A new algorithm on unsupervised adversarial finetuning of visual encoder models via constrained optimization.

We establish the theoretical limits of clustering under the Popularity-Adjusted Block Model and show that even without edge-density signals, cluster recovery is possible by leveraging differences in intra- and inter-cluster popularity parameters

We introduce an iterative framework to jointly learn hierarchical representations of both samples and features using tree-Wasserstein distance and data-driven Haar wavelet filters.

OracleAD detects multivariate time series anomalies via prediction error and deviation from emergent latent structure, enabling interpretable, fine-grained, and root-cause diagnosis.

MetaSlot is a plug-and-play Slot Attention variant that adapts to variable object counts by deduplicating slots using a learned prototype codebook, significantly improving object-centric learning performance and interpretability.

This paper proposes a visual-text complementary guidance method to solve the problem of unsupervised class incremental learning

Underwater refractive distortion removal via unsupervised learning of the gradient of the water surface.

We propose the first comprehensive and unified benchmark for deep graph clustering, offer an open-source package named PyDGC, and point out promising research directions for DGC from extensive experimental analyses.

Our work effectively bridges Hebbian principles with explicit representation learning objectives, demonstrating considerable potential and biological plausibility.

We introduce Deep Taxonomic Networks, a deep latent variable approach that uses a complete binary-tree mixture-of-Gaussians prior in a VAE framework to discover interpretable hierarchical taxonomies and prototype clusters from unlabeled data

We enhance visible-infrared person re-identification by leveraging attribute arrays extracted using a large vision-language model.

Training PD-DL reconstruction from routine clinical images without raw data to improve fast MRI access

We makes an early attempt at fully unsupervised LLM reasoning incentivization.

AsymDSD is a self-supervised learning framework for 3D point clouds that combines masked modeling and invariance learning through latent space prediction.