Domain Generalization and Robustness in Human Activity Recognition and Image Classification

The field of human activity recognition and image classification is moving towards developing more robust and domain-invariant models. Researchers are exploring new architectures and techniques to improve the generalization capability of models across different domains and datasets. One of the key challenges in human activity recognition is addressing cross-user variability, and recent studies have proposed innovative solutions such as integrating anatomical correlation knowledge into graph neural networks and using variational edge feature extractors. In image classification, researchers are focusing on developing domain-invariant features through neuro-inspired neural response normalization layers and higher-order convolutional neural networks. Notably, the TAROT algorithm has shown superior performance in robust domain adaptation, and the NeuRN layer has demonstrated effectiveness in enhancing domain generalization. The paper on Domain-Adversarial Anatomical Graph Networks is particularly noteworthy for its state-of-the-art performance on the OPPORTUNITY and DSADS datasets. The paper on TAROT is also notable for its theoretical justification and superior performance on the DomainNet dataset.

Sources

Domain-Adversarial Anatomical Graph Networks for Cross-User Human Activity Recognition

Compact and Efficient Neural Networks for Image Recognition Based on Learned 2D Separable Transform

TAROT: Towards Essentially Domain-Invariant Robustness with Theoretical Justification

Activity and Subject Detection for UCI HAR Dataset with & without missing Sensor Data

NeuRN: Neuro-inspired Domain Generalization for Image Classification

Mice to Machines: Neural Representations from Visual Cortex for Domain Generalization

Higher-Order Convolution Improves Neural Predictivity in the Retina

Graph-based Online Monitoring of Train Driver States via Facial and Skeletal Features

Detecting Sybil Addresses in Blockchain Airdrops: A Subgraph-based Feature Propagation and Fusion Approach

Correlating Account on Ethereum Mixing Service via Domain-Invariant feature learning

SOS: A Shuffle Order Strategy for Data Augmentation in Industrial Human Activity Recognition

Built with on top of