Abstract:Human intent prediction (HIP) and human activity recognition (HAR) are important for human–robot interactions. However, human–robot interface signals are user-dependent. A classifier trained on labeled source subjects performs poorly on unlabeled target subjects. Besides, previous methods used a single learner, which may only learn a subset of features and degrade their performance on target subjects. Last, HIP and HAR require real-time computing on edge devices whose computational capabilities limit the model size. To address these issues, this paper designs an ensemble diverse hypotheses (EDH) and knowledge distillation (EDHKD) method. EDH mitigates the cross-subject divergence by training feature generators to minimize the upper bound of the classification discrepancy among multiple classifiers. EDH also maximizes the discrepancy among multiple feature generators to learn diverse and complete features. After training EDH, a lightweight student network (EDHKD) distills the knowledge from EDH to a single feature generator and classifier to significantly decrease the model size but remain accurate. The performance of EDHKD is theoretically demonstrated and experimentally validated. Results show that EDH can learn diverse features and adapt well to unknown target subjects. With only soft labels provided by EDH, the student network (EDHKD) can inherit the knowledge learned by EDH and classify unlabeled target data of a 2D moon dataset and two human locomotion datasets with the accuracy at 96.9%, 94.4%, and 97.4%, respectively, in no longer than 1 millisecond. Compared to the benchmark method, EDHKD lifts the target-domain classification accuracy by 1.3% and 7.1% in the two human locomotion datasets. EDHKD also stabilizes learning curves. Therefore, EDHKD significantly increases the generalization ability and efficiency of the HIP and HAR.

Gaussian-guided feature alignment for unsupervised cross-subject adaptation

Privacy-Preserving and Cross-Domain Human Sensing by Federated Domain Adaptation with Semantic Knowledge Correction

ContrasGAN: Unsupervised Domain Adaptation in Human Activity Recognition Via Adversarial and Contrastive Learning

Unsupervised Domain Adaptation in Activity Recognition: A GAN-Based Approach

Local and Global Alignments for Generalizable Sensor-Based Human Activity Recognition

Improving Human Activity Recognition Performance by Data Fusion and Feature Engineering.

Ensemble diverse hypotheses and knowledge distillation for unsupervised cross-subject adaptation

Unsupervised Domain Adaptation for Disguised-gait-based Person Identification on Micro-Doppler Signatures

Discriminative Feature Alignment: Improving Transferability of Unsupervised Domain Adaptation by Gaussian-guided Latent Alignment

Local Domain Adaptation for Cross-Domain Activity Recognition

XHAR: Deep Domain Adaptation for Human Activity Recognition with Smart Devices

Category-Conditional Gradient Alignment for Domain Adaptive Face Anti-Spoofing

Comparing Cross-Subject Performance on Human Activities Recognition Using Learning Models

ATFA: Adversarial Time–Frequency Attention network for sensor-based multimodal human activity recognition

Graph-Based Local Feature Adaptation for Cross-Domain Person Re-Identification

Homeomorphism Alignment for Unsupervised Domain Adaptation

Cross-Domain Detection Via Graph-Induced Prototype Alignment

Few shot learning-based fast adaptation for human activity recognition

Poster Abstract: A Multi-source Unsupervised Domain Adaptation Method for Wearable Sensor based Human Activity Recognition

Deep Generative Domain Adaptation with Temporal Attention for Cross-User Activity Recognition

Unsupervised domain adaptation via coarse-to-fine feature alignment method using contrastive learning