arXiv subjects

Eduardo Sanchez

Publications and source records attributed to Eduardo Sanchez.

3 recordsLinked to original sources

arXiv2026

PolyFact: Comparing Consistency-Driven Post-training Methods for Cross-Lingual Factual Recall

Large language models (LLMs) trained predominantly on English data encode substantial world knowledge, yet often fail to express it reliably in other languages, a phenomenon known as cross-lingual factual inconsistency. To study this, we introduce PolyFact, a fully parallel multilingual factual QA dataset of 60K Wikidata-grounded facts across 12 typologically diverse languages, and propose consistency-driven GRPO with cross-lingual reward pooling. We compare our method against supervised fine-tuning (SFT) and the consistency-enhancement baselines DCO and CM-Align on OLMo-2-1124-7B and Qwen-2.5-7B, and analyze whether light continual pretraining (CPT) on parallel data provides a useful foundation for post-training. No single method dominates: SFT maximises in-distribution accuracy but not consistency, DCO yields the strongest consistency gains but fails to transfer to free-form generation, and our GRPO variant achieves the strongest transfer to free-form recall and unseen languages on the multilingual base model. CPT mildly aids monolingual models but harms multilingual ones. Mechanistic analyses suggest that GRPO is associated with reduced language specialization, consistent with greater sharing of representations across languages. We release our code, models, and dataset publicly.

cs.CL↗

arXiv2025

LDMX -- The Light Dark Matter eXperiment

The Light Dark Matter eXperiment (LDMX) is an electron fixed-target experiment optimized to search for sub-GeV dark matter production through the missing momentum signature. LDMX is designed to operate in End Station A at SLAC, using an 8 GeV electron beam accelerated alongside the LCLS-II drive beam. The design of the apparatus is strongly motivated by the performance requirements of a high-rate missing momentum search and leverages detector technologies and designs from other experiments along with existing facilities at SLAC. LDMX will improve on previous results by up to three orders of magnitude, enabling broad sensitivity to dark sector scenarios including the dark matter interaction strengths motivated by freeze-out of MeV-GeV mass dark matter to the observed relic abundance. With hermetic forward coverage, LDMX also has sensitivity to visible signatures of dark sectors and provides a unique probe of electron-nuclear interactions important to interpreting data from accelerator-based neutrino experiments. This report encompasses the technical design of the LDMX Detector, its simulated performance, and the physics capabilities of the experiment.

hep-ex↗

arXiv2019

Learning Disentangled Representations of Satellite Image Time Series

In this paper, we investigate how to learn a suitable representation of satellite image time series in an unsupervised manner by leveraging large amounts of unlabeled data. Additionally , we aim to disentangle the representation of time series into two representations: a shared representation that captures the common information between the images of a time series and an exclusive representation that contains the specific information of each image of the time series. To address these issues, we propose a model that combines a novel component called cross-domain autoencoders with the variational autoencoder (VAE) and generative ad-versarial network (GAN) methods. In order to learn disentangled representations of time series, our model learns the multimodal image-to-image translation task. We train our model using satellite image time series from the Sentinel-2 mission. Several experiments are carried out to evaluate the obtained representations. We show that these disentangled representations can be very useful to perform multiple tasks such as image classification, image retrieval, image segmentation and change detection.

cs.CV↗