Orals

ORAL 1 (11:30-12:30)

• 11:30 - Víctor Ramos González (US) - Prototyping Logic-Based AI Services with LogicUS (AAAI23)

Abstract: Currently, there is renewed interest in logic-related solutions for AI and Computer Science. The availability of software tools to support the realization of such studies (both as powerful and versatile prototyping tools and as teaching tools) has become a necessity. Intending to contribute to this field, we present a tool that allows the unification of different logic tasks, focused on Computer Logic but adaptable to the treatment in several subfields, contexts, and abstraction levels (LogicUS-LIB, LogicUS-NB, LogicUS-GUI).

The tool provides a sound framework for two activity fields. On the one hand, in the topic of logic-based systems re- search, prototyping is facilitated in a relatively fast, simple, and highly adaptable way. On the other hand, in Education, by allowing the student abstract from low-level execution of algorithms whilst preserving the conceptual structures and procedural methodologies underlying the logical foundations.

• 11:38 - Nadine Kroher (Time Machine Capital 2) - DAACI-VoDAn: Improving Vocal Detection with New Data and Methods (EUSIPCO23)

Abstract: Currently, Vocal detection (VD) algorithms aim to detect the presence of vocals in music recordings and are an essential pre-processing step for other tasks, including singer identification and lyrics transcription. However, the lack of large-scale annotated datasets has slowed down research in the field, in particular w.r.t. the application of modern deep learning methods. This paper introduces DAACI-VoDAn, a novel dataset for VD that contains 706 full-length music tracks and vocal segment annotations. In addition, we propose a new method for the task that outperforms state of the art methods on DAACI-VoDAn as well as on an existing VD dataset. Our approach combines a convolutional head, that is pre-trained on large amounts of weakly-labeled data, with a temporal-convolutional architecture which models the occurrence of two-dimensional patterns over time.

• 11:46 - Alvaro Gonzalez Jimenez (University of Basel) - Robust T-Loss for Medical Image Segmentation (MICCAI23)

Abstract: This paper presents a new robust loss function, the T-Loss, for medical image segmentation. The proposed loss is based on the negative log-likelihood of the Student-t distribution and can effectively handle outliers in the data by controlling its sensitivity with a single parameter. This parameter is updated during the backpropagation process, eliminating the need for additional computation or prior information about the level and spread of noisy labels. Our experiments show that the T-Loss outperforms traditional loss functions in terms of dice scores on two public medical datasets for skin lesion and lung segmentation. We also demonstrate the ability of T-Loss to handle different types of simulated label noise, resembling human error. Our results provide strong evidence that the T-Loss is a promising alternative for medical image segmentation where high levels of noise or outliers in the dataset are a typical phenomenon in practice.

• 11:54 - María Teresa Martín Valdivia (UJA) - Automatic counter-narrative generation for hate speech in Spanish (SEPLN)

Abstract: This paper analyzes the use of language models to automatically generate counter-narratives for hate speech in Spanish. Despite the existence of a few studies in English and other languages, no previous work has explored this topic focused on Spanish. The article shows that the use of GPT-3 outperforms other models in generating non-offensive and informative counter-narratives, which sometimes present compelling arguments. We have used few-shot learning algorithms applying different prompt strategies and analyzing the results for each of them. Additionally, a new corpus called CONAN-SP, which consists of 238 pairs of hate speech and counter-narratives in Spanish, has been made available to the research community to facilitate further investigations in this area. These findings highlight the potential of language models to combat hate speech in Spanish by counter-narrative generation.

• 12:02 - Ignacio Peis (Technical University of Denmark) - Variational Mixture of HyperGenerators for Learning Distributions Over Functions (ICML23)

Abstract: Recent approaches build on implicit neural representations (INRs) to propose generative models over function spaces. However, they are computationally intensive when dealing with inference tasks, such as missing data imputation, or directly cannot tackle them. In this work, we propose a novel deep generative model, named VAMoH. VAMoH combines the capabilities of modeling continuous functions using INRs and the inference capabilities of Variational Autoencoders (VAEs). In addition, VAMoH relies on a normalizing flow to define the prior, and a mixture of hypernetworks to parametrize the data log-likelihood. This gives VAMoH a high expressive capability and interpretability. Through experiments on a diverse range of data types, such as images, voxels, and climate data, we show that VAMoH can effectively learn rich distributions over continuous functions. Furthermore, it can perform inference-related tasks, such as conditional super-resolution generation and in-painting, as well or better than previous approaches, while being less computationally demanding.

• 12:10 - 12:30 - Q&A

ORAL 2 (12:45-13:45)

• 12:45 - Cristina Zuheros Montes (UGR) - Deep learning models for processing social networks opinions for crowd decisions (Information Fusion)

Abstract: There exist a high demand to provide explainability to artificial intelligence systems, where decision making models are included. This paper focuses on crowd decision making using natural language evaluations from social media with the aim to provide explainability. We present the Explainable Crowd Decision Making based on Subgroup Discovery and Attention Mechanisms (ECDM-SDAM) methodology as an a posteriori explainable process that captures the wisdom of crowds that is naturally provided in social media opinions. It extracts the opinions from social media texts using a deep learning based sentiment analysis approach called Attention based Sentiment Analysis Method. The methodology includes a backward process that provides explanations to justify its sense-making procedure by applying mainly the attention mechanism on texts and subgroup discovery on opinions. We evaluate the methodology in the real case study of the TripR-2020Large dataset for restaurant choice. The results show that the ECDM-SDAM methodology provides easy understandable explanations that elucidates the key reasons that support the output of the decision process.

• 12:53 - Ricardo Garcia Pinel (INRIA) - PolarNet: 3D Point Clouds for Language-Guided Robotic Manipulation (CoRL)

Abstract: The ability for robots to comprehend and execute manipulation tasks based on natural language instructions is a long-term goal in robotics. The dominant approaches for language-guided manipulation use 2D image representations, which face difficulties in combining multi-view cameras and inferring precise 3D positions and relationships. To address these limitations, we propose a 3D point cloud based policy called PolarNet for language-guided manipulation. It leverages carefully designed point cloud inputs, efficient point cloud encoders, and multi-modal transformers to learn 3D point cloud representations and integrate them with language instructions for action prediction. PolarNet is shown to be effective and data efficient in a variety of experiments conducted on the RLBench benchmark. It outperforms state-of-the-art 2D and 3D approaches in both single-task and multi-task learning. It also achieves promising results on a real robot.

• 13:01 - José Raúl Ruiz Sarmiento (UMA) - The Robot@Home2 dataset: A new release with improved usability tools (SoftwareX)

Abstract: The development of intelligent robots relies heavily on data-intensive Artificial Intelligence (AI) techniques, demanding large and diverse datasets. To this end, and with the focus on indoor robots, the MAPIR-UMA group contributed the Robot@Home dataset, a repository of +87K RGB-D images and 2D laser scans collected by a mobile robot in different apartment settings. Additionally, the dataset is enriched with 2D geometric maps and 3D reconstructions of these environments, being all the data thoroughly annotated with ground truth information about the categories of objects and rooms. The dataset has been used in the design and validation of AI techniques for a variety of tasks, such as object/room recognition, semantic segmentation, semantic mapping, odometry, path planning, navigation, localization, or SLAM. Recently, its second version was released, bringing a number of innovations to improve usability. These include a recasting of the data in the form of a relational database, the development of a python toolkit for ease of use, and the creation of a suite of learning Jupyter Notebooks for reducing the learning curve to the minimum. With this iteration, publicly available at https://github.com/goyoambrosio/RobotAtHome2, we aim to further accelerate and advance the research on exciting AI techniques.

• 13:09 - Miguel Ángel Armengol de la Hoz (Junta de Andalucía) - A guide to sharing open healthcare data under the General Data Protection Regulation (Nature Scientific Data)

Abstract: Sharing healthcare data is increasingly essential for developing data-driven improvements in patient care at the Intensive Care Unit (ICU). However, it is also very challenging under the strict privacy legislation of the European Union (EU). Therefore, we explored four successful open ICU healthcare databases to determine how open healthcare data can be shared appropriately in the EU. A questionnaire was constructed based on the Delphi method. Then, follow-up questions were discussed with experts from the four databases. These experts encountered similar challenges and regarded ethical and legal aspects to be the most challenging. Based on the approaches of the databases, expert opinion, and literature research, we outline four distinct approaches to openly sharing healthcare data, each with varying implications regarding data security, ease of use, sustainability, and implementability. Ultimately, we formulate seven recommendations for sharing open healthcare data to guide future initiatives in sharing open healthcare data to improve patient care and advance healthcare.

• 13:17 - Juan José Murillo-Fuentes (US) - Cross-point detection with deep learning for thread counting in canvases by Velázquez and Murillo (Engineering Applications of AI)

Abstract: In forensic studies in art, the analysis of thread countings in plain weave canvases plays a central role. The comparison between the thread densities along the canvases allows the curators to conclude about the authorship or integrity of a masterpiece. Fourier transform (FT) has been used as an unsupervised robust tool in the last decade. However, in some fabrics, detected in paintings by Velázquez and Murillo, the frequency analysis fails. In this work, we present an alternative solution based on deep learning. We aim at locating crossing points. Densities can be then easily estimated. We checked for various designs based on inception and the U-Net model. The normalized error was reduced from 7.47% with FT to 1.61% using DL. The method is applied in the comparison of canvases at the Museo Nacional del Prado.

• 13:25 - 13:45 - Q&A

ORAL 3 (15:45-16:45)

• 15:45 - Marina Escobar Planas (UPV / EC JRC) - “That Robot Played with Us!” Children’s Perceptions of a Robot after a Child-Robot Group Interaction (ACM Human-Computer Interaction)

Abstract: The design of child-centred, intelligent and collaborative robots is a challenging endeavour, which requires to understand how the implemented robot behaviours and collaboration paradigms affect children’s perception about the robot. This paper presents the results of a set of semi-structured interviews of N=81, 5 to 8 years old children who previously interacted in pairs with a robot in the context of a problem-solving task. We manipulated two different factors of the robot behaviour: cognitive reliability in logic game movements (optimal vs sub-optimal) and expressivity in the communication (expressive vs neutral) and we assigned the children in one of the four conditions. At post-intervention interviews, we examined children’s perceptions on the robot’s attributions, collaboration and social role. Results indicate that a robot’s cognitive reliability shapes the helping relationship between the children and the robot, while the robot’s expressivity impacts children perception of the robot supportive ability and friendship. Finally, results also indicate that, even if children interact in pairs with the robot, their perceptions about it remain individual, although a good collective task-performance seems to empower children perception of the robot in terms of friendship and reliability.

• 15:53 - Ruben Vera-Rodriguez (UAM) - Artificial Intelligence for Biometric Recognition (ICCV23)

Abstract: In the contemporary landscape of technology, the convergence of Artificial Intelligence (AI) and biometric recognition has introduced a new era of innovation and security. Biometric recognition, which encompasses the identification and authentication of individuals through unique physical and behavioral traits, stands as a critical component in diverse domains, from access control and surveillance to healthcare and finance. With the rapid advancement of AI, machine learning, and deep learning techniques, this synergy presents a powerful and transformative toolset for enhancing the accuracy, efficiency, and security of biometric systems. In the proposed talk I will cover recent advances made in my research group in the field of face recognition through the usage of synthetically generated face images through state of the art GAN and Diffusion models covering our recent two papers at IJCB and ICCV and our Workshop organized on this matter at WACV 2024. Also, I will cover the topic of behavioral biometrics (online signature verification, keystroke dynamics, gait recognition and mobile human computer interaction) using the latest deep learning algorithms (LSTM, GRU and Transformers).

• 16:01 - Alejandro Moreo Fernández (Instituto di Scienza e Tecnologie dell'Informazione) - Learning to Quantify (JAIR / TKDD)

Abstract: Quantification (or “supervised prevalence estimation“), is the task of estimating the percentages of items of a test sample that belong to each class. Quantification is interesting in all applications of classification in which the final goal is not determining which class (or classes) individual data items belong to, but estimating the distribution of the items across the classes. Examples of the latter include the social sciences, political science, market research, epidemiology, etc. The literature has convincingly shown that solving this task by classifying and counting yields poor accuracy in situations characterized by prior probability shift. In this presentation, I will offer a comprehensive overview of the problem, the methods that have been proposed so far, and the potential applications of quantification, by also covering research results we have published during 2023 in top-tier conferences like ECML-PKDD, as well as in international journals like JAIR and TKDD.

• 16:09 - José Manuel Camacho Rodríguez (ICMAT CSIC) - Comparing Machine Learning Methods for Bot Detection on Twitter with a Case Study on COVID-19 Discourse and the Implementation of a Bot Content Analysis Tool Modeling of Inter- and Intra-observer Variability in Medical Image Segmentation (Social Science Computer Review)

Abstract: Bot Detection is crucial in a world where Online Social Networks (OSNs) play a pivotal role in our lives as public communication channels. This task becomes highly relevant in crises like the Covid-19 pandemic when there is a growing risk of proliferation of automated accounts designed to produce misinformation content. To address this issue, we first introduce a comparison between supervised Bot Detection models using Data Selection. The techniques used to develop the bot detection models use features such as the tweets’ metadata or accounts’ Digital Fingerprint. The techniques implemented in this work proved effective in detecting bots with different behaviors. Social Fingerprint-based methods have been found to be effective with bots that behave in a coordinated manner. Furthermore, all these approaches have produced excellent results compared to the Botometer v3. Second, we present and discuss a case study related to the Covid-19 pandemic that analyses the differences in the discourse between bots and humans on Twitter, a platform used worldwide to express opinions and engage in dialogue in a public arena. While bots and humans generally express themselves alike, the tweets’ content and sentiment analysis reveal some dissimilitudes, especially in tweets concerning President Trump. When the discourse switches to pandemic management by Trump, sentiment-related values display a drastic difference, showing that tweets generated by bots have a predominantly negative attitude. However, according to our findings, while automated accounts are numerous and active in discussing controversial issues, they usually do not seem to increase exposure to negative and inflammatory content for human users.  

• 16:17 - Rocío Gonzalez-Diaz (US) - Topology-based representative datasets to reduce neural network training resources (Neural Computing and Applications)

Abstract: One of the main drawbacks of the practical use of neural networks is the long time required in the training process. Such a training process consists of an iterative change of parameters trying to minimize a loss function. These changes are driven by a dataset, which can be seen as a set of labeled points in an n-dimensional space. In this paper, we explore the concept of a representative dataset which is a dataset smaller than the original one, satisfying a nearness condition independent of isometric transformations. Representativeness is measured using persistence diagrams (a computational topology tool) due to its computational efficiency. We theoretically prove that the accuracy of a perceptron evaluated on the original dataset coincides with the accuracy of the neural network evaluated on the representative dataset when the neural network architecture is a perceptron, the loss function is the mean squared error, and certain conditions on the representativeness of the dataset are imposed. These theoretical results accompanied by experimentation open a door to reducing the size of the dataset to gain time in the training process of any neural network.

• 16:25 - 16:45 - Q&A