NeFut Logo NeFut
Admin Login

[CS.AI] IIns-VAE+: A Robust Transfer Learning Framework for Wireless Sensing

Published at: 2026-09-10 22:00 Last updated: 2026-09-12 06:35
#algorithm #AI #Machine Learning

Environmental identification is essential for 6G integrated sensing and communication (ISAC) systems to achieve reliable situational awareness. Deep learning models, however, often struggle to generalize under domain shift across diverse environments. The Inter‑Instance Variational Auto‑encoder (IIns‑VAE) learns rich latent representations, yet its neural classifier remains vulnerable to distribution changes. To address this, we propose IIns‑VAE+, a hybrid model that integrates IIns‑VAE with Minimax Risk Classifiers (MRC) to enhance adaptability in transfer learning scenarios.

Key techniques

Experimental setup We evaluate on a public wireless channel dataset across three transfer learning tasks: general‑to‑specific room environments, high‑to‑low label resolution, and mixed‑to‑specific environments. Baselines include vanilla IIns‑VAE, standard CNN, and adversarially trained models. Metrics are classification accuracy and F1 score.

Results In the general→specific room task, IIns‑VAE+ improves accuracy by ~8%; in the label resolution reduction task, by ~12%; and in the mixed→specific environment task, by ~10%. The gains stem from MRC’s explicit modeling of distribution uncertainty, enabling a decision boundary that is robust to worst‑case shifts in the latent space.

Review This work demonstrates that combining variational auto‑encoder representation learning with minimax risk classification yields a powerful approach to handle environmental domain shifts in wireless sensing, offering a practical pathway for robust perception in future 6G ISAC deployments.

Original Source: https://arxiv.org/abs/2609.06131

[h] Back to Home