Guillermo Cabrera-Vives

The Legacy Survey of Space and Time (LSST) on the Vera C. Rubin Observatory will generate a massive collection of time series (light curves) of the measured flux of transient and variable astronomical objects. With each new flux observation, light curve classifiers need to generate updated probability distributions over candidate classes, which will then be shared with the global community for the purpose of identifying interesting targets for follow-up observations as well as less time-sensitive analysis applications. Using the synthetic light curves and classification results of participating classifiers from the Extended LSST Astronomical Time-series Classification Challenge (ELAsTiCC), we investigate a novel framework to enhance existing light curve classifications by incorporating their classification histories and the temporal evolution of these histories. To demonstrate the potential of this approach, we introduce a model that combines a recurrent neural network and an additive attention module, which shows improved classification accuracy and more balanced precision–recall performance compared to existing classifiers from the challenge. Furthermore, at this stage, most, if not all, of the existing classifiers are evaluated by their final classification results on complete light curves; we propose new metrics that evaluate the stability, accuracy, and early classification performance of a classifier’s predictions when using limited data by considering the Wasserstein distance between the temporally evolving classification probability distributions. Our metrics offer a more comprehensive perspective for model assessment by supplementing classical methods such as the confusion matrix and precision–recall.

We developed a text-to-structured query language (SQL) system based on large language models (LLMs) using in-context learning and applied it to the Automatic Learning for the Rapid Classification of Events (ALeRCE) astronomical database. ALeRCE is a community broker for the Zwicky Transient Facility and the Vera C. Rubin Observatory. The system enables users to query the database in natural language (NL) and generates executable SQL queries. To develop and evaluate the system, we constructed a set of 110 NL-SQL pairs. We propose a step-by-step generation framework comprising four modules: schema linking, query classification, prompt decomposition, and self-correction. The performance of 13 LLMs was evaluated using in-context learning and prompt engineering techniques. Text-to-SQL performance was assessed using the perfect-match (PM) rate for row identifiers (e.g., object identifiers) and column identifiers (i.e., column names). The proposed step-by-step framework consistently outperforms a direct-inference baseline, while the self-correction module consistently reduces execution errors. For Claude Opus 4.6, PM performance on row (column) identifiers is high for simple queries, reaching 0.97 (0.94), and decreases with query complexity to 0.44 (0.72) for medium queries and 0.59 (0.49) for hard queries. Among the 13 models evaluated, the best-performing LLMs for the text-to-SQL task are Claude Opus 4.6, Gemini 2.5 Pro, Gemini 3 Flash, and GPT-5.2-Codex.

We applied pretrained vision transformers (ViTs), originally developed for image recognition, to the analysis of astronomical spectral data. By converting traditional 1D spectra into 2D image representations, we enable ViTs to capture both local and global spectral features through spatial self-attention. We fine-tuned a ViT pretrained on ImageNet using millions of spectra from the Sloan Digital Sky Survey (SDSS; Kollmeier et al. 2019, in BAAS, 51, 274; Stoughton et al. 2002, AJ, 123, 485) and Large Sky Area Multi-Object Fiber Spectroscopic Telescope (LAMOST; Luo et al. 2015, Res. Astron. Astrophys., 15, 1095) surveys, represented as spectral plots. Our model is evaluated on key tasks including stellar object classification and redshift ( z ) estimation, where it demonstrates strong performance and scalability. We achieved classification accuracy higher than that of support vector machines and random forests, and we attain R ² values comparable to AstroCLIP’s spectrum encoder, even when generalizing across diverse object types. These results demonstrate the effectiveness of using pretrained vision models for spectroscopic data analysis. To our knowledge, this is the first application of ViTs to large-scale astronomical datasets, which also leverages real spectroscopic data and does not rely on synthetic inputs.

Context. Galaxy clusters grow through the accretion of galaxies from groups, filaments, and other clusters. During this process, galaxies may undergo pre-processing in lower density environments, where galaxy–galaxy mergers and other interactions can significantly alter their properties prior to cluster infall. Aims. We investigate the role of galaxy mergers in the pre-processing of galaxies prior to cluster infall by studying the spatial distribution of galaxy mergers across the cosmic web. Methods. We used a sample of 43 922 galaxies being targeted by the 4MOST CHANCES survey in and around 33 low-redshift clusters ( z < 0.07). Using Zoobot, a deep-learning framework trained on Galaxy Zoo data, we identified 698 galaxy mergers. We measured their distances to cosmic web filaments and compared those results to those of non-merging galaxies. Results. We find that galaxy mergers are significantly closer to filaments than the non-merging galaxy population, with this trend being strongest beyond the cluster virial radius. This suggests that filaments provide conditions conducive to mergers, possibly moderating relative velocities and enhancing gas availability. Conclusions. Our findings support a scenario in which filaments play a key role in transforming (pre-processing) galaxies by promoting mergers before entering the cluster cores where star formation is quenched.

Context . Foundational models have emerged as a powerful paradigm within the deep learning field. Their capacity relies on the ability to learn robust representations from large-scale datasets and generalize to diverse downstream applications, such as classification. In this paper, we present Astromer 2, a foundational model designed for extracting light curve embeddings. Aims . We introduce Astromer 2, an enhanced iteration of our self-supervised model for light curve analysis. This paper highlights the advantages of its pretrained embeddings, compares its performance with that of its predecessor, Astromer 1, and provides a detailed empirical analysis of its capabilities, offering deeper insights into the model’s representations. Methods . Astromer 2 is pretrained on 1.5 million single-band light curves from the MACHO survey using a self-supervised learning task that predicts randomly masked observations within sequences. Finetuning on a smaller labeled dataset allows us to assess its performance in classification tasks. The quality of the embeddings is measured by the F1 score of an multilayer perceptron (MLP) classifier trained on Astromer-generated embeddings. Results . Our results demonstrate that Astromer 2 significantly outperforms Astromer 1 across all evaluated scenarios, including limited datasets of 20,100, and 500 samples per class. The use of weighted per-sample embeddings, which integrate intermediate representations from Astromer’s attention blocks, is particularly impactful. Notably, Astromer 2 achieves a 15% improvement in F1 score on the ATLAS dataset compared to prior models, showcasing robust generalization to new datasets. This enhanced performance, especially with minimal labeled data, underscores the potential of Astromer 2 for more efficient and scalable light curve analysis.

The Response Evaluation Criteria in Solid Tumors (RECIST 1.1) protocol is the gold standard for assessing treatment response in oncological clinical trials and routine practice. It requires radiologists to review and select appropriate target lesions and perform precise diameter measurements, making the process labor-intensive and variable. Artificial Intelligence (AI) holds great promise for automating this workflow, but progress is hindered by the lack of public datasets with comprehensive lesion annotations and RECIST-compliant measurements. We address this gap by presenting a dataset of 1,246 manually segmented lesions from 58 CT scans of 22 cancer patients treated at the Clinical Hospital of the University of Chile (HCUCH). All cases were evaluated under RECIST 1.1, with diameter measurements reported for 82 target lesions. This resource supports diverse applications, including validating automated RECIST tools, applying radiomics to study metastatic heterogeneity, benchmarking segmentation algorithms, and advancing foundation models in medical imaging. By including data from a Latin American institution, this dataset also promotes global representation in the development of generalizable medical AI tools.

Context . The advent of large-scale sky surveys, such as the Vera C. Rubin Observatory Legacy Survey of Space and Time (LSST), will generate vast volumes of photometric data, necessitating automatic classification of light curves to identify variable stars and transient events. However, challenges such as irregular sampling, multi-band observations, and diverse flux distributions across bands demand advanced models for accurate classification. Aims . This study investigates the potential of a pre-trained vision Transformer (VT) model, specifically the Swin Transformer V2 (SwinV2), to classify photometric light curves without the need for feature extraction or multi-band preprocessing. The goal is to assess whether this image-based approach can accurately differentiate astronomical phenomena and if it can serve as a viable option for working with multi-band photometric light curves. Methods . We transformed each multi-band light curve into an image. These images served as input to the SwinV2 model, which was pre-trained on ImageNet-21K. The datasets employed include the public Catalog of Variable Stars from the Massive Compact Halo Object (MACHO) survey, using both one and two bands, and the first round of the recent Extended LSST Astronomical Time-Series Classification Challenge (ELAsTiCC), which includes six bands. The model’s performance was evaluated based on six classes for the MACHO dataset and 20 distinct classes of variable stars and transient events for the ELAsTiCC dataset. Results . The fine-tuned SwinV2 model achieved a better performance than models specifically designed for light curves, such as Astromer and the Astronomical Transformer for time series And Tabular data (ATAT). When trained on the “full dataset” of MACHO, it attained a macro F1-score of 80.2% and outperformed Astromer in single-band experiments. Incorporating a second band further improved performance, increasing the F1-score to 84.1%. In the ELAsTiCC dataset, SwinV2 achieved a macro F1-score of 65.5%, slightly surpassing ATAT by 1.3%. Conclusions . SwinV2, a pre-trained VT model, effectively classifies photometric light curves. It outperforms traditional models and offers a promising approach for large-scale surveys. This highlights the potential of using visual representations of light curves, with future prospects including the integration of tabular data, textual information, and multi-modal learning to enhance analysis and classification in time-domain astronomy.

Multiband astronomical time series exhibit heterogeneous variability patterns, sampling cadences, and signal characteristics across bands. Standard transformers apply shared parameters to all bands, potentially limiting their ability to model this rich structure. In this work, we introduce Astro-MoE, a foundational transformer architecture that enables dynamic processing via a Mixture of Experts module. We validate our model on both simulated (ELAsTiCC-1) and real-world datasets (Pan-STARRS1).

Context. Classifying variable stars is key to understanding stellar evolution and galactic dynamics. With the demands of large astronomical surveys, machine learning models, especially attention-based neural networks, have become the state of the art. While achieving high accuracy is crucial, improving model interpretability and uncertainty estimation is equally important to ensuring that insights are both reliable and comprehensible. Aims. We aim to enhance transformer-based models for classifying astronomical light curves by incorporating uncertainty estimation techniques to detect misclassified instances. We tested our methods on labeled datasets from MACHO, OGLE-III, and ATLAS, introducing a framework that significantly improves the reliability of automated classification for next-generation surveys. Methods. We used Astromer, a transformer-based encoder designed to capture representations of single-band light curves. We enhanced its capabilities by applying three methods for quantifying uncertainty: Monte Carlo dropout (MC Dropout), hierarchical stochastic attention, and a novel hybrid method that combines the two approaches (HA-MC Dropout). We compared these methods against a baseline of deep ensembles. To estimate uncertainty scores for the misclassification task, we used the following uncertainty estimates: the sampled maximum probability, probability variance (PV), and Bayesian active learning by disagreement. Results. In predictive performance tests, HA-MC Dropout outperforms the baseline, achieving macro F1-scores of 79.8 ± 0.5 on OGLE, 84 ± 1.3 on ATLAS, and 76.6 ± 1.8 on MACHO. When comparing the PV score values, the quality of uncertainty estimation by HA-MC Dropout surpasses that of all other methods, with improvements of 2.5 ± 2.3 for MACHO, 3.3 ± 2.1 for ATLAS, and 8.5 ± 1.6 for OGLE-III.

Context. ALeRCE (Automatic Learning for the Rapid Classification of Events) is currently processing the Zwicky Transient Facility (ZTF) alert stream, in preparation for the Vera C. Rubin Observatory, and classifying objects using a broad taxonomy. The ALeRCE light curve classifier is a balanced random forest (BRF) algorithm with a two-level scheme that uses variability features computed from the ZTF alert stream, and colors obtained from AllWISE and ZTF photometry. Aims. This work develops an updated version of the ALeRCE broker light curve classifier that includes tidal disruption events (TDEs) as a new subclass. For this purpose we incorporated 24 new features, notably including the distance to the nearest source detected in ZTF science images and a parametric model of the power-law decay for transients. We also expanded the labeled set to include 219 792 spectroscopically classified sources, including 60 TDEs. Methods. To effectively integrate TDEs into the ALeRCE’s taxonomy, we identified specific characteristics that set them apart from other transient classes, such as their central position in a galaxy, the typical decay pattern displayed when fully disrupted, and the lack of color variability after disruption. Based on these attributes, we developed features to distinguish TDEs from other transient events. Results. The modified classifier can distinguish between a broad range of classes with a better performance compared to the previous version and it can integate the TDE class achieving 91% recall, also identifying a large number of potential TDE candidates in ZTF alert stream unlabeled data.