Proxy-based metric learning has enhanced semantic similarity with class representatives and exhibited noteworthy performance in deep metric learning (DML) tasks. While these methods alleviate computational demands by learning instance-to-class relationships rather than instance-to-instance relationships, they often limit features to be class-specific, thereby degrading generalization performance for unseen class. In this paper, we introduce a novel perspective called Disentangled Deep Metric Learning (DDML), grounded in the framework of information bottleneck, which applies class-agnostic regularization to existing DML methods. Unlike conventional NormSoftmax methods, which primarily emphasize distinct class-specific features, our DDML enables a diverse feature representation by seamlessly transitioning between class-specific features with the aid of class-agnostic features. It smooths decision boundaries, allowing unseen classes to have stable semantic representations in the embedding space. To achieve this, we learn disentangled representations of both class-specific and class-agnostic features in the context of DML. Empirical results demonstrate that our method addresses the limitations of conventional approaches. Our method easily integrates into existing proxy-based algorithms, consistently delivering improved performance.

Multivariate Time Series Forecasting (MTSF) remains challenging due to the need to capture complex temporal dependencies and ensure robustness against distribution shift. While Transformer-based models excel at capturing long-range patterns, and diffusion models offer generative flexibility, directly applying diffusion often leads to excessive uncertainty. To address these limitations, we propose a unified forecasting framework (IB-TransDiff) that combines global modeling based on Transformers and local refinement based on diffusion, guided by the Information Bottleneck (IB) principle. To improve generalization, we introduce a novel regularization strategy that penalizes the second moment of the conditional embedding to reduce its mutual information with the input. Experiments on multiple benchmarks demonstrate that our method consistently outperforms existing approaches in MSE and MAE. In addition, our method reduces uncertainty, achieving notable improvements in QICE, CRPS, and PICP metrics. These results highlight the value of integrating information-theoretic constraints into conditional diffusion models for robust and accurate time series forecasting.