Advances in Multimodal Learning and Out-of-Distribution Detection

The field of machine learning is witnessing significant developments in multimodal learning and out-of-distribution detection. Researchers are exploring innovative approaches to improve the accuracy and reliability of models in complex scenarios, such as multimodal data, distribution shifts, and limited training data. Notably, meta-learning and physics-informed methods are being leveraged to enhance model performance and adaptability. Furthermore, feature disentanglement and energy-guided calibration are being investigated to improve the robustness of time series classification models. Overall, these advancements have the potential to significantly impact various applications, including employment predictions, anomaly detection, and multimedia interactive systems. Noteworthy papers include: CTRL, which introduces a meta-learning method for improving overall accuracy and preserving source-level heterogeneity. M3OOD, a meta-learning-based framework for automatic selection of multimodal out-of-distribution detectors. ERIS, an energy-guided feature disentanglement framework for out-of-distribution time series classification.

Sources

CTRL Your Shift: Clustered Transfer Residual Learning for Many Small Datasets

Physics-Informed Diffusion Models for Unsupervised Anomaly Detection in Multivariate Time Series

M3OOD: Automatic Selection of Multimodal OOD Detectors

Multi-source Multimodal Progressive Domain Adaption for Audio-Visual Deception Detection

MAGNeT: Multimodal Adaptive Gaussian Networks for Intent Inference in Moving Target Selection across Complex Scenarios

ERIS: An Energy-Guided Feature Disentanglement Framework for Out-of-Distribution Time Series Classification

Probability Density from Latent Diffusion Models for Out-of-Distribution Detection

Built with on top of