arXiv subjects

Mariia Drozdova

Publications and source records attributed to Mariia Drozdova.

15 recordsLinked to original sources

arXiv2026

Let It Go or Learn to Self-Correct: Continuous Diffusion for Constrained Discrete Tasks

Denoising Diffusion Probabilistic Models (DDPMs) generate samples by starting from noise and repeatedly denoising while keeping each update close to the current noisy state. This behavior is effective in many continuous domains, but its role is less clear for globally constrained discrete tasks, such as Sudoku, graph connectivity, Latin squares, and N-queens. In such settings, early discrete errors can be difficult to undo. As a result, standard diffusion sampling may preserve early mistakes, even when the model's clean predictions are informative. We compare standard samplers to sampling directly from the model's clean prediction. Without retraining, this single change improves Sudoku validity from 31% to 95%, with consistent gains across the other discrete tasks. We hypothesize that staying close to the current noisy state is harmful because the reverse trajectory can drift off the forward noising distribution the model was trained on. To reduce this train-test mismatch, we further introduce self-correction training, which exposes the model to its own predictions, improving robustness to errors that arise during inference. This substantially improves the performance of standard samplers. Our results suggest that continuous diffusion models can learn nontrivial global constraints, but discrete reasoning tasks require better alignment between training and inference: either through samplers that reduce commitment to early decisions, or through training that teaches the model to correct its own inference-time errors.

cs.LG↗

arXiv2026

Diffusion as a Training Curriculum for Timestep-Free Iterative Reasoning

Diffusion models and recursive reasoners are both iterative, but they carry information across iterations differently. We add a persistent hidden state to a diffusion denoiser and remove its timestep conditioning, leaving a single shared update that can be run to arbitrary depth. The result is an anytime solver: accuracy keeps improving with inference depth far beyond the rollout lengths and backpropagation window used in training, reaching 99.90% exact solve on Sudoku-Extreme. We also obtain 98.93% solve rate on Maze-Unique. Surprisingly, progressive denoising is unnecessary at inference: holding corruption at its maximum by replacing every non-clue variable with fresh Gaussian noise at each step retains near-perfect solving and converges to stable solutions. This simple noise-injection mechanism enables a single trajectory to efficiently explore the solution space and settle on the correct answer without parallel rollouts, candidate selection, or external verifiers required by prior reasoning models. Nonetheless, ordered annealed corruption remains critical during training, which suggests that diffusion's primary contribution to our anytime solver is not a sampling procedure at inference, but a denoising training curriculum.

cs.LG↗

arXiv2026

DINOspec: Efficient Multimodal Alignment of Vision and Spectral Foundation Models for Astronomy

Astronomical observations provide multimodal views of physical systems, with images and spectra capturing complementary properties of celestial objects. Scientific foundation models can learn powerful representations from these observations, but representations learned by separate models remain difficult to combine. We investigate whether physical representations learned by separate vision and spectral models can be aligned without retraining their encoders. We introduce DINOspec, a multimodal framework that aligns a frozen DINOv3 image encoder with a pre-trained AION-1 spectral tokenizer using lightweight adapters and contrastive learning on 20,472 paired images and spectra of astronomical objects. DINOspec improves galaxy morphology classification (F1: 0.72$\rightarrow$0.78) and spectral classification (F1: 0.70$\rightarrow$0.74) while training at most 21M parameters. Improvements depend on the downstream task, revealing asymmetric transfer between independently learned representations, while spectroscopic redshift prediction remains unchanged ($R^2\approx0.9$). These results demonstrate that scientific foundation models can be composed through lightweight representation alignment.

astro-ph.IM↗

arXiv2026

Learning Radio Astronomical Representations with LeJEPA and Very Small Models

Representations learned by vision foundation models pretrained on natural images have been shown to be useful for out-of-domain astronomical images. Performance on scientific downstream tasks increases with model size, which both carries higher inference costs and limits scalability, even when considering parameter-efficient adaptation. An alternative is to learn representations directly from astronomical observations rather than natural images, through self-supervised pretraining. We evaluate LeJEPA's ability to learn robust representations using very small vision models ($\sim$6M parameters) pretrained on Radio Galaxy Zoo images, comparing with established self-supervised frameworks. We test whether LeJEPA's latent-space regularization leads to better radio galaxy morphology classification. Across three evaluation datasets, LeJEPA achieves performance comparable to a substantially larger foundation model while producing more consistent representations across training and evaluation datasets. These results suggest that the choice of representation learning objective is critical for enabling small domain-specific models to achieve performance competitive with representations transferred from large foundation models in scientific imaging.

astro-ph.IM↗

arXiv2026

Exploring Image-Text Alignment for Radio Galaxy Morphologies

We investigate whether specially constructed text captions can capture the same morphological information as radio galaxy images. Using the MiraBest dataset, we generate captions with a domain-specific prompt and evaluate their alignment with images through the SigLIP-2 vision--language model, with and without LoRA fine-tuning. Results show that caption-based classification of FR-I and FR-II galaxies performs similarly to images, with fine-tuning improving local coherence of embeddings but not global alignment.

astro-ph.IM↗

arXiv2026

Can Continuous-Time Diffusion Models Generate and Solve Globally Constrained Discrete Problems? A Study on Sudoku

Can standard continuous-time generative models represent distributions whose support is an extremely sparse, globally constrained discrete set? We study this question using completed Sudoku grids as a controlled testbed, treating them as a subset of a continuous relaxation space. We train flow-matching and score-based models along a Gaussian probability path and compare deterministic (ODE) sampling, stochastic (SDE) sampling, and DDPM-style discretizations derived from the same continuous-time training. Unconditionally, stochastic sampling substantially outperforms deterministic flows; score-based samplers are the most reliable among continuous-time methods, and DDPM-style ancestral sampling achieves the highest validity overall. We further show that the same models can be repurposed for guided generation: by repeatedly sampling completions under clamped clues and stopping when constraints are satisfied, the model acts as a probabilistic Sudoku solver. Although far less sample-efficient than classical solvers and discrete-geometry-aware diffusion methods, these experiments demonstrate that classic diffusion/flow formulations can assign non-zero probability mass to globally constrained combinatorial structures and can be used for constraint satisfaction via stochastic search.

cs.LG↗

arXiv2025

Radio Astronomy in the Era of Vision-Language Models: Prompt Sensitivity and Adaptation

Vision-Language Models (VLMs), such as recent Qwen and Gemini models, are positioned as general-purpose AI systems capable of reasoning across domains. Yet their capabilities in scientific imaging, especially on unfamiliar and potentially previously unseen data distributions, remain poorly understood. In this work, we assess whether generic VLMs, presumed to lack exposure to astronomical corpora, can perform morphology-based classification of radio galaxies using the MiraBest FR-I/FR-II dataset. We explore prompting strategies using natural language and schematic diagrams, and, to the best of our knowledge, we are the first to introduce visual in-context examples within prompts in astronomy. Additionally, we evaluate lightweight supervised adaptation via LoRA fine-tuning. Our findings reveal three trends: (i) even prompt-based approaches can achieve good performance, suggesting that VLMs encode useful priors for unfamiliar scientific domains; (ii) however, outputs are highly unstable, i.e. varying sharply with superficial prompt changes such as layout, ordering, or decoding temperature, even when semantic content is held constant; and (iii) with just 15M trainable parameters and no astronomy-specific pretraining, fine-tuned Qwen-VL achieves near state-of-the-art performance (3% Error rate), rivaling domain-specific models. These results suggest that the apparent "reasoning" of VLMs often reflects prompt sensitivity rather than genuine inference, raising caution for their use in scientific domains. At the same time, with minimal adaptation, generic VLMs can rival specialized models, offering a promising but fragile tool for scientific discovery.

astro-ph.IM↗

arXiv2024

Semi-Supervised Fine-Tuning of Vision Foundation Models with Content-Style Decomposition

In this paper, we present a semi-supervised fine-tuning approach designed to improve the performance of pre-trained foundation models on downstream tasks with limited labeled data. By leveraging content-style decomposition within an information-theoretic framework, our method enhances the latent representations of pre-trained vision foundation models, aligning them more effectively with specific task objectives and addressing the problem of distribution shift. We evaluate our approach on multiple datasets, including MNIST, its augmented variations (with yellow and white stripes), CIFAR-10, SVHN, and GalaxyMNIST. The experiments show improvements over supervised finetuning baseline of pre-trained models, particularly in low-labeled data regimes, across both frozen and trainable backbones for the majority of the tested datasets.

cs.CV↗

arXiv2024

Self-Supervised Learning on MeerKAT Wide-Field Continuum Images

Self-supervised learning (SSL) applied to natural images has demonstrated a remarkable ability to learn meaningful, low-dimension representations without labels, resulting in models that are adaptable to many different tasks. Until now, applications of SSL to astronomical images have been limited to Galaxy Zoo datasets, which require a significant amount of pre-processing to prepare sparse images centered on a single galaxy. With wide-field survey instruments at the forefront of the Square Kilometer Array (SKA) era, this approach to gathering training data is impractical. We demonstrate that continuum images from surveys like the MeerKAT Galactic Cluster Legacy Survey (MGCLS) can be successfully used with SSL, without extracting single-galaxy cutouts. Using the SSL framework DINO, we experiment with various preprocessing steps, augmentations, and architectures to determine the optimal approach for this data. We train both ResNet50 and Vision Transformer (ViT) backbones. Our models match state-of-the-art results (trained on Radio Galaxy Zoo) for FRI/FRII morphology classification. Furthermore, they predict the number of compact sources via linear regression with much higher accuracy. However, fine-tuning results in similar performance between our models, the state-of-the-art, and open-source models on multi-class morphology classification. Using source-rich crops from wide-field images to train multi-purpose models is an easily scalable approach that significantly reduces data preparation time. For the tasks evaluated in this work, twenty thousand crops is sufficient training data for models that produce results similar to state-of-the-art. In the future, complex tasks like source detection and characterization, together with domain-specific tasks, ought to demonstrate the true advantages of training models with radio astronomy data over natural-image foundation models.

astro-ph.IM↗

arXiv2024

MV-MR: multi-views and multi-representations for self-supervised learning and knowledge distillation

We present a new method of self-supervised learning and knowledge distillation based on the multi-views and multi-representations (MV-MR). The MV-MR is based on the maximization of dependence between learnable embeddings from augmented and non-augmented views, jointly with the maximization of dependence between learnable embeddings from augmented view and multiple non-learnable representations from non-augmented view. We show that the proposed method can be used for efficient self-supervised classification and model-agnostic knowledge distillation. Unlike other self-supervised techniques, our approach does not use any contrastive learning, clustering, or stop gradients. MV-MR is a generic framework allowing the incorporation of constraints on the learnable embeddings via the usage of image multi-representations as regularizers. Along this line, knowledge distillation is considered a particular case of such a regularization. MV-MR provides the state-of-the-art performance on the STL10 and ImageNet-1K datasets among non-contrastive and clustering-free methods. We show that a lower complexity ResNet50 model pretrained using proposed knowledge distillation based on the CLIP ViT model achieves state-of-the-art performance on STL10 linear evaluation. The code is available at: https://github.com/vkinakh/mv-mr

cs.CV↗

arXiv2024

Radio-astronomical Image Reconstruction with Conditional Denoising Diffusion Model

Reconstructing sky models from dirty radio images for accurate source localization and flux estimation is crucial for studying galaxy evolution at high redshift, especially in deep fields using instruments like the Atacama Large Millimetre Array (ALMA). With new projects like the Square Kilometre Array (SKA), there's a growing need for better source extraction methods. Current techniques, such as CLEAN and PyBDSF, often fail to detect faint sources, highlighting the need for more accurate methods. This study proposes using stochastic neural networks to rebuild sky models directly from dirty images. This method can pinpoint radio sources and measure their fluxes with related uncertainties, marking a potential improvement in radio source characterization. We tested this approach on 10164 images simulated with the CASA tool simalma, based on ALMA's Cycle 5.3 antenna setup. We applied conditional Denoising Diffusion Probabilistic Models (DDPMs) for sky models reconstruction, then used Photutils to determine source coordinates and fluxes, assessing the model's performance across different water vapor levels. Our method showed excellence in source localization, achieving more than 90% completeness at a signal-to-noise ratio (SNR) as low as 2. It also surpassed PyBDSF in flux estimation, accurately identifying fluxes for 96% of sources in the test set, a significant improvement over CLEAN+ PyBDSF's 57%. Conditional DDPMs is a powerful tool for image-to-image translation, yielding accurate and robust characterisation of radio sources, and outperforming existing methodologies. While this study underscores its significant potential for applications in radio astronomy, we also acknowledge certain limitations that accompany its usage, suggesting directions for further refinement and research.

astro-ph.IM↗

arXiv2021

Turbo-Sim: a generalised generative model with a physical latent space

We present Turbo-Sim, a generalised autoencoder framework derived from principles of information theory that can be used as a generative model. By maximising the mutual information between the input and the output of both the encoder and the decoder, we are able to rediscover the loss terms usually found in adversarial autoencoders and generative adversarial networks, as well as various more sophisticated related models. Our generalised framework makes these models mathematically interpretable and allows for a diversity of new ones by setting the weight of each loss term separately. The framework is also independent of the intrinsic architecture of the encoder and the decoder thus leaving a wide choice for the building blocks of the whole network. We apply Turbo-Sim to a collider physics generation problem: the transformation of the properties of several particles from a theory space, right after the collision, to an observation space, right after the detection in an experiment.

cs.LG↗

arXiv2021

Generation of data on discontinuous manifolds via continuous stochastic non-invertible networks

The generation of discontinuous distributions is a difficult task for most known frameworks such as generative autoencoders and generative adversarial networks. Generative non-invertible models are unable to accurately generate such distributions, require long training and often are subject to mode collapse. Variational autoencoders (VAEs), which are based on the idea of keeping the latent space to be Gaussian for the sake of a simple sampling, allow an accurate reconstruction, while they experience significant limitations at generation task. In this work, instead of trying to keep the latent space to be Gaussian, we use a pre-trained contrastive encoder to obtain a clustered latent space. Then, for each cluster, representing a unimodal submanifold, we train a dedicated low complexity network to generate this submanifold from the Gaussian distribution. The proposed framework is based on the information-theoretic formulation of mutual information maximization between the input data and latent space representation. We derive a link between the cost functions and the information-theoretic formulation. We apply our approach to synthetic 2D distributions to demonstrate both reconstruction and generation of discontinuous distributions using continuous stochastic networks.

cs.LG↗

arXiv2021

Information-theoretic stochastic contrastive conditional GAN: InfoSCC-GAN

Conditional generation is a subclass of generative problems where the output of the generation is conditioned by the attribute information. In this paper, we present a stochastic contrastive conditional generative adversarial network (InfoSCC-GAN) with an explorable latent space. The InfoSCC-GAN architecture is based on an unsupervised contrastive encoder built on the InfoNCE paradigm, an attribute classifier and an EigenGAN generator. We propose a novel training method, based on generator regularization using external or internal attributes every $n$-th iteration, using a pre-trained contrastive encoder and a pre-trained classifier. The proposed InfoSCC-GAN is derived based on an information-theoretic formulation of mutual information maximization between input data and latent space representation as well as latent space and generated data. Thus, we demonstrate a link between the training objective functions and the above information-theoretic formulation. The experimental results show that InfoSCC-GAN outperforms the "vanilla" EigenGAN in the image generation on AFHQ and CelebA datasets. In addition, we investigate the impact of discriminator architectures and loss functions by performing ablation studies. Finally, we demonstrate that thanks to the EigenGAN generator, the proposed framework enjoys a stochastic generation in contrast to vanilla deterministic GANs yet with the independent training of encoder, classifier, and generator in contrast to existing frameworks. Code, experimental results, and demos are available online at https://github.com/vkinakh/InfoSCC-GAN.

cs.CV↗

arXiv2020

A study of Neural networks point source extraction on simulated Fermi/LAT Telescope images

Astrophysical images in the GeV band are challenging to analyze due to the strong contribution of the background and foreground astrophysical diffuse emission and relatively broad point spread function of modern space-based instruments. In certain cases, even finding of point sources on the image becomes a non-trivial task. We present a method for point sources extraction using a convolution neural network (CNN) trained on our own artificial data set which imitates images from the Fermi Large Area Telescope. These images are raw count photon maps of 10x10 degrees covering energies from 1 to 10 GeV. We compare different CNN architectures that demonstrate accuracy increase by ~15% and reduces the inference time by at least the factor of 4 accuracy improvement with respect to a similar state of the art models.

cs.CV↗