1. Trang chủ
  2. » Tất cả

Training Complex Models with Multi-Task Weak Supervision

31 2 0

Đang tải... (xem toàn văn)

Tài liệu hạn chế xem trước, để xem đầy đủ mời bạn chọn Tải xuống

THÔNG TIN TÀI LIỆU

Cấu trúc

  • 1 Introduction

  • 2 Related Work

  • 3 Programming Machine Learning with Weak Supervision

  • 4 Modeling Multi-Task Weak Supervision

    • 4.1 A Multi-Task Weak Supervision Estimator

    • 4.2 Theoretical Analysis: Scaling with Diverse Multi-Task Supervision

    • 4.3 Extensions: Abstentions & Unipolar Sources

  • 5 Experiments

  • 6 Conclusion

  • A Problem Setup & Modeling Approach

    • A.1 Glossary of Symbols

    • A.2 Problem Setup

    • A.3 Our Approach: Modeling Multi-Task Sources

      • A.3.1 Defining a Multi-Task Source Model

      • A.3.2 Model Estimation without Ground Truth Using Inverse Covariance Structure

      • A.3.3 Handling Non-Singleton Separator Sets

      • A.3.4 Rank-One Settings

      • A.3.5 Recovering the Class Balance P & Computing P(Y|)

      • A.3.6 Predicting Labels with the Label Model

    • A.4 Example: Hierarchical Multi-Task Supervision

  • B Theoretical Results

    • B.1 Conditions for Identifiability

    • B.2 Interpreting the Main Bound

    • B.3 Proof of Theorem 1

    • B.4 Proof of Theorem ??

  • C Experimental Details

    • C.1 Data Balancing and Label Model Training Procedure

    • C.2 End Model Training Procedure

    • C.3 Dataset Statistics

    • C.4 Task Accuracies

    • C.5 Ablation Study: Unipolar Correction and Joint Modeling

Nội dung

Ngày đăng: 30/01/2022, 22:42

TÀI LIỆU CÙNG NGƯỜI DÙNG

TÀI LIỆU LIÊN QUAN