IBM Research - Brazil

Abstract This paper exploits the exogenous variation of Mexico’s Oportunidades conditional cash transfer program on urban households’ time and consumption allocations to identify and structurally estimate a collective labor supply model with home production. I use my structural estimates to show that participation in Oportunidades increased maternal intrahousehold bargaining power by almost 13%, which is associated with an increase of approximately 14% in the production of a child-related public good in dual-earner beneficiary households. Counterfactual exercises show that Oportunidades is as effective as alternative cash transfer programs and wage subsidies at increasing mothers’ bargaining power, control over household monetary resources, and domestic output.

Biodiversity monitoring programs need to deliver accurate, timely, and actionable predictions. To establish a predictive monitoring program for deep-sea benthos of the Santos Basin, Brazil, we developed a two-stage structured model that allowed comparison of biodiversity predictions obtained from environmental simulations (2M-Sim). We also modeled the environmental variables as a function of spatial and temporal variables and compared this model's predictions with predictions obtained from real environmental data (2M). We built unstructured models (1M) as references to evaluate whether the proposed structured approach was reliable. We expected no significant differences between 1M and 2M or between 2M and 2M-Sim. Data were obtained from 100 stations at depths of 25-2400m during two surveys (2019 and 2021). In our model framework, we used 12 benthic macro- and meiofaunal variables, 44 sediment and water column environmental variables, and four spatial and temporal variables. We applied a random forest algorithm to the structured and unstructured models. All comparisons were performed with 20% of the dataset set aside for validation. The average accuracy was 72%, 69%, and 68% for the 1M, 2M, and 2M-Sim models, respectively. Accuracies of 2M ranged from 38% to 84% and were generally higher for macrofauna. The observed accuracy loss from 1M to 2M (3%) and from 2M to 2M-Sim (1%) was not significant for any biodiversity variable. The 2M model identified 30 significant environmental variables; bottom water parameters and sedimentary phytopigment and carbonate concentrations were the best predictors. Our approach supports biodiversity conservation by optimizing data needs and future sampling and by guiding data-driven management decisions for benthic biodiversity.

The use of foundation models has extended from natural language processing to molecular modeling. In this context, large-scale pre-training strategies have been applied to chemical language models to enable representation learning across diverse tasks. Here we introduce a family of encoder-decoder chemical foundation models pre-trained on a curated dataset of 91 million molecular sequences from PubChem. These models support a range of applications, including property estimation and reaction outcome prediction. We evaluate two model variants across several benchmark datasets and show that they match or exceed existing approaches. We also assess the structure of the learned representations and find that the embedding space supports few-shot learning and separates molecules based on chemically relevant features. This structure appears to result from the decoder-based reconstruction objective used during pre-training. These findings suggest that the proposed models can serve as general-purpose tools for molecular analysis and reasoning with minimal supervision.

With advancements in generative Artificial Intelligence (AI), there has been an increasing need for tools that rely on Large Language Model (LLM)s. As these models may produce undesired answers, there is a need to prevent such events, especially in enterprise environments. Even if models are trained on safe data, user inputs and even model behavior can be unpredictable, leading to problems such as leakage of confidential data that could result in revenue loss. In this paper, we describe our experiences on developing tools for “guardrailing” LLMs. We describe how we started with a quick monolith implementation, and later transitioned to a microservices architecture. As results, we share our lessons learned throughout the process, and how the re-architecture to microservices led to runtime performance gains, easier maintenance and extensibility, and also allowed us to open source the main component of the solution, so anyone can contribute to and use it.

Abstract The Covid-19 disease has spread widely over the whole world since the beginning of 2020. Following the epidemic which started in Wuhan, China on January 30, 2020 the World Health Organization (WHO) declared a global health emergency and a pandemic. Researchers of different disciplines work along with public health officials to understand the SARS-CoV-2 pathogenesis and jointly with the policymakers urgently develop strategies to control the spread of this new disease. Recent findings have observed specific image patterns from computed tomography (CT) for patients infected by SARS-CoV-2 which are distinct from the other pulmonary diseases. In this paper, we propose an explainable-by-design that has an integrated image segmentation mechanism based on SLIC that improves the algorithm performance and the interpretability of the resulting model. In order to evaluate the proposed approach, we used the SARS-CoV-2 CT scan dataset that we published recently and has been widely used in the literature. The proposed Super-xDNN could obtain statistically better results than traditional deep learning approaches as DenseNet-201 and Resnet-152. Furthermore, it also improved the explainability and interpretability of its decision mechanism when compared with the xDNN basis approach that uses the whole image as prototype. The segmentation mechanism of Super-xDNN favored a decision structure that is more close to the human logic. Moreover, it also allowed the provision of new insights as a heat-map which highlights the areas with highest similarities with Covid-19 prototypes, and an estimation of the area affected by the disease.

Carbon dioxide (CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$_2$$\\end{document}) trapping in capillary networks of reservoir rocks is a pathway to long-term geological storage. At pore scale, CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$_2$$\\end{document} drainage displacement depends on injection pressure, temperature, and the rock’s interaction with the surrounding fluids. Modeling this interaction requires adequate representations of both capillary volume and surface. For the lack of scalable representations, however, the prediction of a rock’s CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$_2$$\\end{document} storage potential has been challenging. Here, we report how to represent a rock’s pore space by statistically sampled capillary networks (ssCN) that preserve morphological rock characteristics. We have used the ssCN method to simulate CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$_2$$\\end{document} drainage within a representative sandstone sample at reservoir pressures and temperatures, exploring intermediate- and CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$_2$$\\end{document}-wet conditions. This wetting regime is often neglected, despite evidence of plausibility. By raising pressure and temperature we observe increasing CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$_2$$\\end{document} penetration within the capillary network. For contact angles approaching 90∘\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$^\\circ$$\\end{document}, the CO2\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$_2$$\\end{document} saturation exhibits a pronounced maximum reaching 80%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\%$$\\end{document} of the accessible pore volume. This is about twice as high as the saturation values reported previously. For enabling validation of our results and a broader application of our methodology, we have made available the rock tomography data, the digital rock computational workflows, and the ssCN models used in this study.