CASL-VAE: Learning Structured Latent Variables from Unpaired Data for Semi-supervised Clustering and Paired Sample Generation

2026-07-09 · Source: Machine Learning · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Health & Medical Research · Depth: Expert, quick

Summary

CASL-VAE is a deep contrastive latent variable model designed to learn structured latent generative factors from unpaired data, addressing challenges in quantifying population variability, particularly when paired data is absent and target variation is heterogeneous. This model factorizes variation into continuous common latent factors, shared across populations, and hierarchical salient latent factors. These salient factors specifically model target-specific heterogeneity, distinguishing discrete subtypes and continuous within-subtype variation. Utilizing variational inference, CASL-VAE optimizes approximate joint likelihood across reference and target domains using unpaired data, establishing a principled foundation for generating paired samples and conducting cross-domain analysis. Validation on semi-synthetic neuroimaging data demonstrated improved subtype recovery and paired-sample generation compared to baseline clustering and generative models, also revealing biologically plausible heterogeneity in Alzheimer's disease.

Key takeaway

For AI Scientists and Research Scientists working with heterogeneous population data lacking paired samples, CASL-VAE offers a robust approach. You can utilize its ability to factorize variation into common and salient latent factors for improved subtype recovery and principled paired-sample generation. Consider applying this model to complex clinical datasets, such as those in neuroimaging or Alzheimer's research, to uncover hidden biological heterogeneity and enhance cross-domain analysis.

Key insights

CASL-VAE learns structured latent variables from unpaired data to quantify population variability and generate paired samples.

Principles

• Factorize variation into common and hierarchical salient latent factors.
• Model target-specific heterogeneity as discrete subtypes and continuous variation.
• Approximate joint likelihood optimization can use unpaired data.

Method

CASL-VAE employs variational inference to perform approximate joint likelihood optimization over reference and target domains using unpaired data, enabling paired-sample generation and cross-domain analysis.

In practice

• Improved subtype recovery in neuroimaging data.
• Enhanced paired-sample generation capabilities.
• Reveals biologically plausible heterogeneity in Alzheimer's disease.

Topics

• CASL-VAE
• Latent Variable Models
• Semi-supervised Clustering
• Paired Sample Generation
• Neuroimaging
• Alzheimer's Disease

Best for: AI Scientist, Research Scientist

Related on AIssential

• eXact-Prior Variational Autoencoder (X-VAE): Learning Data-Adaptive Gaussian Mixture Priors for Latent Distributions — X-VAE improves VAEs by using a data-adaptive Gaussian mixture prior derived from pretrained autoencoder latents, enhancing control and sample realism.
• Imputation Meets Clustering: Exploiting Latent Subgroup Structure for Missing Data Recovery — CAGI co-optimizes clustering and imputation, using dynamic cluster assignments to guide a GAN for superior missing data recovery.
• Beyond Visual Similarity: Rule-Guided Multimodal Clustering with explicit domain rules — DART-VAE integrates LLM-generated domain rules into VAEs for multimodal clustering, yielding more meaningful, interpretable results than purely data-driven methods.
• Task-guided cross-subject latent alignment: a multi-encoder-decoder VAE — MED-VAE aligns neural activity across subjects using a pretrained ANN scaffold, eliminating the need for shared stimuli.
• Recovering Latent Structures after Variational Bayesian Variable Selection: Fit Assessment and Factor-Number Selection in Partially Exploratory Factor Analysis — A post-selection framework assesses latent structures in PEFA, improving factor number and loading recovery.
• Structured Nonparametric Variational Inference for Dependent Latent Modeling — SN-VI accurately approximates complex posterior distributions by preserving latent variable dependencies using multivariate splines.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Machine Learning.