Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Time and Frequency Synergy for Source-Free Time-Series Domain Adaptations

Muhammad Tanzil Furqon, Mahardhika Pratama, Ary Mazharuddin Shiddiqi, Lin Liu, Habibullah Habibullah, Kutluyil Dogancay

TL;DR

TFDA is developed with a dual branch network structure fully utilizing both time and frequency features in delivering final predictions, and induces pseudo-labels based on a neighborhood concept where predictions of a sample group are aggregated to generate reliable pseudo labels.

Abstract

The issue of source-free time-series domain adaptations still gains scarce research attentions. On the other hand, existing approaches rely solely on time-domain features ignoring frequency components providing complementary information. This paper proposes Time Frequency Domain Adaptation (TFDA), a method to cope with the source-free time-series domain adaptation problems. TFDA is developed with a dual branch network structure fully utilizing both time and frequency features in delivering final predictions. It induces pseudo-labels based on a neighborhood concept where predictions of a sample group are aggregated to generate reliable pseudo labels. The concept of contrastive learning is carried out in both time and frequency domains with pseudo label information and a negative pair exclusion strategy to make valid neighborhood assumptions. In addition, the time-frequency consistency technique is proposed using the self-distillation strategy while the uncertainty reduction strategy is implemented to alleviate uncertainties due to the domain shift problem. Last but not least, the curriculum learning strategy is integrated to combat noisy pseudo labels. Our experiments demonstrate the advantage of our approach over prior arts with noticeable margins in benchmark problems.

Time and Frequency Synergy for Source-Free Time-Series Domain Adaptations

TL;DR

TFDA is developed with a dual branch network structure fully utilizing both time and frequency features in delivering final predictions, and induces pseudo-labels based on a neighborhood concept where predictions of a sample group are aggregated to generate reliable pseudo labels.

Abstract

The issue of source-free time-series domain adaptations still gains scarce research attentions. On the other hand, existing approaches rely solely on time-domain features ignoring frequency components providing complementary information. This paper proposes Time Frequency Domain Adaptation (TFDA), a method to cope with the source-free time-series domain adaptation problems. TFDA is developed with a dual branch network structure fully utilizing both time and frequency features in delivering final predictions. It induces pseudo-labels based on a neighborhood concept where predictions of a sample group are aggregated to generate reliable pseudo labels. The concept of contrastive learning is carried out in both time and frequency domains with pseudo label information and a negative pair exclusion strategy to make valid neighborhood assumptions. In addition, the time-frequency consistency technique is proposed using the self-distillation strategy while the uncertainty reduction strategy is implemented to alleviate uncertainties due to the domain shift problem. Last but not least, the curriculum learning strategy is integrated to combat noisy pseudo labels. Our experiments demonstrate the advantage of our approach over prior arts with noticeable margins in benchmark problems.

Paper Structure

This paper contains 31 sections, 4 theorems, 60 equations, 5 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Time Series Domain Adaptation
  4. Source-Free Domain Adaptation
  5. Problem Formulation
  6. Method
  7. Dual Branch Structure
  8. Pseudo-labelling Strategy
  9. Sample Selection
  10. Self-Distillation Strategy
  11. Contrastive Learning
  12. Time Frequency Consistency
  13. Uncertainty Reduction Strategy
  14. Overall Loss Function
  15. Curriculum Learning
  16. ...and 16 more sections

Key Result

Theorem 1

Suppose that Assumptions (as:1), (as:2) and (as:3) hold for some b, $\mu$ such that $\textnormal{min}_{y\in [K]}P(\left\{ x:G^*(x)=y \right\}) > max \left\{ 2/(b-1),2 \right\}\mu$. Then any minimizer $\widehat{G}$ of satisfies

Figures (5)

  • Figure 1: Performance of the model with and without frequency domain.
  • Figure 2: TFDA algorithm starts with the neighborhood pseudo-labelling strategy. After that, data samples are divided into two parts: reliable and non-reliable parts learned differently with class-balanced cross-entropy loss and label propagation loss under the curriculum learning framework. In addition, the contrastive learning framework, the self-distillation learning framework and the uncertainty learning framework are integrated.
  • Figure 3: TFDA architecture: it comprises the temporal encoder processing the temporal features and the frequency encoder mining the frequency features. The Fourier transform is applied to extract the frequency spectrum. Each encoder is assigned with its own classifier where the final output is aggregated based on the confidence of their predictions.
  • Figure 4: MF1 score results with different number of neighbors
  • Figure 5: T-SNE of SSC Dataset

Theorems & Definitions (4)

  • Theorem 1
  • Theorem 2
  • Lemma 1
  • Lemma 2