![[2602.18216] Generative Model via Quantile Assignment](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.18216] Generative Model via Quantile Assignment
Summary
The paper introduces NeuroSQL, a novel generative model that eliminates auxiliary networks, achieving superior image quality and reduced training time compared to existing models like GANs and VAEs.
Why It Matters
NeuroSQL addresses significant challenges in deep generative models by simplifying architecture and improving performance. Its ability to generate high-quality synthetic data efficiently has implications for various applications in machine learning, particularly in scenarios with limited training samples.
Key Takeaways
- NeuroSQL eliminates the need for auxiliary networks, enhancing stability and speed.
- It achieves lower mean pixel distance and better perceptual fidelity than GANs and VAEs.
- NeuroSQL is particularly effective for generating synthetic data with limited samples.
- The model leverages optimal transportation principles for latent variable representation.
- Benchmarking shows significant computational advantages over traditional methods.
Computer Science > Machine Learning arXiv:2602.18216 (cs) [Submitted on 20 Feb 2026] Title:Generative Model via Quantile Assignment Authors:Georgi Hrusanov, Oliver Y. Chén, Julien S. Bodelet View a PDF of the paper titled Generative Model via Quantile Assignment, by Georgi Hrusanov and 2 other authors View PDF Abstract:Deep Generative models (DGMs) play two key roles in modern machine learning: (i) producing new information (e.g., image synthesis) and (ii) reducing dimensionality. However, traditional architectures often rely on auxiliary networks such as encoders in Variational Autoencoders (VAEs) or discriminators in Generative Adversarial Networks (GANs), which introduce training instability, computational overhead, and risks like mode collapse. We present NeuroSQL, a new generative paradigm that eliminates the need for auxiliary networks by learning low-dimensional latent representations implicitly. NeuroSQL leverages an asymptotic approximation that expresses the latent variables as the solution to an optimal transportation problem. Specifically, NeuroSQL learns the latent variables by solving a linear assignment problem and then passes the latent information to a standalone generator. We benchmark its performance against GANs, VAEs, and a budget-matched diffusion baseline on four datasets: handwritten digits (MNIST), faces (CelebA), animal faces (AFHQ), and brain images (OASIS). Compared to VAEs, GANs, and diffusion models: (1) in terms of image quality, NeuroSQL ach...
