PUBLICACIONES

Abstract

The presentation turns around the subject of explainable AI. More specifically, we deal with attribution numerical scores that are assigned to features values of an entity under classification, to identify and rank their importance for the obtained classification label. We concentrate on the popular SHAP score [2] that can be applied with black-box and open models. We show that, in contrast to its general #P-hardness, it can be computed in polynomial time for classifiers that are based on decomposable and deterministic Boolean decision circuits. This class of classifiers includes decision trees and ordered binary decision diagrams. This result was established in [1]. The presentation illustrates how the proof heavily relies on the connection to SAT-related computational problems.

ABSTRACT

Word embeddings are vital descriptors of words in unigram representations of documents for many tasks in natural language processing and information retrieval. The representation of queries has been one of the most critical challenges in this area because it consists of a few terms and has little descriptive capacity. Strategies such as average word embeddings can enrich the queries' descriptive capacity since they favor the identification of related terms from the continuous vector representations that characterize these approaches. We propose a data-driven strategy to combine word embeddings. We use Idf combinations of embeddings to represent queries, showing that these representations outperform the average word embeddings recently proposed in the literature. Experimental results on benchmark data show that our proposal performs well, suggesting that data-driven combinations of word embeddings are a promising line of research in ad-hoc information retrieval.

ABSTRACT

Medical images are an essential input for the timely diagnosis of pathologies. Despite its wide use in the area, searching for images that can reveal valuable information to support decision-making is difficult and expensive. However, the possibilities that open when making large repositories of images available for search by content are unsuspected. We designed a content-based image retrieval system for medical imaging, which reduces the gap between access to information and the availability of useful repositories to meet these needs. The system operates on the principle of query-by-example, in which users provide medical images, and the system displays a set of related images. Unlike metadata match-driven searches, our system drives content-based search. This allows the system to conduct searches on repositories of medical images that do not necessarily have complete and curated metadata. We explore our system’s feasibility in computational tomography (CT) slices for SARS-CoV-2 infection (COVID-19), showing that our proposal obtains promising results, advantageously comparing it with other search methods.

ABSTRACT

This paper proposes a novel online self-learning detection system for different types of objects. It allows users to random select detection target, generating an initial detection model by selecting a small piece of image sample and continue training the detection model automatically. The proposed framework is divided into two parts: First, the initial detection model and the online reinforcement learning. The detection model is based on the proportion of users of the Haar-like features to generate feature pool, which is used to train classifiers and get positive-negative (PN) classifier model. Second, as the videos plays, the detecting model detects the new sample by Nearest Neighbor (NN) Classifier to get the PN similarity for new model. Online reinforcement learning is used to continuously update classifier, PN model and new classifier. The experiment shows the result of less detection sample with automatic online reinforcement learning is satisfactory.

ABSTRACT

Every year physicians face an increasing demand of image-based diagnosis from patients, a problem that can be addressed with recent artificial intelligence methods. In this context, we survey works in the area of automatic report generation from medical images, with emphasis on methods using deep neural networks, with respect to: (1) Datasets, (2) Architecture Design, (3) Explainability and (4) Evaluation Metrics. Our survey identifies interesting developments, but also remaining challenges. Among them, the current evaluation of generated reports is especially weak, since it mostly relies on traditional Natural Language Processing (NLP) metrics, which do not accurately capture medical correctness.

ABSTRACT

In this paper the Single Particle Model is used to describe the behavior of a Li-ion battery. The main goal is to design a feedback input current in order to regulate the State of Charge (SOC) to a prescribed reference trajectory. In order to do that, we use the boundary ion concentration as output. First, we measure it directly and then we assume the existence of an appropriate estimator, which has been established in the literature using voltage measurements. By applying backstepping and Lyapunov tools, we are able to build observers and to design output feedback controllers giving a positive answer to the SOC tracking problem. We provide convergence proofs and perform some numerical simulations to illustrate our theoretical results.

ABSTRACT

Decision support systems have become increasingly popular in the domain of agriculture. With the development of automated machine learning, agricultural experts are now able to train, evaluate and make predictions using cutting edge machine learning (ML) models without the need for much ML knowledge. Although this automated approach has led to successful results in many scenarios, in certain cases (e.g., when few labeled datasets are available) choosing among different models with similar performance metrics is a difficult task. Furthermore, these systems do not commonly allow users to incorporate their domain knowledge that could facilitate the task of model selection, and to gain insight into the prediction system for eventual decision making. To address these issues, in this paper we present AHMoSe, a visual support system that allows domain experts to better understand, diagnose and compare different regression models, primarily by enriching model-agnostic explanations with domain knowledge. To validate AHMoSe, we describe a use case scenario in the viticulture domain, grape quality prediction, where the system enables users to diagnose and select prediction models that perform better. We also discuss feedback concerning the design of the tool from both ML and viticulture experts.

ABSTRACT

Variations in human behavior correspond to the adaptation of the nervous system to different internal and environmental demands. Attention, a cognitive process for weighing environmental demands, changes over time. Pupillary activity, which is affected by fluctuating levels of cognitive processing, appears to identify neural dynamics that relate to different states of attention. In mice, for example, pupil dynamics directly correlate with brain state fluctuations. Although, in humans, alpha-band activity is associated with inhibitory processes in cortical networks during visual processing, and its amplitude is modulated by attention, conclusive evidence linking this narrowband activity to pupil changes in time remains sparse. We hypothesize that, as alpha activity and pupil diameter indicate attentional variations over time, these two measures should be comodulated. In this work, we recorded the electroencephalographic (EEG) and pupillary activity of 16 human subjects who had their eyes fixed on a gray screen for 1 min. Our study revealed that the alpha-band amplitude and the high-frequency component of the pupil diameter covariate spontaneously. Specifically, the maximum alpha-band amplitude was observed to occur ∼300 ms before the peak of the pupil diameter. In contrast, the minimum alpha-band amplitude was noted to occur ∼350 ms before the trough of the pupil diameter. The consistent temporal coincidence of these two measurements strongly suggests that the subject’s state of attention, as indicated by the EEG alpha amplitude, is changing moment to moment and can be monitored by measuring EEG together with the diameter pupil.

ABSTRACT

In the automotive industry, light-alloy aluminum castings are an important element for determining roadworthiness. X-ray testing with computer vision is used during automated inspections of aluminum castings to identify defects inside of the test object that are not visible to the naked eye. In this article, we evaluate eight state-of-the-art deep object detection methods (based on YOLO, RetinaNet, and EfficientDet) that are used to detect aluminum casting defects. We propose a training strategy that uses a low number of defect-free X-ray images of castings with superimposition of simulated defects (avoiding manual annotations). The proposed solution is simple, effective, and fast. In our experiments, the YOLOv5s object detector was trained in just 2.5 h, and the performance achieved on the testing dataset (with only real defects) was very high (average precision was 0.90 and the F1 factor was 0.91). This method can process 90 X-ray images per second, i.e. ,this solution can be used to help human operators conduct real-time inspections. The code and datasets used in this paper have been uploaded to a public repository for future studies. It is clear that deep learning-based methods will be used more by the aluminum castings industry in the coming years due to their high level of effectiveness. This paper offers an academic contribution to such efforts.

ABSTRACT

We describe how answer-set programs can be used to declaratively specify counterfactual interventions on entities under classification, and reason about them. In particular, they can be used to define and compute responsibility scores as attribution-based explanations for outcomes from classification models. The approach allows for the inclusion of domain knowledge and supports query answering. A detailed example with a naive-Bayes classifier is presented.