Iago C. Chaves

Mecanismos de seleção diferencialmente privados (DP) obtem o elemento de maior qualidade em um conjunto finito, garantindo forte privacidade. No entanto, a maioria dos métodos existentes usa sensibilidade global, frequentemente adicionando ruído excessivo e prejudicando a utilidade dos resultados. Propomos o algoritmo Smooth Noisy Max (SNM), que emprega sensibilidade suave para obter limites de erro mais precisos e maior acurácia sob condições moderadas. Experimentos mostram que o SNM supera os melhores métodos de seleção com DP em três aplicações: seleção de percentis, árvores de decisão gananciosas e florestas aleatórias.

Recommendation systems personalize user experiences but often introduce the “rich-get-richer” effect, where popular items dominate visibility. This bias disadvantages new items or those with few interactions, where historical data limits the exposure of new items. To address these challenges, fairness in exposure has been studied through balanced exposure, which ensures equal visibility but may reduce recommendation utility, and quality-weighted exposure, which prioritizes high-quality items. To mitigate the trade-off between fairness and utility, we propose TWIX, a method that applies quality-weighted and balanced exposures determined by a threshold. Our approach reduces popularity bias and cold start issues while maintaining recommendation utility.

Computer manufacturers offer platforms for users to describe device faults using textual reports such as “My screen is flickering”. Identifying the faulty component from the report is essential for automating tests and improving user experience. However, such reports are often ambiguous and lack detail, making this task challenging. Large Language Models (LLMs) have shown promise in addressing such issues. This study evaluates 27 open-source models (1B–72B parameters) and 2 proprietary LLMs using four prompting strategies: Zero-Shot, Few-Shot, Chain-of-Thought (CoT), and CoT+Few-Shot (CoT+FS). We conducted 98,948 inferences, processing over 51 million input tokens and generating 13 million output tokens. We achieve f1-score up to 0.76. Results show that three models offer the best balance between size and performance: mistral-small-24b-instruct and two smaller models, llama-3.2-1b-instruct and gemma-2-2b-it, that offer competitive performance with lower VRAM usage, enabling efficient inference on end-user devices as modern laptops or smartphones with NPUs.

Differentially private selection mechanisms offer strong privacy guarantees for queries aiming to identify the top-scoring element r from a finite set R, based on a dataset-dependent utility function. While selection queries are fundamental in data science, few mechanisms effectively ensure their privacy. Furthermore, most approaches rely on global sensitivity to achieve differential privacy (DP), which can introduce excessive noise and impair downstream inferences. To address this limitation, we propose the Smooth Noisy Max (SNM) mechanism, which leverages smooth sensitivity to yield provably tighter (upper bounds on) expected errors compared to global sensitivity-based methods. Empirical results demonstrate that SNM is more accurate than state-of-the-art differentially private selection methods in three applications: percentile selection, greedy decision trees, and random forests.

Computer manufacturers typically offer platforms for users to report faults. However, there remains a significant gap in these platforms' ability to effectively utilize textual reports, which impedes users from describing their issues in their own words. In this context, Natural Language Processing (NLP) offers a promising solution, by enabling the analysis of user-generated text. This paper presents an innovative approach that employs NLP models to classify user reports for detecting faulty computer components, such as CPU, memory, motherboard, video card, and more. In this work, we build a dataset of 341 user reports obtained from many sources. Additionally, through extensive experimental evaluation, our approach achieved an accuracy of 79% with our dataset.

Este artigo propõe uma nova abordagem de particionamento de dados categóricos para aplicar a privacidade diferencial em Gradient Boosting Decision Trees. Nele estudamos aprimoramentos no tratamento de atributos categóricos e seleção aleatória de pontos de particionamento enquanto oferecemos garantias de privacidade diferencial. Nossa abordagem define uma nova função de ganho para esses atributos e determina os limites de sensibilidade dessa função. Além disso, realizamos uma análise empírica em 6 conjuntos de dados reais, mostrando que a abordagem proposta alcança taxas de erro menores ou iguais aos modelos de referência.

Advancements in electronic fabrication technologies have facilitated the large-scale production of computer components, which are prone to faults over time. Despite the availability of fault-reporting tools provided by hardware manufacturers, there is a significant gap in effectively utilizing textual reports due to data scarcity. In this paper, we introduce FACTO dataset, a comprehensive collection of user reports on faulty computer components such as video cards, storage devices, motherboards, memory, and others. Data was gathered through a survey of hardware specialists, web scraping of internet forums, and synthetic text generation from real manufacturer data using large language models. This dataset aims to provide insights for correlating user reports with faulty components, thus enhancing diagnostic capabilities and improving hardware reliability and customer satisfaction.

Predicting failures in Hard Disk Drives (HDD) is a major challenge that has been faced by both industry and academy in recent years. Being able to predict failure events may incur in avoiding data losses and also improve service availability. Among all failure prediction strategies, the health degree prediction is one of the most popular. The task of health degree prediction consists of, given a finite set of health states that are related to the degradation of the equipment, estimate which state reflects the actual degradation of the equipment. This problem is usually modeled as a classification task. Although many health degree prediction methods have been proposed, some practical details regarding this prediction task have been neglected in previous works. In this work we tackle two of these aspects: the ordinal nature of the problem and the different costs associated with miss-classifications. Such aspects are considered in a framework based on Deep Recurrent Neural Networks (DRNN). The choice of DRNN is given its remarkable performances in many applications including HDDs failure prediction. The resulting methods outperformed state-of-the-art approaches in a metric that consider the new aspects that motivated our proposal.