaussian  process,  a  canonical  nonparametric  model,  known  for  its  poor  scaling  with  dataset  size.  Existing  methods  accelerate  Gaussian  processes  at  the  cost  of  modeling  biases.  I  analyze  two  common  techniques  --  early  truncated  conjugate  gradients  and  random  Fourier  features  --  showing  that  they  find  hyperparameters  that  underfit  and  overfit  the  data,  respectively.  I  then  propose  to  eliminate  these  biases  in  exchange  of  increased  variance,  via  randomized  truncation  estimators.  In  In  Chapter  3,  I  investigate  continual  learning,  or  "finetuning",  in  large  language  models  (LLMs)  with  billions  of  weights.  Training  these  models  requires  more  memory  than  typically  available  in  academic  clusters.  Low-Rank  Adaptation  (LoRA)  is  a  widely-used  technique  that  saves  memory  by  training  only  low  rank  perturbations  to  selected  weight  matrices  in  a  so-called  "base  model'".  I  compare  the  performance  of  LoRA  and  full  finetuning  on  two  target  domains,  programming  and  mathematics,  across  different  data  regimes.  I  find  that  in  most  common  settings,  LoRA  underperforms  full  finetuning,  but  it  nevertheless  exhibits  a  desirable  form  of  regularization:  it  better  maintains  the  base  model's  performance  on  tasks  outside  the  target  domain.  I  then  propose  best  practices  for  finetuning  with  LoRA.In  summary,  applying  state-of-the-art  models  to  large  scientific  datasets  necessitates  taking  computational  shortcuts.  This  thesis  highlights  the  implications  of  these  shortcuts  and  emphasizes  the  need  for  careful  empirical  and  theoretical  investigation  to  find  favorable  trade-offs  between  accuracy  and  resource  allocation.
■590    ▼aSchool  code:  0054.
■650  4▼aNeurosciences
■650  4▼aStatistics
■650  4▼aComputer  science
■650  4▼aInformation  science
■653    ▼aComputer  vision
■653    ▼aGaussian  processes
■653    ▼aLarge  language  models
■653    ▼aNeuroethology
■653    ▼aPose  estimation
■653    ▼aSoftware  systems
■690    ▼a0317
■690    ▼a0800
■690    ▼a0463
■690    ▼a0984
■690    ▼a0723
■71020▼aColumbia  University▼bNeurobiology  and  Behavior.
■7730  ▼tDissertations  Abstracts  International▼g86-01A.
■790    ▼a0054
■791    ▼aPh.D.
■792    ▼a2024
■793    ▼aEnglish
■85640▼uhttp://www.riss.kr/pdu/ddodLink.do?id=T17162625▼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
    TF14086 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.