Adversarial Bi-Regressor Network for Domain Adaptive Regression

TL;DR

This work tackles domain adaptive regression for cross-domain indoor localization by introducing ABRNet, an architecture that uses a feature generator, two parallel regressors, and a discriminator to learn domain-invariant representations. The core idea is to maximize the disagreement between the two regressors to identify target samples outside the source distribution and then align features via adversarial training, augmented by intermediate domains to bridge large domain gaps. The approach yields significant improvements over strong baselines on both real-world (SPA WC2021) and synthetic (dSprites) regressive DAR tasks, with ablations confirming the efficacy of the bi-regressor discrepancy and intermediate-domain strategies. The method has practical impact for robust localization in changing environments using multi-sensor RF signals, enabling more reliable cross-domain predictions without target labels.

Abstract

Domain adaptation (DA) aims to transfer the knowledge of a well-labeled source domain to facilitate unlabeled target learning. When turning to specific tasks such as indoor (Wi-Fi) localization, it is essential to learn a cross-domain regressor to mitigate the domain shift. This paper proposes a novel method Adversarial Bi-Regressor Network (ABRNet) to seek more effective cross-domain regression model. Specifically, a discrepant bi-regressor architecture is developed to maximize the difference of bi-regressor to discover uncertain target instances far from the source distribution, and then an adversarial training mechanism is adopted between feature extractor and dual regressors to produce domain-invariant representations. To further bridge the large domain gap, a domain-specific augmentation module is designed to synthesize two source-similar and target-similar intermediate domains to gradually eliminate the original domain mismatch. The empirical studies on two cross-domain regressive benchmarks illustrate the power of our method on solving the domain adaptive regression (DAR) problem.

Figures (5)

• Figure 1: The floor map of the SPAWC2021 multi-modal indoor localization datasetArnoldSchaich21, where red squares and red circles denote, respectively, UWB and Wi-Fi anchors, while blue line denotes the path of a robot. The green box denotes the area where certain furniture was moved between offline fingerprinting and online test data collection.
• Figure 2: Overview of our proposed adversarial bi-regressors network (ABRNet) including feature generator $F(\cdot)$, two regressors $\{{\hat{G},\hat{R}}\}$, $\{{\widetilde{G},\widetilde{R}}\}$ and discriminator $D(\cdot)$. The main training process involves three stages. During the first step, ABRNet utilizes all source supervisions to train the network. For the second step, we introduce soft-similarity to maximize the bi-regressor difference. The third one aims to generate domain-invariant representations by updating feature generator and discriminator with the frozen regressors.
• Figure 3: Before adaption (a): target samples far from the source distribution are named as source-dissimilar ones; After adaption (b): target features are more aligned over the source support.
• Figure 4: Visualization of average MSEs over $50 \times 50$$\text{cm}^2$ on the floor map. The $x$-axis and $y$-axis are divided several intervals with 0.5 meter, which forms the chessboard. The number of each grid is computed by taking the average MSE of all test samples falling into the grid.
• Figure 5: Empirical Analysis: (a) Comparison of ABRNet variants, (b) Selection of parameter $\lambda$, and (c) Training stability with iterations.