Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Weakly Supervised Label Learning Flows

You Lu, Wenzhuo Song, Chidubem Arachie, Bert Huang

TL;DR

This paper develops label learning flows (LLF), a general framework for weakly supervised learning problems, and develops a training method for LLF that trains the conditional flow inversely and avoids estimating the labels.

Abstract

Supervised learning usually requires a large amount of labelled data. However, attaining ground-truth labels is costly for many tasks. Alternatively, weakly supervised methods learn with cheap weak signals that only approximately label some data. Many existing weakly supervised learning methods learn a deterministic function that estimates labels given the input data and weak signals. In this paper, we develop label learning flows (LLF), a general framework for weakly supervised learning problems. Our method is a generative model based on normalizing flows. The main idea of LLF is to optimize the conditional likelihoods of all possible labelings of the data within a constrained space defined by weak signals. We develop a training method for LLF that trains the conditional flow inversely and avoids estimating the labels. Once a model is trained, we can make predictions with a sampling algorithm. We apply LLF to three weakly supervised learning problems. Experiment results show that our method outperforms many baselines we compare against.

Weakly Supervised Label Learning Flows

TL;DR

This paper develops label learning flows (LLF), a general framework for weakly supervised learning problems, and develops a training method for LLF that trains the conditional flow inversely and avoids estimating the labels.

Abstract

Supervised learning usually requires a large amount of labelled data. However, attaining ground-truth labels is costly for many tasks. Alternatively, weakly supervised methods learn with cheap weak signals that only approximately label some data. Many existing weakly supervised learning methods learn a deterministic function that estimates labels given the input data and weak signals. In this paper, we develop label learning flows (LLF), a general framework for weakly supervised learning problems. Our method is a generative model based on normalizing flows. The main idea of LLF is to optimize the conditional likelihoods of all possible labelings of the data within a constrained space defined by weak signals. We develop a training method for LLF that trains the conditional flow inversely and avoids estimating the labels. Once a model is trained, we can make predictions with a sampling algorithm. We apply LLF to three weakly supervised learning problems. Experiment results show that our method outperforms many baselines we compare against.
Paper Structure (32 sections, 1 theorem, 25 equations, 6 figures, 7 tables)

This paper contains 32 sections, 1 theorem, 25 equations, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Weakly Supervised Learning.
  4. Conditional Normalizing Flows.
  5. Proposed Method
  6. Case Studies
  7. Weakly Supervised Classification
  8. Weakly Supervised Regression
  9. Unpaired Point Cloud Completion
  10. Related Work
  11. Weakly Supervised Learning.
  12. Normalizing Flows.
  13. Point Cloud Modeling.
  14. Empirical Study
  15. Weakly Supervised Classification
  16. ...and 17 more sections

Key Result

Theorem 1

Suppose that for any $i$, $\Omega^*_i$ satisfies that $\hat{\mathbf{y}}_i \in \Omega_i^*$, and for any two $\hat{\mathbf{y}}_i \not= \hat{\mathbf{y}}_j$, the $\Omega^*_i$ and $\Omega^*_j$ are disjoint. The volume of each $\Omega^*_i$ is bounded such that $\frac{1}{|\Omega^*_i|} \le M$, where $M$ is

Figures (6)

  • Figure 1: Evolution of accuracy, likelihood and violation of weak signal constraints. Training with likelihood makes LLF accumulate more probability mass to the constrained space, so that the generated $\mathbf{y}$ are more likely to be within $\Omega$, and the predictions are more accurate.
  • Figure 2: Random sample point clouds generated by different methods. The point clouds generated by LLF are as realistic as mm-pc2pc. Mm-pc2pc has a higher diversity in samples. However, sometimes it may generate unreasonable or invalid shapes. Shape-inversion and KT-net cannot generate qualified complete shapes, when the input has limited information, e.g., the input is only a lampshade.
  • Figure C.1: Model architecture of LLF for unpaired point cloud completion. The $E$ represents the encoder, and the $D$ represents the GAN discriminator.
  • Figure C.2: Random chair samples generated by LLF. The first row is partial point clouds, and the second row is generated complete point clouds.
  • Figure C.3: Random lamp samples generated by LLF.
  • ...and 1 more figures

Theorems & Definitions (1)

  • Theorem 1