representation-learning

This paper proposes a Multi-Granularity Hypergraph Representation Learning (MGHRL) framework that adaptively generates hyperedges at multiple granularities using granular-ball splitting to capture high-order relationships in graphs, outperforming baseline models on benchmark datasets.

NOAH introduces a generative transformer model for comprehensive representation and forecasting of longitudinal multimodal patient data, enabling tasks like zero-shot classification and counterfactual simulation in clinical settings.

This paper proposes TraveL, a Transformer-based multi-view framework for learning distributional representations of paths in road networks, capturing varied traveler behaviors and regional correlations, and outperforming state-of-the-art methods in travel time estimation, path similarity, and destination prediction.

This paper experimentally demonstrates that representational disentanglement in neural networks reduces collateral damage during unlearning, supporting long-held interpretability intuitions.

SMart is a new time series representation learning framework that uses multi-phase recurrence plots recovery and a source dataset selector to enhance representation transfer from multiple datasets, showing improved performance in classification and regression tasks.

This paper proposes a source-free method to diagnose whether forgotten classes can be recovered after class unlearning, introducing the Source-Free Relearning Audit (SFRA) and a relearning score to quantify recoverability.

This paper introduces function-aware masking, a pretraining algorithm for antibody language models that aligns mask placement with functional priors, yielding significant improvements on structure and CDR-related tasks.

The paper introduces Local Reference Geometry (LRG), a lightweight post-hoc feature augmentation module that enhances local feature reliability for minority classes in imbalanced time series classification by measuring and repairing geometry failures.

The study demonstrates that different representation learning objectives recover distinct latent structures from psychometric data, with contrastive objectives enhancing teacher-child retrieval but PCA-based methods better preserving behavioral phenotype organization.

The paper introduces AVA, a framework to evaluate NLP embeddings' ability to capture ontological reasoning, finding significant limitations and challenging the assumption that strong NLP performance translates to Semantic Web competence.

This paper proposes an unsupervised framework combining manifold learning and interpretable graph embeddings to address geometric and interpretability gaps in visual representations, enhancing performance in image retrieval and GCN classification tasks.

The paper proposes SciJEPA, a citation-free framework for scientific document representation using asymmetric within-document predictive learning, and demonstrates that regularization improves performance.

Pixel Linguist II improves visual text representation learning via variable resolution training, natural image-text grounding, layout-aware rendering, and multilingual curricula, achieving state-of-the-art results and robust compression performance.

This paper explores the synergy between visual understanding and generation in unified multimodal models, showing that task-decoupled architectures and end-to-end optimization can enhance performance by turning coexistence into synergy.

The paper introduces OpEmbed, a framework for learning operational fingerprints of LLM cloud services from production incident metadata to enhance operational forecasting and reliability management in cloud environments.

This paper introduces Centroid Intervention Fusion (CIF), a projection fusion framework that unifies multilingual intervention operators to enhance cross-lingual representation learning in large language models, achieving performance gains especially for low-resource languages.

This paper introduces Drift Variation autoencoder, which uses conditional posterior flow matching to unify generative and representation learning, achieving high performance in controlled multimodal benchmarks.

This paper tests whether decodable empathy directions in LLMs can reliably shift automated empathy scores, finding that affective facet control is partial and cognitive steering is inconsistent, highlighting that detection does not imply control.

The paper proposes Winder, a phase-equivariant self-supervised learning model for cardiac cycle analysis, achieving competitive diagnostic accuracy with a parameter-efficient architecture.

This paper diagnoses category-conditional collapse in Graph-JEPA models, where standard metrics indicate healthy representations but usable instance information is absent, and proposes a repair method while proving structural reducibility in the retrieval target.