Impact of Label Noise on Learning Complex Features
원문 초록
Neural networks trained with stochastic gradient descent exhibit an inductive bias towards simpler decision boundaries, typically converging to a narrow family of functions, and often fail to capture more complex features. This phenomenon raises concerns about the capacity of deep models to adequately learn and represent real-world datasets. Traditional approaches such as explicit regularization, data augmentation, architectural modifications, etc., have largely proven ineffective in encouraging the models to learn diverse features. In this work, we investigate the impact of pre-training models with noisy labels on the dynamics of SGD across various architectures and datasets. We show that pretraining promotes learning complex functions and diverse features in the presence of noise. Our experiments demonstrate that pre-training with noisy labels encourages gradient descent to find alternate minima that do not solely depend upon simple features, rather learns more complex and broader set of features, without hurting performance.
함께 살펴볼 논문
공통 태그를 근거로 연결했습니다. 실제 인용 관계나 논문의 우수성을 뜻하지 않습니다.
입력 경로는 소형 트랜스포머의 퓨샷 바인딩을 형성하지만 제로샷은 아니다: 완전 열거 가능한 연구
Input Pathways Shape Few-Shot, Not Zero-Shot, Binding in Tiny Transformers: A Fully-Enumerable Study
Coachable Agents for Interactive Gameplay
VIGOR: 모델 기반 강화학습에서 잠재 공간 일관성을 통한 제로샷 시각적 일반화
VIGOR: Zero-Shot Visual Generalization via Latent-Space Consistency in Model-Based Reinforcement Learning
Spectral Margins and Memory Routing in Looped Transformers
Ocular Verification for Virtual Reality
Solver-Aware Decompositions for Programming-by-Example: When Dividing Requires Knowing How to Conquer
토론
이 논문과 연결된 포럼 글이 아직 없습니다.