Translating Embeddings (TransE) is a widely adopted model for knowledge graph completion which represents relationships as vector translations in an embedding space. While it is known for its efficiency and simplicity, TransE struggles with capturing complex relational patterns, particularly in one-to-many (1-to-N), many-to-one (N-to-1), and many-to-many (N-to-N) relationships. This challenge arises from its rigid distance-based formulation. In this paper, we propose an enhancement to TransE that integrates fuzzy type constraints, which provide a soft regularization of entity embeddings based on their degree of membership in semantic categories (e.g., city, person). This extension aims to improve the model’s ability to represent intricate relationships and enhance the overall performance in knowledge graph tasks.

Mitigating biases in neural networks is crucial to reduce or eliminate the predictive model’s unfair responses, which may arise from unbalanced training, defective architectures, or even social prejudices embedded in the data. This study proposes a novel and fully differentiable framework for mitigating neural network bias using Saliency Maps and Fuzzy Logic. We focus our analysis on a simulation study for recommendation systems, where neural networks are crucial in classifying job applicants based on relevant and sensitive attributes. Leveraging the interpretability of a set of Fuzzy implications and the importance of features attributed by Saliency Maps, our approach penalizes models when they overly rely on biased predictions during training. In this way, we ensure that bias mitigation occurs within the gradient-based optimization process, allowing efficient model training and evaluation.

Smart agriculture has seen impressive progresses in monitoring the quality of the crop and early detecting the onset of pathogens.However, this is typically achieved through smart, expensive, and energy-demanding robots and autonomous systems.We propose an AI-empowered portable low-cost shortwave near-infrared spectroscopy (sw-NIRS) solution that allows non-destructive measurements from plants and vegetables.In this pilot study, we specifically targeted an orange fruit and showed that it is possible to classify its different parts through sw-NIRS in the range 1350-2150 nm by using AI models, exceeding 97% accuracy.Also, we explored the minimum amount of energy needed to reach such high classification performance.In the future, we aim to extend this investigation to other targets (e.g., bean plants), to develop AI architectures to more accurately model the physiological conditions of the target, and to create a network of sw-NIRS sensors to simultaneously monitor a largescale crop.

The classification of EEG during motor imagery (MI) represents a challenging task in neuro-rehabilitation. In 2016, a deep learning (DL) model called EEGNet (based on CNN) and its variants attracted much attention for their ability to reach 80% accuracy in a 4-class MI classification. However, they can poorly explain their output decisions, preventing them from definitely solving questions related to inter-subject variability, generalization, and optimal classification. In this paper, we propose vEEGNet, a new model based on EEGNet, whose objective is now two-fold: it is used to classify MI, but also to reconstruct (and eventually generate) EEG signals. The work is still preliminary, but we are able to show that vEEGNet is able to classify 4 types of MI with performances at the state of the art, and, more interestingly, we found out that the reconstructed signals are consistent with the so-called motor-related cortical potentials, very specific and well-known motorrelated EEG patterns. Thus, jointly training vEEGNet to both classify and reconstruct EEG might lead it, in the future, to decrease the inter-subject performance variability, and also to generate new EEG samples to augment small datasets to improve classification, with a consequent strong impact on neuro-rehabilitation.

Recent studies in the context of machine learning have shown the effectiveness of deep attentional mechanisms for identifying important communities and relationships within a given input network. These studies can be effectively applied in those contexts where capturing specific dependencies, while downloading useless content, is essential to take decisions and provide accurate inference. This is the case, for example, of current recommender systems that exploit social information as a clever source of recommendations and / or explanations. In this paper we extend the social engine of our educational platform “WhoTeach” to leverage social information for educational services. In particular, we report our work in progress for providing “WhoTeach” with an attentional-based recommander system oriented to the design of programmes and courses for new teachers.

An emerging paradigm of Explainable Recommender Systems (ERS) leverages social friend information to supply users with his/her friends' public interests as explained recommendation. In this work, we consider this issue from a theoretical pont of view. We define a computational problem aimed to support the formation of communities of patients (users) and health services (items), thus stimulating targeted communication in social spaces. Using this formulation, dedicated ERS can benefit from social information to supply optimized recommended clarification. In particular, we report the conceptual framework with numerical results applying random graph models.

Recommender Systems have became extremely appealing for all technology enhanced learning researches aimed to design, develop and test technical innovations which support and enhance learning and teaching practices of both individuals and organizations. In this scenario a new emerging paradigm of explainable Recommander Systems leverages social friend information to provide (social) explanations in order to supply users with his/her friends’ public interests as explained recommendation. In this paper we introduce our educational platform called “WhoTeach”, an innovative and original system to integrate knowledge discovery, social networks analysis, and educational services. In particular, we report here our work in progress for providing “WhoTeach” environment with optimized Social Explainable Recommandations oriented to design new teachers’ programmes and courses.

Social interaction allows to support the disease management by creating online spaces where patients can interact with clinicians, and share experiences with other patients. Therefore, promoting targeted communication in online social spaces is a means to group patients around shared goals, offer emotional support, and finally engage patients in their healthcare decision making process. In this paper, we approach the argument from a theoretical perspective: we design an optimization problem aimed to encourage the creation of (induced) sub-networks of patients which, being recently diagnosed, wish to deepen the knowledge about their medical treatment with some other similar profiled patients, which have already been followed up by specific (even alternative) care centers. In particular, due to the computational hardness of the proposed problem, we provide approximated solutions based on distributed heuristics (i.e., Genetic Algorithms). Results are given for simulated data using Erdos-Renyi random graphs.

Inspired by a recent approach for genome reconstruction from incomplete data, we consider a variant of the longest common subsequence problem for the comparison of two sequences, one of which is incomplete, i.e. it has some missing elements. The new combinatorial problem, called Longest Filled Common Subsequence, given two sequences A and B, and a multiset M of symbols missing in B, asks for a sequence B* obtained by inserting the symbols of M into B so that B* induces a common subsequence with A of maximum length. First, we investigate the computational and approximation complexity of the problem and we show that it is NP-hard and APX-hard when A contains at most two occurrences of each symbol. Then, we give a 3/5 approximation algorithm for the problem. Finally, we present a fixed-parameter algorithm, when the problem is parameterized by the number of symbols inserted in B that "match" symbols of A.

Since 2001, in the Cystic Fibrosis Center of the Pediatric Hospital Bambino Gesù in Rome, we use telemedicine for monitoring of our patients. While in our first published works reporting this experience, we showed statistically significant reduction in hospital admissions and a tendency over time towards a better stability of the respiratory function for telehomecare (THC) patients, here we focus on the trend of the Forced Expiratory Volume in the first second (FEV1). In particular, we investigate the evolution of the clinical trend of the FEV1 index, by monitoring the activities of home patients from 2011 to 2014. THC is applied in addition to the standard therapeutic protocol by following 16 Cystic Fibrosis (CF) patients with specialized doctors. Our results show that THC patients improve their FEV1 values with a trend which can be considered significantly better than the one reported by the control group.

Tamoxifen is currently used for the treatment of breast cancer. Response to tamoxifen in metastatic conditions is a primary issue in cancer development. We used a cohort of breast cancer patients, treated or not with tamoxifen, and combined these data with the gene signature of metastatic samples in order to investigate the genetic mechanism of metastasis development, in search of a possible therapeutic effect of tamoxifen in metastatic conditions,. The analysis revealed a group of 21 genes common both to the set of up regulated genes in metastatic BC patients and to the set of down regulated genes in tamoxifen treated patients. These genes could be used as biomarkers for tamoxifen-sensitivity in order to optimize BC treatment.

"Signal" alignments play critical roles in many clinical setting. This is the case of mass spectrometry data, an important component of many types of proteomic analysis. A central problem occurs when one needs to integrate (mass spectrometry) data produced by different sources, e.g., different equipment and/or laboratories. In these cases some form of "data integration'" or "data fusion'" may be necessary in order to discard some source specific aspects and improve the ability to perform a classification task such as inferring the "disease classes'" of patients. The need for new high performance data alignments methods is therefore particularly important in these contexts. In this paper we propose an approach based both on an information theory perspective, generally used in a feature construction problem, and on the application of a mathematical programming task (i.e. the weighted bipartite matching problem). We present the results of a competitive analysis of our method against other approaches. The analysis was conducted on data from plasma/ethylenediaminetetraacetic acid (EDTA) of "control" and Alzheimer patients collected from three different hospitals. The results point to a significant performance advantage of our method with respect to the competing ones tested.

We present a hybrid system for managing both symbolic and subsymbolic knowledge in a uniform way. Our aim is to solve problems where some gap in formal theories occurs which stops us from getting a fully symbolical solution. The idea is to use neural modules to functionally connect pieces of symbolical knowledge, such as mathematical formulas and deductive rules. The whole system is trained through a backpropagation learning algorithm where all (symbolic or subsymbolic) free parameters are updated piping back the error through each component of the system. The structure of this system is very general, possibly varying over time, possibly managing fuzzy variables and decision trees. We use as a test-bed the problem of sorting a file, where suitable suggestions on next sorting moves are supplied by the network also on the basis of hints provided by some conventional sorters. A comprehensive discussion of system performance is provided in order to understand behaviors and capabilities of the proposed hybrid system.