Reducing Oversmoothing through Informed Weight Initialization in Graph Neural Networks

In this work, we generalize the ideas of Kaiming initialization to Graph Neural Networks (GNNs) and propose a new scheme (G-Init) that reduces oversmoothing, leading to very good results in node and graph classification tasks. GNNs are commonly initialized using methods designed for other types of Neural Networks, overlooking the underlying graph topology. We analyze theoretically the variance of signals flowing forward and gradients flowing backward in the class of convolutional GNNs. We then simplify our analysis to the case of the GCN and propose a new initialization method. Our results indicate that the new method (G-Init) reduces oversmoothing in deep GNNs, facilitating their effective use. Experimental validation supports our theoretical findings, demonstrating the advantages of deep networks in scenarios with no feature information for unlabeled nodes (i.e., ``cold start'' scenario).

Partially Trained Graph Convolutional Networks Resist Oversmoothing

In this work we investigate an observation made by Kipf \& Welling, who suggested that untrained GCNs can generate meaningful node embeddings. In particular, we investigate the effect of training only a single layer of a GCN, while keeping the rest of the layers frozen. We propose a basis on which the effect of the untrained layers and their contribution to the generation of embeddings can be predicted. Moreover, we show that network width influences the dissimilarity of node embeddings produced after the initial node features pass through the untrained part of the model. Additionally, we establish a connection between partially trained GCNs and oversmoothing, showing that they are capable of reducing it. We verify our theoretical results experimentally and show the benefits of using deep networks that resist oversmoothing, in a ``cold start'' scenario, where there is a lack of feature information for unlabeled nodes.

Complex Event Recognition with Symbolic Register Transducers: Extended Technical Report

We present a system for Complex Event Recognition (CER) based on automata. While multiple such systems have been described in the literature, they typically suffer from a lack of clear and denotational semantics, a limitation which often leads to confusion with respect to their expressive power. In order to address this issue, our system is based on an automaton model which is a combination of symbolic and register automata. We extend previous work on these types of automata, in order to construct a formalism with clear semantics and a corresponding automaton model whose properties can be formally investigated. We call such automata Symbolic Register Transducers (SRT). We show that SRT are closed under various operators, but are not in general closed under complement and they are not determinizable. However, they are closed under these operations when a window operator, quintessential in Complex Event Recognition, is used. We show how SRT can be used in CER in order to detect patterns upon streams of events, using our framework that provides declarative and compositional semantics, and that allows for a systematic treatment of such automata. For SRT to work in pattern detection, we allow them to mark events from the input stream as belonging to a complex event or not, hence the name "transducers". We also present an implementation of SRT which can perform CER. We compare our SRT-based CER engine against other state-of-the-art CER systems and show that it is both more expressive and more efficient.

Overview of BioASQ 2023: The eleventh BioASQ challenge on Large-Scale Biomedical Semantic Indexing and Question Answering

This is an overview of the eleventh edition of the BioASQ challenge in the context of the Conference and Labs of the Evaluation Forum (CLEF) 2023. BioASQ is a series of international challenges promoting advances in large-scale biomedical semantic indexing and question answering. This year, BioASQ consisted of new editions of the two established tasks b and Synergy, and a new task (MedProcNER) on semantic annotation of clinical content in Spanish with medical procedures, which have a critical role in medical practice. In this edition of BioASQ, 28 competing teams submitted the results of more than 150 distinct systems in total for the three different shared tasks of the challenge. Similarly to previous editions, most of the participating systems achieved competitive performance, suggesting the continuous advancement of the state-of-the-art in the field.

Financial misstatement detection: a realistic evaluation

In this work, we examine the evaluation process for the task of detecting financial reports with a high risk of containing a misstatement. This task is often referred to, in the literature, as ``misstatement detection in financial reports''. We provide an extensive review of the related literature. We propose a new, realistic evaluation framework for the task which, unlike a large part of the previous work: (a) focuses on the misstatement class and its rarity, (b) considers the dimension of time when splitting data into training and test and (c) considers the fact that misstatements can take a long time to detect. Most importantly, we show that the evaluation process significantly affects system performance, and we analyze the performance of different models and feature types in the new realistic framework.

Analysing Biomedical Knowledge Graphs using Prime Adjacency Matrices

Most phenomena related to biomedical tasks are inherently complex, and in many cases, are expressed as signals on biomedical Knowledge Graphs (KGs). In this work, we introduce the use of a new representation framework, the Prime Adjacency Matrix (PAM) for biomedical KGs, which allows for very efficient network analysis. PAM utilizes prime numbers to enable representing the whole KG with a single adjacency matrix and the fast computation of multiple properties of the network. We illustrate the applicability of the framework in the biomedical domain by working on different biomedical knowledge graphs and by providing two case studies: one on drug-repurposing for COVID-19 and one on important metapath extraction. We show that we achieve better results than the original proposed workflows, using very simple methods that require no training, in considerably less time.

Large-scale fine-grained semantic indexing of biomedical literature based on weakly-supervised deep learning

Semantic indexing of biomedical literature is usually done at the level of MeSH descriptors, representing topics of interest for the biomedical community. Several related but distinct biomedical concepts are often grouped together in a single coarse-grained descriptor and are treated as a single topic for semantic indexing. This study proposes a new method for the automated refinement of subject annotations at the level of concepts, investigating deep learning approaches. Lacking labelled data for this task, our method relies on weak supervision based on concept occurrence in the abstract of an article. The proposed approach is evaluated on an extended large-scale retrospective scenario, taking advantage of concepts that eventually become MeSH descriptors, for which annotations become available in MEDLINE/PubMed. The results suggest that concept occurrence is a strong heuristic for automated subject annotation refinement and can be further enhanced when combined with dictionary-based heuristics. In addition, such heuristics can be useful as weak supervision for developing deep learning models that can achieve further improvement in some cases.

Overview of BioASQ 2022: The tenth BioASQ challenge on Large-Scale Biomedical Semantic Indexing and Question Answering

This paper presents an overview of the tenth edition of the BioASQ challenge in the context of the Conference and Labs of the Evaluation Forum (CLEF) 2022. BioASQ is an ongoing series of challenges that promotes advances in the domain of large-scale biomedical semantic indexing and question answering. In this edition, the challenge was composed of the three established tasks a, b, and Synergy, and a new task named DisTEMIST for automatic semantic annotation and grounding of diseases from clinical content in Spanish, a key concept for semantic indexing and search engines of literature and clinical records. This year, BioASQ received more than 170 distinct systems from 38 teams in total for the four different tasks of the challenge. As in previous years, the majority of the competing systems outperformed the strong baselines, indicating the continuous advancement of the state-of-the-art in this domain.

Efficient multi-relational network representation using primes

Multi-relational networks play an important role in today's world and are utilized to capture complex relationships between the data. Their applications span many domains such as biomedical, financial, social, etc., and because of their increasing usability, it becomes crucial to find efficient ways to deal with the added complexity of multiple layers. In this work, we propose a novel approach to represent these complex networks using a single aggregated adjacency matrix, by utilizing primes as surrogates for the relations. Due to the fundamental theorem of arithmetic, this allows for a lossless, compact representation of the whole multi-relational graph, using a single adjacency matrix. Moreover, this representation enables the fast computation of multi-hop adjacency matrices, that can be useful for a variety of downstream tasks. We present simple and complex tasks in which this representation can be useful and showcase its efficiency and performance. Finally, we also provide insights on the advantages and the open challenges that still need to be addressed and motivate future work.

Learning Automata-Based Complex Event Patterns in Answer Set Programming

Complex Event Recognition and Forecasting (CER/F) techniques attempt to detect, or even forecast ahead of time, event occurrences in streaming input using predefined event patterns. Such patterns are not always known in advance, or they frequently change over time, making machine learning techniques, capable of extracting such patterns from data, highly desirable in CER/F. Since many CER/F systems use symbolic automata to represent such patterns, we propose a family of such automata where the transition-enabling conditions are defined by Answer Set Programming (ASP) rules, and which, thanks to the strong connections of ASP to symbolic learning, are directly learnable from data. We present such a learning approach in ASP and an incremental version thereof that trades optimality for efficiency and is capable to scale to large datasets. We evaluate our approach on two CER datasets and compare it to state-of-the-art automata learning techniques, demonstrating empirically a superior performance, both in terms of predictive accuracy and scalability.