Search Results
Fiber-based SORS-SERDS system and chemometrics for the diagnostics and therapy monitoring of psoriasis inflammatory disease in vivo
Psoriasis is considered a widespread dermatological disease that can strongly affect the quality of life. Currently, the treatment is continued until the skin surface appears clinically healed. However, lesions appearing normal may contain modifications in deeper layers. To terminate the treatment too early can highly increase the risk of relapses. Therefore, techniques are needed for a better knowledge of the treatment process, especially to detect the lesion modifications in deeper layers. In this study, we developed a fiber-based SORS-SERDS system in combination with machine learning algorithms to non-invasively determine the treatment efficiency of psoriasis. The system was designed to acquire Raman spectra from three different depths into the skin, which provide rich information about the skin modifications in deeper layers. This way, it is expected to prevent the occurrence of relapses in case of a too short treatment. The method was verified with a study of 24 patients upon their two visits: the data is acquired at the beginning of a standard treatment (visit 1) and four months afterwards (visit 2). A mean sensitivity of ≥85% was achieved to distinguish psoriasis from normal skin at visit 1. At visit 2, where the patients were healed according to the clinical appearance, the mean sensitivity was ≈65%.
Application of machine learning to object manipulation with bio-inspired microstructures
Bioinspired fibrillar adhesives have been proposed for novel gripping systems with enhanced scalability and resource efficiency. Here, we propose an in-situ optical monitoring system of the contact signatures, coupled with image processing and machine learning. Visual features were extracted from the contact signature images recorded at maximum compressive preload and after lifting a glass object. The algorithm was trained to cope with several degrees of misalignment and with unbalanced weight distributions by off-center gripping. The system allowed an assessment of the picking process for objects of various mass (200, 300, and 400 g). Several classifiers showed a high accuracy of about 90 % for successful prediction of attachment, depending on the mass of the object. The results promise improved reliability of handling objects, even in difficult situations.
Topology of products similarity network for market forecasting
The detection and prediction of risk in financial markets is one of the main challenges of economic forecasting, and draws much attention from the scientific community. An even more challenging task is the prediction of the future relative gain of companies. We here develop a novel combination of product text analysis, network theory and topological based machine learning to study the future performance of companies in financial markets. Our network links are based on the similarity of firms’ products and constructed using the Securities Exchange Commission (SEC) filings of US listed firms. We find that several topological features of this network can serve as good precursors of risks or future gain of companies. We then apply machine learning to network attributes vectors for each node to predict successful and failing firms. The resulting accuracies are much better than current state of the art techniques. The framework presented here not only facilitates the prediction of financial markets but also provides insight and demonstrates the power of combining network theory and topology based machine learning.
Metadata analysis of open educational resources
Open Educational Resources (OERs) are openly licensed educational materials that are widely used for learning. Nowadays, many online learning repositories provide millions of OERs. Therefore, it is exceedingly difficult for learners to find the most appropriate OER among these resources. Subsequently, the precise OER metadata is critical for providing high-quality services such as search and recommendation. Moreover, metadata facilitates the process of automatic OER quality control as the continuously increasing number of OERs makes manual quality control extremely difficult. This work uses the metadata of 8,887 OERs to perform an exploratory data analysis on OER metadata. Accordingly, this work proposes metadata-based scoring and prediction models to anticipate the quality of OERs. Based on the results, our analysis demonstrated that OER metadata and OER content qualities are closely related, as we could detect high-quality OERs with an accuracy of 94.6%. Our model was also evaluated on 884 educational videos from Youtube to show its applicability on other educational repositories.
Machine learning for additive manufacturing: Predicting materials characteristics and their uncertainty
Additive manufacturing (AM) is known for versatile fabrication of complex parts, while also allowing the synthesis of materials with desired microstructures and resulting properties. These benefits come at a cost: process control to manufacture parts within given specifications is very challenging due to the relevance of a large number of processing parameters. Efficient predictive machine learning (ML) models trained on small datasets, can minimize this cost. They also allow to assess the quality of the dataset inclusive of uncertainty. This is important in order for additively manufactured parts to meet property specifications not only on average, but also within a given variance or uncertainty. Here, we demonstrate this strategy by developing a heteroscedastic Gaussian process (HGP) model, from a dataset based on laser powder bed fusion of a glass-forming alloy at varying processing parameters. Using amorphicity as the microstructural descriptor, we train the model on our Zr52.5Cu17.9Ni14.6Al10Ti5 (at.%) alloy dataset. The HGP model not only accurately predicts the mean value of amorphicity, but also provides the respective uncertainty. The quantification of the aleatoric and epistemic uncertainty contributions allows to assess intrinsic inaccuracies of the dataset, as well as identify underlying physical phenomena. This HGP model approach enables to systematically improve ML-driven AM processes.
Using Bayesian Networks to Investigate the Influence of Subseasonal Arctic Variability on Midlatitude North Atlantic Circulation
Recent enhanced warming and sea ice depletion in the Arctic have been put forward as potential drivers of severe weather in the midlatitudes. Evidence of a link between Arctic warming and midlatitude atmospheric circulation is growing, but the role of Arctic processes relative to other drivers remains unknown. Arctic–midlatitude connections in the North Atlantic region are particularly complex but important due to the frequent occurrence of severe winters in recent decades. Here, dynamic Bayesian networks with hidden variables are introduced to the field to assess their suitability for teleconnection analyses. Climate networks are constructed to analyze North Atlantic circulation variability at 5-day to monthly time scales during the winter months of the years 1981–2018. The inclusion of a number of Arctic, midlatitude, and tropical variables allows for an investigation into the relative role of Arctic influence compared to internal atmospheric variability and other remote drivers. A robust covariability between regions of amplified Arctic warming and two definitions of midlatitude circulation is found to occur entirely within winter at submonthly time scales. Hidden variables incorporated in networks represent two distinct modes of stratospheric polar vortex variability, capturing a periodic shift between average conditions and slower anomalous flow. The influence of the Barents–Kara Seas region on the North Atlantic Oscillation is found to be the strongest link at 5- and 10-day averages, while the stratospheric polar vortex strongly influences jet variability on monthly time scales.
Calibrating mini-mental state examination scores to predict misdiagnosed dementia patients
Mini-Mental State Examination (MMSE) is used as a diagnostic test for dementia to screen a patient’s cognitive assessment and disease severity. However, these examinations are often inaccurate and unreliable either due to human error or due to patients’ physical disability to correctly interpret the questions as well as motor deficit. Erroneous data may lead to a wrong assessment of a specific patient. Therefore, other clinical factors (e.g., gender and comorbidities) existing in electronic health records, can also play a significant role, while reporting her examination results. This work considers various clinical attributes of dementia patients to accurately determine their cognitive status in terms of the Mini-Mental State Examination (MMSE) Score. We employ machine learning models to calibrate MMSE score and classify the correctness of diagnosis among patients, in order to assist clinicians in a better understanding of the progression of cognitive impairment and subsequent treatment. For this purpose, we utilize a curated real-world ageing study data. A random forest prediction model is employed to estimate the Mini-Mental State Examination score, related to the diagnostic classification of patients.This model uses various clinical attributes to provide accurate MMSE predictions, succeeding in correcting an important percentage of cases that contain previously identified miscalculated scores in our dataset. Furthermore, we provide an effective classification mechanism for automatically identifying patient episodes with inaccurate MMSE values with high confidence. These tools can be combined to assist clinicians in automatically finding episodes within patient medical records where the MMSE score is probably miscalculated and estimating what the correct value should be. This provides valuable support in the decision making process for diagnosing potential dementia patients.
The flare likelihood and region eruption forecasting (FLARECAST) project: flare forecasting in the big data & machine learning era
The European Union funded the FLARECAST project, that ran from January 2015 until February 2018. FLARECAST had a research-to-operations (R2O) focus, and accordingly introduced several innovations into the discipline of solar flare forecasting. FLARECAST innovations were: first, the treatment of hundreds of physical properties viewed as promising flare predictors on equal footing, extending multiple previous works; second, the use of fourteen (14) different machine learning techniques, also on equal footing, to optimize the immense Big Data parameter space created by these many predictors; third, the establishment of a robust, three-pronged communication effort oriented toward policy makers, space-weather stakeholders and the wider public. FLARECAST pledged to make all its data, codes and infrastructure openly available worldwide. The combined use of 170+ properties (a total of 209 predictors are now available) in multiple machine-learning algorithms, some of which were designed exclusively for the project, gave rise to changing sets of best-performing predictors for the forecasting of different flaring levels, at least for major flares. At the same time, FLARECAST reaffirmed the importance of rigorous training and testing practices to avoid overly optimistic pre-operational prediction performance. In addition, the project has (a) tested new and revisited physically intuitive flare predictors and (b) provided meaningful clues toward the transition from flares to eruptive flares, namely, events associated with coronal mass ejections (CMEs). These leads, along with the FLARECAST data, algorithms and infrastructure, could help facilitate integrated space-weather forecasting efforts that take steps to avoid effort duplication. In spite of being one of the most intensive and systematic flare forecasting efforts to-date, FLARECAST has not managed to convincingly lift the barrier of stochasticity in solar flare occurrence and forecasting: solar flare prediction thus remains inherently probabilistic.
Ranking facts for explaining answers to elementary science questions
In multiple-choice exams, students select one answer from among typically four choices and can explain why they made that particular choice. Students are good at understanding natural language questions and based on their domain knowledge can easily infer the question's answer by “connecting the dots” across various pertinent facts. Considering automated reasoning for elementary science question answering, we address the novel task of generating explanations for answers from human-authored facts. For this, we examine the practically scalable framework of feature-rich support vector machines leveraging domain-targeted, hand-crafted features. Explanations are created from a human-annotated set of nearly 5000 candidate facts in the WorldTree corpus. Our aim is to obtain better matches for valid facts of an explanation for the correct answer of a question over the available fact candidates. To this end, our features offer a comprehensive linguistic and semantic unification paradigm. The machine learning problem is the preference ordering of facts, for which we test pointwise regression versus pairwise learning-to-rank. Our contributions, originating from comprehensive evaluations against nine existing systems, are (1) a case study in which two preference ordering approaches are systematically compared, and where the pointwise approach is shown to outperform the pairwise approach, thus adding to the existing survey of observations on this topic; (2) since our system outperforms a highly-effective TF-IDF-based IR technique by 3.5 and 4.9 points on the development and test sets, respectively, it demonstrates some of the further task improvement possibilities (e.g., in terms of an efficient learning algorithm, semantic features) on this task; (3) it is a practically competent approach that can outperform some variants of BERT-based reranking models; and (4) the human-engineered features make it an interpretable machine learning model for the task.
Master Memory Function for Delay-Based Reservoir Computers With Single-Variable Dynamics
We show that many delay-based reservoir computers considered in the literature can be characterized by a universal master memory function (MMF). Once computed for two independent parameters, this function provides linear memory capacity for any delay-based single-variable reservoir with small inputs. Moreover, we propose an analytical description of the MMF that enables its efficient and fast computation. Our approach can be applied not only to single-variable delay-based reservoirs governed by known dynamical rules, such as the Mackey–Glass or Stuart–Landau-like systems, but also to reservoirs whose dynamical model is not available.