Preventing profiling for ethical fake news detection.

A news article's online audience provides useful insights about the article's identity. However, fake news classifiers using such information risk relying on profiling. In response to the rising demand for ethical AI, we present a profiling-avoiding algorithm that leverages Twitter users during model optimisation while excluding them when an article's veracity is evaluated. For this, we take inspiration from the social sciences and introduce two objective functions that maximise correlation between the article and its spreaders, and among those spreaders. We applied our profiling-avoiding algorithm to three popular neural classifiers and obtained results on fake news data discussing a variety of news topics. The positive impact on prediction performance demonstrates the soundness of the proposed objective functions to integrate social context in text-based classifiers. Moreover, statistical visualisation and dimension reduction techniques show that the user-inspired classifiers better discriminate between unseen fake and true news in their latent spaces. Our study serves as a stepping stone to resolve the underexplored issue of profiling-dependent decision-making in user-informed fake news detection.

Causal Factor Disentanglement for Few-Shot Domain Adaptation in Video Prediction.

An important challenge in machine learning is performing with accuracy when few training samples are available from the target distribution. If a large number of training samples from a related distribution are available, transfer learning can be used to improve the performance. This paper investigates how to do transfer learning more effectively if the source and target distributions are related through a Sparse Mechanism Shift for the application of next-frame prediction. We create Sparse Mechanism Shift-TempoRal Intervened Sequences (SMS-TRIS), a benchmark to evaluate transfer learning for next-frame prediction derived from the TRIS datasets. We then propose to exploit the Sparse Mechanism Shift property of the distribution shift by disentangling the model parameters with regard to the true causal mechanisms underlying the data. We use the Causal Identifiability from TempoRal Intervened Sequences (CITRIS) model to achieve this disentanglement via causal representation learning. We show that encouraging disentanglement with the CITRIS extensions can improve performance, but their effectiveness varies depending on the dataset and backbone used. We find that it is effective only when encouraging disentanglement actually succeeds in increasing disentanglement. We also show that an alternative method designed for domain adaptation does not help, indicating the challenging nature of the SMS-TRIS benchmark.

Justifying diagnosis decisions by deep neural networks.

An integrated approach is proposed across visual and textual data to both determine and justify a medical diagnosis by a neural network. As deep learning techniques improve, interest grows to apply them in medical applications. To enable a transition to workflows in a medical context that are aided by machine learning, the need exists for such algorithms to help justify the obtained outcome so human clinicians can judge their validity. In this work, deep learning methods are used to map a frontal X-ray image to a continuous textual representation. This textual representation is decoded into a diagnosis and the associated textual justification that will help a clinician evaluate the outcome. Additionally, more explanatory data is provided for the diagnosis by generating a realistic X-ray that belongs to the nearest alternative diagnosis. We demonstrate in a clinical expert opinion study on a subset of the X-ray data set from the Indiana University hospital network that our justification mechanism significantly outperforms existing methods that use saliency maps. While performing multi-task training with multiple loss functions, our method achieves excellent diagnosis accuracy and captioning quality when compared to current state-of-the-art single-task methods.

A deep learning approach to bilingual lexicon induction in the biomedical domain.

BACKGROUND: Bilingual lexicon induction (BLI) is an important task in the biomedical domain as translation resources are usually available for general language usage, but are often lacking in domain-specific settings. In this article we consider BLI as a classification problem and train a neural network composed of a combination of recurrent long short-term memory and deep feed-forward networks in order to obtain word-level and character-level representations. RESULTS: The results show that the word-level and character-level representations each improve state-of-the-art results for BLI and biomedical translation mining. The best results are obtained by exploiting the synergy between these word-level and character-level representations in the classification model. We evaluate the models both quantitatively and qualitatively. CONCLUSIONS: Translation of domain-specific biomedical terminology benefits from the character-level representations compared to relying solely on word-level representations. It is beneficial to take a deep learning approach and learn character-level representations rather than relying on handcrafted representations that are typically used. Our combined model captures the semantics at the word level while also taking into account that specialized terminology often originates from a common root form (e.g., from Greek or Latin).

Semi-supervised Learning for the BioNLP Gene Regulation Network.

BACKGROUND: The BioNLP Gene Regulation Task has attracted a diverse collection of submissions showcasing state-of-the-art systems. However, a principal challenge remains in obtaining a significant amount of recall. We argue that this is an important quality for Information Extraction tasks in this field. We propose a semi-supervised framework, leveraging a large corpus of unannotated data available to us. In this framework, the annotated data is used to find plausible candidates for positive data points, which are included in the machine learning process. As this is a method principally designed for gaining recall, we further explore additional methods to improve precision on top of this. These are: weighted regularisation in the SVM framework, and filtering out unlabelled examples based on a probabilistic rule-finding method. The latter method also allows us to add candidates for negatives from unlabelled data, a method not viable in the unfiltered approach. RESULTS: We replicate one of the original participant systems, and modify it to incorporate our methods. This allows us to test the extent of our proposed methods by applying them to the GRN task data. We find a considerable improvement in recall compared to the baseline system. We also investigate the evaluation metrics and find several mechanisms explaining a bias towards precision. Furthermore, these findings uncover an intricate precision-recall interaction, depriving recall of its habitual immediacy seen in traditional machine learning set-ups. CONCLUSION: Our contributions are twofold.

Structured learning for spatial information extraction from biomedical text: bacteria biotopes.

BACKGROUND: We aim to automatically extract species names of bacteria and their locations from webpages. This task is important for exploiting the vast amount of biological knowledge which is expressed in diverse natural language texts and putting this knowledge in databases for easy access by biologists. The task is challenging and the previous results are far below an acceptable level of performance, particularly for extraction of localization relationships. Therefore, we aim to design a new system for such extractions, using the framework of structured machine learning techniques. RESULTS: We design a new model for joint extraction of biomedical entities and the localization relationship. Our model is based on a spatial role labeling (SpRL) model designed for spatial understanding of unrestricted text. We extend SpRL to extract discourse level spatial relations in the biomedical domain and apply it on the BioNLP-ST 2013, BB-shared task. We highlight the main differences between general spatial language understanding and spatial information extraction from the scientific text which is the focus of this work. We exploit the text's structure and discourse level global features. Our model and the designed features substantially improve on the previous systems, achieving an absolute improvement of approximately 57 percent over F1 measure of the best previous system for this task. CONCLUSIONS: Our experimental results indicate that a joint learning model over all entities and relationships in a document outperforms a model which extracts entities and relationships independently. Our global learning model significantly improves the state-of-the-art results on this task and has a high potential to be adopted in other natural language processing (NLP) tasks in the biomedical domain.

Critical Analysis of Deconfounded Pretraining to Improve Visio-Linguistic Models.

An important problem with many current visio-linguistic models is that they often depend on spurious correlations. A typical example of a spurious correlation between two variables is one that is due to a third variable causing both (a "confounder"). Recent work has addressed this by adjusting for spurious correlations using a technique of deconfounding with automatically found confounders. We will refer to this technique as AutoDeconfounding. This article dives more deeply into AutoDeconfounding, and surfaces a number of issues of the original technique. First, we evaluate whether its implementation is actually equivalent to deconfounding. We provide an explicit explanation of the relation between AutoDeconfounding and the underlying causal model on which it implicitly operates, and show that additional assumptions are needed before the implementation of AutoDeconfounding can be equated to correct deconfounding. Inspired by this result, we perform ablation studies to verify to what extent the improvement on downstream visio-linguistic tasks reported by the works that implement AutoDeconfounding is due to AutoDeconfounding, and to what extent it is specifically due to the deconfounding aspect of AutoDeconfounding. We evaluate AutoDeconfounding in a way that isolates its effect, and no longer see the same improvement. We also show that tweaking AutoDeconfounding to be less related to deconfounding does not negatively affect performance on downstream visio-linguistic tasks. Furthermore, we create a human-labeled ground truth causality dataset for objects in a scene to empirically verify whether and how well confounders are found. We show that some models do indeed find more confounders than a random baseline, but also that finding more confounders is not correlated with performing better on downstream visio-linguistic tasks. Finally, we summarize the current limitations of AutoDeconfounding to solve the issue of spurious correlations and provide directions for the design of novel AutoDeconfounding methods that are aimed at overcoming these limitations.

Extracting relations from traditional Chinese medicine literature via heterogeneous entity networks.

OBJECTIVE: Traditional Chinese medicine (TCM) is a unique and complex medical system that has developed over thousands of years. This article studies the problem of automatically extracting meaningful relations of entities from TCM literature, for the purposes of assisting clinical treatment or poly-pharmacology research and promoting the understanding of TCM in Western countries. METHODS: Instead of separately extracting each relation from a single sentence or document, we propose to collectively and globally extract multiple types of relations (eg, herb-syndrome, herb-disease, formula-syndrome, formula-disease, and syndrome-disease relations) from the entire corpus of TCM literature, from the perspective of network mining. In our analysis, we first constructed heterogeneous entity networks from the TCM literature, in which each edge is a candidate relation, then used a heterogeneous factor graph model (HFGM) to simultaneously infer the existence of all the edges. We also employed a semi-supervised learning algorithm estimate the model's parameters. RESULTS: We performed our method to extract relations from a large dataset consisting of more than 100,000 TCM article abstracts. Our results show that the performance of the HFGM at extracting all types of relations from TCM literature was significantly better than a traditional support vector machine (SVM) classifier (increasing the average precision by 11.09%, the recall by 13.83%, and the F1-measure by 12.47% for different types of relations, compared with a traditional SVM classifier). CONCLUSION: This study exploits the power of collective inference and proposes an HFGM based on heterogeneous entity networks, which significantly improved our ability to extract relations from TCM literature.

Machine Reading for Extraction of Bacteria and Habitat Taxonomies.

There is a vast amount of scientific literature available from various resources such as the internet. Automating the extraction of knowledge from these resources is very helpful for biologists to easily access this information. This paper presents a system to extract the bacteria and their habitats, as well as the relations between them. We investigate to what extent current techniques are suited for this task and test a variety of models in this regard. We detect entities in a biological text and map the habitats into a given taxonomy. Our model uses a linear chain Conditional Random Field (CRF). For the prediction of relations between the entities, a model based on logistic regression is built. Designing a system upon these techniques, we explore several improvements for both the generation and selection of good candidates. One contribution to this lies in the extended exibility of our ontology mapper that uses an advanced boundary detection and assigns the taxonomy elements to the detected habitats. Furthermore, we discover value in the combination of several distinct candidate generation rules. Using these techniques, we show results that are significantly improving upon the state of art for the BioNLP Bacteria Biotopes task.