An arXiv paper (2026-10-05) introduces polyview, a Python package unifying multi-view machine learning workflows in a scikit-learn-compatible framework. The release fills a gap in actively maintained Python tooling for multi-view learning, offering embedding, clustering, fusion, and augmentation under one interface. It is demonstrated on five real datasets and benchmarked against two established libraries.
A new arxiv paper introduces a multi-view clustering framework that estimates how much each view encodes sensitive attributes and aligns views asymmetrically to limit fairness degradation. It targets the problem of highly sensitive-dependent views contaminating less sensitive ones during cross-view learning. Experiments report a favorable balance between clustering quality and group fairness.
A new arxiv paper introduces a multi-view clustering framework that estimates how much each view encodes sensitive attributes and aligns views asymmetrically to limit fairness degradation. It targets the problem of highly sensitive-dependent views contaminating less sensitive ones during cross-view learning. Experiments report a favorable balance between clustering quality and group fairness.
Researchers introduce LatentVerse, a representation-analysis resource combining a web-based visual analytics platform with a command-line CLI. It unifies embedding-quality diagnostics and decomposes multimodal embeddings into shared and modality-specific components, targeting biomedical foundation-model workflows where evaluation is currently fragmented across custom scripts.
Researchers introduce LatentVerse, a representation-analysis resource combining a web-based visual analytics platform with a command-line CLI. It unifies embedding-quality diagnostics and decomposes multimodal embeddings into shared and modality-specific components, targeting biomedical foundation-model workflows where evaluation is currently fragmented across custom scripts.
An arXiv paper introduces CURE, a confounder-aware framework for multi-view urban region representation learning. Existing multi-view methods can propagate spurious correlations from shared latent factors; CURE encodes each view's regional graph, estimates and reduces a shared latent component, then fuses residual views hierarchically. Experiments on three real-world cities show improved predictive performance and robustness to missing or noisy views.
An arXiv paper introduces CURE, a confounder-aware framework for multi-view urban region representation learning. Existing multi-view methods can propagate spurious correlations from shared latent factors; CURE encodes each view's regional graph, estimates and reduces a shared latent component, then fuses residual views hierarchically. Experiments on three real-world cities show improved predictive performance and robustness to missing or noisy views.
An arXiv paper (2026-09-30) proposes an inductive framework for Gromov-Wasserstein multidimensional scaling based on barycentric distillation. A GW-MDS teacher learns a latent support and optimal transport plan, and a neural student learns an explicit out-of-sample mapping, avoiding relational-matrix construction and GW optimization at inference. Experiments show distilled models preserve teacher geometry on unseen samples and outperform direct neural GW training.
An arXiv paper (2026-09-30) proposes an inductive framework for Gromov-Wasserstein multidimensional scaling based on barycentric distillation. A GW-MDS teacher learns a latent support and optimal transport plan, and a neural student learns an explicit out-of-sample mapping, avoiding relational-matrix construction and GW optimization at inference. Experiments show distilled models preserve teacher geometry on unseen samples and outperform direct neural GW training.
TensoRAG is an open-source hobby project implementing multi-tenant/multi-domain vector compression via multilinear GSVD, based on a 2024 TU Ilmenau PhD thesis. It runs entirely offline in NumPy with no API calls, targeting RAG builders who need compressed embeddings across multiple domains. An interactive demo and agent simulator are referenced.
TensoRAG is an open-source hobby project implementing multi-tenant/multi-domain vector compression via multilinear GSVD, based on a 2024 TU Ilmenau PhD thesis. It runs entirely offline in NumPy with no API calls, targeting RAG builders who need compressed embeddings across multiple domains. An interactive demo and agent simulator are referenced.