Renato Vicente

GENERAL APPRAISAL I think the premise of questioning the objectivity of ground truth is timely and resonates with contemporary discourse in the machine learning community, particularly within the subfields of data perspectivism, subjective NLP, and...

With urbanisation, cities face increasing temperatures, which are further increased by climate change. In this context, urban greenery can be a strategy to reduce surface temperatures in cities, providing cooling through shade and evapotranspiration. However, little is known about how long different types of urban greenery take to reach their maximum surface temperature reduction capacity in different climates across the world. To fill this gap, we previously developed a method using remote sensing data to quantify this time span, calling it “Cooling Establishment Time” (CET). To increase the number of case studies to those previously investigated in Zurich (Switzerland), our main challenges are to automate the identification of green areas, their selection, and quantification of their cooling dynamics through time in a computationally effective way. As a starting point, this ongoing research quantified the Cooling Establishment Time of green areas in São Paulo (Brazil), generating new information about this time measurement in a different climatic and urban context. São Paulo’s green areas took around 6 to 20 years to reach peak Land Surface Temperature reduction, which were longer than the time spans identified in Zurich. This contrast may be explained by the differences in local predominant vegetation and built environment. We expect to generate a dataset of green areas’ Cooling Establishment Times throughout different cities in the world, leading to a better understanding of what drives the temporal dynamics of vegetation cooling. Such results can be useful for policymakers to best plan green areas, improving heat mitigation and adaptation strategies depending on local environmental conditions and social needs.

Social structure may have changed from hierarchical to egalitarian and back along the evolutionary line of humans. Within the tradition of sociophysics, we construct a mathematical model of a society of agents subject to competing cognitive and social navigation constraints and predict, using statistical mechanics methods, that its degree of hierarchy decreases with encephalization and increases with group size, hence suggesting human societies were driven from hierarchical to egalitarian structures by the encephalization during the last few million years and back to hierarchical due to fast demographic changes during the Neolithic. In addition, applied to a different problem, the theory leads to the following predictions for modern pre-literary humans: (i) an intermediate hierarchy degree in mild climates. In harsher climates, societies will be (ii) more egalitarian if organized in small groups (of less than 100 persons) but (iii) more hierarchical if in larger (of more than 1000 persons) groups. The predicted bifurcation, characteristic of a phase transition, is also seen in the empirical cross-cultural record (248 cultures in the Ethnographic Atlas).

The objective of this study is to assess the performance of an innovative AI‐powered tool for sex determination using panoramic radiographs (PR) and to explore factors affecting the performance of the convolutional neural network (CNN). The study involved 207,946 panoramic dental X‐rays and their corresponding reports from 15 clinical centers in São Paulo, Brazil. The PRs were acquired with four different devices, and 58% of the patients were female. Data preprocessing included anonymizing the exams, extracting pertinent information from the reports, such as sex, age, type of dentition, and number of missing teeth, and organizing the data into a PostgreSQL database. Two neural network architectures, a standard CNN and a ResNet, were utilized for sex classification, with both undergoing hyperparameter tuning and cross‐validation to ensure optimal performance. The CNN model achieved 95.02% accuracy in sex estimation, with image resolution being a significant influencing factor. The ResNet model attained over 86% accuracy in subjects older than 6 years and over 96% in those over 16 years. The algorithm performed better on female images, and the area under the curve (AUC) exceeded 96% for most age groups, except the youngest. Accuracy values were also assessed for different dentition types (deciduous, mixed, and permanent) and missing teeth. This study demonstrates the effectiveness of an AI‐driven tool for sex determination using PR and emphasizes the role of image resolution, age, and sex in determining the algorithm's performance.

In supervised learning, training and test datasets are often sampled from distinct distributions. Domain adaptation techniques are thus required. Covariate shift adaptation yields good generalization performance when domains differ only by the marginal distribution of features. Covariate shift adaptation is usually implemented using importance weighting, which may fail, according to common wisdom, due to small effective sample sizes (ESS). Previous research argues this scenario is more common in high-dimensional settings. However, how effective sample size, dimensionality, and model performance/generalization are formally related in supervised learning, considering the context of covariate shift adaptation, is still somewhat obscure in the literature. Thus, a main challenge is presenting a unified theory connecting those points. Hence, in this paper, we focus on building a unified view connecting the ESS, data dimensionality, and generalization in the context of covariate shift adaptation. Moreover, we also demonstrate how dimensionality reduction or feature selection can increase the ESS, and argue that our results support dimensionality reduction before covariate shift adaptation as a good practice.

We dissect an experimental credit scoring model developed with real data and demonstrate - without access to protected attributes - how the use of location information introduces racial bias. We analyze the tree gradient boosting model with the aid of a game-theoretic inspired machine learning explainability technique, counterfactual experiments and Brazilian census data. By exposing algorithmic racial bias explaining the trained machine learning model inner mechanisms, this experiment comprises an interesting artifact to aid the endeavor of theoretical understanding of the emergence of racial bias in machine learning systems. Without access to individuals' racial categories, we show how classification parity measures using geographically defined groups could carry information about model racial bias. The experiment testifies to the need for methods and language that do not presuppose access to protected attributes when auditing ML models, the importance of considering regional specifics when addressing racial issues, and the central role of census data in the AI research community. To the best of our knowledge, this is the first documented case of algorithmic racial bias in ML-based credit scoring in Brazil, the country with the second largest Black population in the world.

In this work, we propose a Bayesian online reconstruction algorithm for sparse signals based on Compressed Sensing and inspired by L1-regularization schemes. A previous work has introduced a mean-field approximation for the Bayesian online algorithm and has shown that it is possible to saturate the offline performance in the presence of Gaussian measurement noise when the signal generating distribution is known. Here, we build on these results and show that reconstruction is possible even if prior knowledge about the generation of the signal is limited, by introduction of a Laplace prior and of an extra Kullback–Leibler divergence minimization step for hyper-parameter learning.

Empirical evidence suggests that social structure may have changed from hierarchical to egalitarian and back along the evolutionary line of humans. We model a society subject to competing cognitive and social navigation constraints. The theory predicts that the degree of hierarchy decreases with encephalization and increases with group size. Hence hominin groups may have been driven from a phase with hierarchical order to a phase with egalitarian structures by the encephalization during the last two million years, and back to hierarchical due to fast demographical changes during the Neolithic. The dynamics in the perceived social network shows evidence in the egalitarian phase of the observed phenomenon of Reverse Dominance. The theory also predicts for modern hunter-gatherers in mild climates a trend towards an intermediate hierarchy degree and a phase transition for harder ecological conditions. In harsher climates societies would tend to bemore egalitarian if organized in small groups but more hierarchical if in large groups. The theoretical model permits organizing the available data in the cross-cultural record (Ethnographic Atlas, N=248 cultures) where the symmetry breaking transition can be clearly seen.