서브메뉴
검색
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
- 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.


