LALuca de Alfaro

Luca de Alfaro

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Luca de Alfaro has a total of 58 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    A Generalized Approach to Exploring Subgroup Divergence

    by Eliana Pastor, Luca de Alfaro, Elena Baralis

    As the adoption of machine learning models and automated decision-making systems grows, understanding their behavior in different data subgroups is critical to ensuring their reliability and fairness. Current approaches predominantly focus on characterizing classification performance within subgroups or identifying subgroups where a model performs poorly. However, disparities in the subgroup behavior arise from various sources and encompass a wider range of models and tasks beyond classification performance. There is a need to explore and comprehend subgroup behavior in a broader context, including applications related to ranking, regression models, scoring functions, and studying specific data attributes. We introduce the notion of divergence as a general measure of subgroup anomaly in model or data behavior. Divergence quantifies the difference between a subgroup's performance statistics and the overall dataset statistics. The proposed approach explores all possible subgroups, using frequent pattern-mining techniques to analyze their divergence efficiently. The approach is target-agnostic and model-agnostic, making it suitable for various applications, including classification, regression, and ranking systems. Experimental results on real-world datasets demonstrate the effectiveness and efficiency of the proposed approach in uncovering critical subgroup behaviors.

  • Open Access

    EcoScape: Species-specific Functional Habitat Connectivity Via Efficient Parallel Propagation Simulations

    by Natalia Ocampo-Peñuela, Luca de Alfaro, Jasmine Tai, Natalie Valett, Coen Adler, Artie Nazarov, Montague Neate-Clegg

    Well-connected habitats sustain viable populations through gene flow, repopulation potential, and functional habitat extent. Conserving and enhancing habitat connectivity has been identified as a priority to halt extinctions and safeguard 30% of the planet. Conservation prioritization would therefore benefit from habitat connectivity modeling that is applicable at broad spatial and taxonomic scales. We introduce EcoScape, a species-specific computational tool capable of efficiently modeling habitat connectivity across large geographical areas. We describe how it estimates connectivity and flow relative to existing models. To evaluate its outputs, we map connectivity for three avian species across the contiguous United States, and we validate the resulting estimates against community-science occurrence data from eBird. EcoScape performs hundreds of GPU-parallelized dispersal simulations seeded randomly in habitat, with propagation informed by each species' dispersal ability and habitat preferences. Connectivity is the probability that a habitat pixel can be repopulated from other habitat locations; flow is a pixel's contribution to downstream repopulation, highlighting matrix elements that facilitate connectivity. We demonstrate EcoScape on a synthetic landscape, then map both connectivity and flow layers for Acorn Woodpecker, Pileated Woodpecker, and Steller's Jay from Area of Habitat maps, habitat preferences, and dispersal abilities derived from a global trait dataset. We then use eBird checklists across each species' range to test whether connectivity predicts local occurrence and abundance after controlling for observer effort and habitat patch size. In the synthetic landscape, we show how EcoScape predicts higher connectivity in larger and better-connected habitat patches, while flow between habitat is higher in a more permeable matrix. For the three focal species, we show that the computed connectivity positively predicts both local abundance and probability of occurrence in habitat grid cells beyond habitat patch size alone. We show that simulations can be efficiently performed in parallel on GPUs, leading to fast computation times that make it feasible to compute connectivity at continental scales. EcoScape can be applied to any species with sufficient input data, is computationally efficient, and can be validated with readily available data. These advantages enable habitat connectivity mapping at large spatial scales and fine resolutions for hundreds of species, offering guidance for current and future conservation efforts.

  • Open Access

    Variation in anthropogenic tolerance alters dispersal capacity of a large carnivore

    by K. Whitney Hansen, John J. Morgan, Luca De Alfaro, Christopher C. Wilmers, Natalia Ocampo-Peñuela

    Population connectivity is essential for long-term species persistence, and a major challenge to the conservation of isolated wildlife populations globally. Understanding how individual variation in movement behavior scales up to affect population connectivity across human-dominated landscapes is essential for the development and implementation of effective conservation and management plans, particularly for wide-ranging species with potential for human-wildlife conflict. Based on the observed movement behavior of 84 pumas ( Puma concolor ) in the Santa Cruz Mountains, we characterized individual variation in tolerance towards anthropogenic presence and used EcoScape, a novel functional habitat connectivity algorithm, to predict how individual differences in behavior affect the connectivity and dispersal capacity of an at-risk carnivore population. We found that individual differences in human tolerance have a greater effect than sex-based differences on modeled ability to disperse through anthropogenic landscapes, and strongly influence connectivity. This work highlights how individual variation in animal responses to human disturbance affects dispersal capacity, and offers a novel approach to predicting the persistence of metapopulations in the context of rapid global change.

  • Open Access

    Human development, societal stability and bird capital predict global tourist eBirding activity

    by Natalia Ocampo‐Peñuela, Sam de Alfaro, Montague H. C. Neate‐Clegg, Luca de Alfaro, Katarina Bjegovic, R. Scott Winton

    Birdwatching tourism is one of the fastest growing sectors within nature‐based tourism. This specialized form of ecotourism has the potential to bring economic and conservation benefits to tropical countries, which house the richest avifaunas, but little is known about the factors driving birdwatchers' decisions about which country to visit when they travel abroad. We assessed the growth of birdwatching tourism from 2010 to 2022 using data from the community‐science platform eBird. We then used a Bayesian hierarchical model to evaluate the factors that predict variation in eBirding tourism rates, including bird capital (total and small‐range species richness) and socio‐economic factors (Global Peace Index and Human Development Index) for 155 countries. Our results suggest that overall species richness and relative small‐range species richness are positive predictors of eBirding tourism activity, as are the Human Development Index and relative societal stability. Our findings highlight cases that illustrate the observed disparities among countries and make recommendations on how nations with unrealized potential can accrue economic and conservation benefits from birdwatching tourism. Read the free Plain Language Summary for this article on the Journal blog.


  • Open Access

    Detecting Interpretable Subgroup Drifts

    by Flavio Giobergia, Eliana Pastor, Luca de Alfaro, Elena Baralis


  • A Synthetic Benchmark to Explore Limitations of Localized Drift Detections

    by Flavio Giobergia, Eliana Pastor, Luca de Alfaro, Elena Baralis

  • Mitigating Subgroup Disparities in Speech Models: A Divergence-Aware Dual Strategy

    by Alkis Koudounas, Eliana Pastor, Luca de Alfaro, Elena Baralis

    Speech models may exhibit disparities in performance across different population subgroups. Prior mitigation efforts often rely on the manual user-driven selection of predefined data subgroups of interest. However, they fail to correctly identify all relevant subgroups associated with performance issues. We propose to mitigate performance disparities of subgroups that underperform, i.e., exhibit a divergence, relative to overall model performance. We tackle the performance disparities from two alternative perspectives - an in-processing one, implementing mitigation measures during model development, and a post-processing one, refining already trained models. For the in-processing scenario, we propose two approaches: a divergence-based regularization and a data augmentation technique to boost subgroup performance during model fine-tuning. The post-processing strategy introduces a divergence-aware data acquisition method to prioritize acquiring samples from underperforming subgroups. Experiments on a dataset for Automatic Speech Recognition, one for Emotion Recognition, and two datasets for Intent Classification in English and Italian highlight the improvement achieved by the divergence-aware strategies, which significantly reduce performance disparities and outperform traditional clustering-, KNN-, error-driven-, and random-based methods.


  • Open Access

    A Synthetic Benchmark to Explore Limitations of Localized Drift Detections

    by Flavio Giobergia, Eliana Pastor, Luca de Alfaro, Elena Baralis

    Concept drift is a common phenomenon in data streams where the statistical properties of the target variable change over time. Traditionally, drift is assumed to occur globally, affecting the entire dataset uniformly. However, this assumption does not always hold true in real-world scenarios where only specific subpopulations within the data may experience drift. This paper explores the concept of localized drift and evaluates the performance of several drift detection techniques in identifying such localized changes. We introduce a synthetic dataset based on the Agrawal generator, where drift is induced in a randomly chosen subgroup. Our experiments demonstrate that commonly adopted drift detection methods may fail to detect drift when it is confined to a small subpopulation. We propose and test various drift detection approaches to quantify their effectiveness in this localized drift scenario. We make the source code for the generation of the synthetic benchmark available at https://github.com/fgiobergia/subgroup-agrawal-drift.

  • Towards Comprehensive Subgroup Performance Analysis in Speech Models

    by Alkis Koudounas, Eliana Pastor, Giuseppe Attanasio, Vittorio Mazzia, Manuel Giollo, Thomas Gueudre, Elisa Reale, Luca Cagliero, Sandro Cumani, Luca de Alfaro, et al.

    The evaluation of spoken language understanding (SLU) systems is often restricted to assessing their global performance or examining predefined subgroups of interest. However, a more detailed analysis at the subgroup level has the potential to uncover valuable insights into how speech system performance differs across various subgroups. In this work, we identify biased data subgroups and describe them at the level of user demographics, recording conditions, and speech targets. We propose a new task-, model- and dataset-agnostic approach to detect significant intra- and cross-model performance gaps. We detect problematic data subgroups in SLU models by leveraging the notion of subgroup divergence. We also compare the outcome of different SLU models on the same dataset and task at the subgroup level. We identify significant gaps in subgroup performance between models different in size, architecture, or pre-training objectives, including multi-lingual and mono-lingual models, yet comparable to each other in overall performance. The results, obtained on two SLU models, four datasets, and three different tasks–intent classification, automatic speech recognition, and emotion recognition–confirm the effectiveness of the proposed approach in providing a nuanced SLU model assessment.


  • Open Access

    EcoScape: A Model for Species-Specific Functional Habitat Connectivity

    by Natalia Ocampo-Peñuela, Luca de Alfaro, Jasmine Tai, Natalie Valett, Coen Adler, Artie Nazarov

    Habitat connectivity is essential to maintain viable and healthy populations due to increased genetic flow, enhanced functional habitat, and repopulation potential [1–4]. Conserving, enhancing, and restoring habitat connectivity has been identified as a priority to achieve global targets of halting extinctions and safeguarding 30% of our planet [5]. Computational constraints of existing connectivity models have restricted these analyses to coarse scales and to a few charismatic species and/or landscapes [6–8]. We introduce EcoScape, an efficient algorithm for computing the pixel-level functional connectivity of species habitats. Informed by metapopulation theory, Ecoscape computes the connectivity of each habitat pixel as the probability that the pixel can be repopulated via propagation from other habitat locations, taking into account the species’ ecological preferences and dispersal abilities [9]. We provide a methodology based on open science data for applying EcoScape to any bird species. We show that for two forest/woodland bird species in California, EcoScape habitat connectivity values are highly correlated with relative abundances reported in eBird [10]. EcoScape leverages Graphic Processing Units (GPUs), and is hundreds of times faster than existing connectivity models. This opens the way to mapping habitat connectivity at large scale and fine resolutions for multiple species, offering guidance for current and future conservation efforts.