Javier Marin

Pau Gallés, Katalin Takats, Javier Marin

A lot of work has been done to reach the best possible performance of predictive models on images. There are fewer studies about the resilience of these models when they are trained on image datasets that suffer modifications altering their original quality. Yet this is a common problem that is often encountered in the industry. A good example of that is with earth observation satellites that are capturing many images. The energy and time of connection to the earth of an orbiting satellite are limited and must be carefully used. An approach to mitigate that is to compress the images on board before downloading. The compression can be regulated depending on the intended usage of the image and the requirements of this application. We present a new software tool with the name iquaflow that is designed to study image quality and model performance variation given an alteration of the image dataset. Furthermore, we do a showcase study about oriented object detection models adoption on a public image dataset DOTA Xia_2018_CVPR given different compression levels. The optimal compression point is found and the usefulness of iquaflow becomes evident.

Javier Marin

This work proposes a method to evaluate synthetic tabular data generated to augment small sample datasets. While data augmentation techniques can increase sample counts for machine learning applications, traditional validation approaches fail when applied to extremely limited sample sizes. Our experiments across four datasets reveal significant inconsistencies between global metrics and topological measures, with statistical tests producing unreliable significance values due to insufficient sample sizes. We demonstrate that common metrics like propensity scoring and MMD often suggest similarity where fundamental topological differences exist. Our proposed normalized Bottleneck distance based metric provides complementary insights but suffers from high variability across experimental runs and occasional values exceeding theoretical bounds, showing inherent instability in topological approaches for very small datasets. These findings highlight the critical need for multi-faceted evaluation methodologies when validating synthetic data generated from limited samples, as no single metric reliably captures both distributional and structural similarity.

Javier Marin

This research addresses two fundamental challenges in Marketing Mix Modeling: the tendency of models to over-attribute influence to high-investment channels and the difficulty in quantifying cross-channel effects. We propose integrating the Michaelis-Menten equation and Maxwell-Boltzmann kinetic theory into hierarchical Bayesian models to overcome these limitations. Our approach uses the Michaelis-Menten model to characterize shape effects with spending-independent parameters and Boltzmann-type equations to systematically quantify cross-channel dynamics. Experimental results show that this physics-inspired approach maintains predictive accuracy while providing superior analytical insights into channel effectiveness and interactions. The normalized Michaelis-Menten constant offers an investment-independent measure of channel efficacy, while the N-particle system simulation reveals previously ignored channel interdependencies, enabling more accurate attribution and informed resource allocation decisions.

Diego Velazquez, Pau Rodriguez López, Sergio Alonso et al.

This paper presents EarthView, a comprehensive dataset specifically designed for self-supervision on remote sensing data, intended to enhance deep learning applications on Earth monitoring tasks. The dataset spans 15 tera pixels of global remote-sensing data, combining imagery from a diverse range of sources, including NEON, Sentinel, and a novel release of 1m spatial resolution data from Satellogic. Our dataset provides a wide spectrum of image data with varying resolutions, harnessed from different sensors and organized coherently into an accessible HuggingFace dataset in parquet format. This data spans five years, from 2017 to 2022. Accompanying the dataset, we introduce EarthMAE, a tailored Masked Autoencoder, developed to tackle the distinct challenges of remote sensing data. Trained in a self-supervised fashion, EarthMAE effectively processes different data modalities such as hyperspectral, multispectral, topographical data, segmentation maps, and temporal structure. This model helps us show that pre-training on Satellogic data improves performance on downstream tasks. While there is still a gap to fill in MAE for heterogeneous data, we regard this innovative combination of an expansive, diverse dataset and a versatile model adapted for self-supervised learning as a stride forward in deep learning for Earth monitoring.

Javier Marin

Understanding how consumers respond to business advertising efforts is essential for optimizing marketing investment. This research introduces a new modeling approach based on the concepts of symmetries and scaling laws in physics to describe consumer response to advertising dynamics. Drawing from mathematical frameworks used in physics and social sciences, we propose a model that accounts for a key aspect: the saturation effect. The model is validated against commonly used models, including the Michaelis-Menten and Hill equations, showing its ability to better capture nonlinearities in advertising effects. We introduce new key parameters like Marketing Sensitivity, Response Sensitivity, and Behavioral Sensitivit, that offer additional insights into the drivers of audience engagement and advertising performance. Our model provides a rigorous yet practical tool for understanding audience behavior, contributing to the improvement of budget allocation strategies.

Javier Marin, Aritro Biswas, Ferda Ofli et al.

In this paper, we introduce Recipe1M+, a new large-scale, structured corpus of over one million cooking recipes and 13 million food images. As the largest publicly available collection of recipe data, Recipe1M+ affords the ability to train high-capacity modelson aligned, multimodal data. Using these data, we train a neural network to learn a joint embedding of recipes and images that yields impressive results on an image-recipe retrieval task. Moreover, we demonstrate that regularization via the addition of a high-level classification objective both improves retrieval performance to rival that of humans and enables semantic vector arithmetic. We postulate that these embeddings will provide a basis for further exploration of the Recipe1M+ dataset and food and cooking in general. Code, data and models are publicly available.

Enes Kocabey, Mustafa Camurcu, Ferda Ofli et al.

A person's weight status can have profound implications on their life, ranging from mental health, to longevity, to financial income. At the societal level, "fat shaming" and other forms of "sizeism" are a growing concern, while increasing obesity rates are linked to ever raising healthcare costs. For these reasons, researchers from a variety of backgrounds are interested in studying obesity from all angles. To obtain data, traditionally, a person would have to accurately self-report their body-mass index (BMI) or would have to see a doctor to have it measured. In this paper, we show how computer vision can be used to infer a person's BMI from social media images. We hope that our tool, which we release, helps to advance the study of social aspects related to body weight.