Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

General General

Interpretable PID parameter tuning for control engineering using general dynamic neural networks: An extensive comparison.

In PloS one ; h5-index 176.0

Modern automation systems largely rely on closed loop control, wherein a controller interacts with a controlled process via actions, based on observations. These systems are increasingly complex, yet most deployed controllers are linear Proportional-Integral-Derivative (PID) controllers. PID controllers perform well on linear and near-linear systems but their simplicity is at odds with the robustness required to reliably control complex processes. Modern machine learning techniques offer a way to extend PID controllers beyond their linear control capabilities by using neural networks. However, such an extension comes at the cost of losing stability guarantees and controller interpretability. In this paper, we examine the utility of extending PID controllers with recurrent neural networks--namely, General Dynamic Neural Networks (GDNN); we show that GDNN (neural) PID controllers perform well on a range of complex control systems and highlight how they can be a scalable and interpretable option for modern control systems. To do so, we provide an extensive study using four benchmark systems that represent the most common control engineering benchmarks. All control environments are evaluated with and without noise as well as with and without disturbances. The neural PID controller performs better than standard PID control in 15 of 16 tasks and better than model-based control in 13 of 16 tasks. As a second contribution, we address the lack of interpretability that prevents neural networks from being used in real-world control processes. We use bounded-input bounded-output stability analysis to evaluate the parameters suggested by the neural network, making them understandable for engineers. This combination of rigorous evaluation paired with better interpretability is an important step towards the acceptance of neural-network-based control approaches for real-world systems. It is furthermore an important step towards interpretable and safely applied artificial intelligence.

Günther Johannes, Reichensdörfer Elias, Pilarski Patrick M, Diepold Klaus


Surgery Surgery

Mapping risk of ischemic heart disease using machine learning in a Brazilian state.

In PloS one ; h5-index 176.0

Cardiovascular diseases are the leading cause of deaths globally. Machine learning studies predicting mortality rates for ischemic heart disease (IHD) at the municipal level are very limited. The goal of this paper was to create and validate a Heart Health Care Index (HHCI) to predict risk of IHD based on location and risk factors. Secondary data, geographical information system (GIS) and machine learning were used to validate the HHCI and stratify the IHD municipality risk in the state of Paraná. A positive spatial autocorrelation was found (Moran's I = 0.6472, p-value = 0.001), showing clusters of high IHD mortality. The Support Vector Machine, which had an RMSE of 0.789 and error proportion close to one (0.867), was the best for prediction among eight machine learning algorithms after validation. In the north and northwest regions of the state, HHCI was low and mortality clusters patterns were high. By creating an HHCI through ML, we can predict IHD mortality rate at municipal level, identifying predictive characteristics that impact health conditions of these localities' guided health management decisions for improvements for IHD within the emergency care network in the state of Paraná.

Bergamini Marcela, Iora Pedro Henrique, Rocha Thiago Augusto Hernandes, Tchuisseu Yolande Pokam, Dutra Amanda de Carvalho, Scheidt João Felipe Herman Costa, Nihei Oscar Kenji, de Barros Carvalho Maria Dalva, Staton Catherine Ann, Vissoci João Ricardo Nickenig, de Andrade Luciano


General General

Comprehensive nutrient analysis in agricultural organic amendments through non-destructive assays using machine learning.

In PloS one ; h5-index 176.0

Portable X-ray fluorescence (pXRF) and Diffuse Reflectance Fourier Transformed Mid-Infrared (DRIFT-MIR) spectroscopy are rapid and cost-effective analytical tools for material characterization. Here, we provide an assessment of these methods for the analysis of total Carbon, Nitrogen and total elemental composition of multiple elements in organic amendments. We developed machine learning methods to rapidly quantify the concentrations of macro- and micronutrient elements present in the samples and propose a novel system for the quality assessment of organic amendments. Two types of machine learning methods, forest regression and extreme gradient boosting, were used with data from both pXRF and DRIFT-MIR spectroscopy. Cross-validation trials were run to evaluate generalizability of models produced on each instrument. Both methods demonstrated similar broad capabilities in estimating nutrients using machine learning, with pXRF being suitable for nutrients and contaminants. The results make portable spectrometry in combination with machine learning a scalable solution to provide comprehensive nutrient analysis for organic amendments.

Towett Erick K, Drake Lee B, Acquah Gifty E, Haefele Stephan M, McGrath Steve P, Shepherd Keith D


General General

Decoding the neural dynamics of free choice in humans.

In PLoS biology

How do we choose a particular action among equally valid alternatives? Nonhuman primate findings have shown that decision-making implicates modulations in unit firing rates and local field potentials (LFPs) across frontal and parietal cortices. Yet the electrophysiological brain mechanisms that underlie free choice in humans remain ill defined. Here, we address this question using rare intracerebral electroencephalography (EEG) recordings in surgical epilepsy patients performing a delayed oculomotor decision task. We find that the temporal dynamics of high-gamma (HG, 60-140 Hz) neural activity in distinct frontal and parietal brain areas robustly discriminate free choice from instructed saccade planning at the level of single trials. Classification analysis was applied to the LFP signals to isolate decision-related activity from sensory and motor planning processes. Compared with instructed saccades, free-choice trials exhibited delayed and longer-lasting HG activity during the delay period. The temporal dynamics of the decision-specific sustained HG activity indexed the unfolding of a deliberation process, rather than memory maintenance. Taken together, these findings provide the first direct electrophysiological evidence in humans for the role of sustained high-frequency neural activation in frontoparietal cortex in mediating the intrinsically driven process of freely choosing among competing behavioral alternatives.

Thiery Thomas, Saive Anne-Lise, Combrisson Etienne, Dehgan Arthur, Bastin Julien, Kahane Philippe, Berthoz Alain, Lachaux Jean-Philippe, Jerbi Karim


Public Health Public Health

Ruling In and Ruling Out COVID-19: Computing SARS-CoV-2 Infection Risk From Symptoms, Imaging and Test Data.

In Journal of medical Internet research ; h5-index 88.0

BACKGROUND : Assigning meaningful probabilities of SARS-CoV-2 infection risk presents a diagnostic challenge across the continuum of care.

OBJECTIVE : Develop and clinically validate an adaptable, personalized diagnostic model to assist clinicians in ruling-in and ruling-out COVID-19 in potential patients. This study compares the diagnostic performance of probabilistic, graphical, and machine-learning models against a previously published benchmark model.

METHODS : We integrated patient symptom and test data using machine learning and Bayesian inference to quantify individual patient risk of SARS-CoV-2 infection. We trained models with 100,000 simulated patient profiles based on thirteen symptoms, estimated local prevalence, imaging, and molecular diagnostic performance from published reports. We tested these models with consecutive patients who presented with a COVID-19 compatible illness at the University of California San Diego Medical Center over 14 days starting in March 2020.

RESULTS : We included 55 consecutive patients with fever (78%) or cough (77%) presenting for ambulatory (n=11) or hospital care (n=44). 51% (n=28) were female, 49% were age <60. Common comorbidities included diabetes (22%), hypertension (27%), cancer (16%) and cardiovascular disease (13%). 69% of these (n=38) were RT-PCR confirmed positive for SARS-CoV-2 infection, 11 had repeated negative nucleic acid testing and an alternate diagnosis. Bayesian inference network, distance metric-learning, and ensemble models discriminated between patients with SARS-CoV-2 infection and alternate diagnoses with sensitivities of 81.6 - 84.2%, specificities of 58.8 - 70.6%, and accuracies of 61.4 - 71.8%. After integrating imaging and laboratory test statistics with the predictions of the Bayesian inference network, changes in diagnostic uncertainty at each step in the simulated clinical evaluation process were highly sensitive to location, symptom, and diagnostic test choices.

CONCLUSIONS : Decision support models that incorporate symptoms and available test results can help providers diagnose SARS-CoV-2 infection in real world settings.


D’Ambrosia Christopher, Christensen Henrik, Aronoff-Spencer Eliah


Radiology Radiology

Artificial intelligence can predict the mortality of COVID-19 patients at the admission time using routine blood samples.

In Journal of medical Internet research ; h5-index 88.0

BACKGROUND : COVID-19, which is accompanied by acute respiratory distress, multiple organ failure, and death, has spread worldwide much faster than previously thought. However, at present, it has limited treatments.

OBJECTIVE : To overcome this issue, we developed an artificial intelligence (AI) model of COVID-19, named EDRnet, to predict in-hospital mortality using a routine blood sample at the time of hospital admission.

METHODS : We selected 28 blood biomarkers and used the age and gender information of patients as model inputs. To improve the mortality prediction, we adopted an ensemble approach combining a deep neural network and random forest model. We trained our model with a database of blood samples from 361 COVID-19 patients in Wuhan, China, and applied it to 106 COVID-19 patients in three Korean medical institutions.

RESULTS : In the testing datasets, EDRnet provided high sensitivity (100%), specificity (91.35%), and accuracy (91.51%). To extend the number of patient data, we developed a web application (, where anyone can access the model to predict the mortality and can register his or her own blood laboratory results.

CONCLUSIONS : Our new AI model, EDRnet, accurately predicts the mortality rate for COVID-19. It is publicly available and aims to help healthcare providers fight COVID1-19 and improve patients' outcome.


Ko Hoon, Chung Heewon, Kang Wu Seong, Park Chul, Kim Do Wan, Kim Seong Eun, Chung Chi Ryang, Ko Ryoung Eun, Lee Hooseok, Seo Jae Ho, Choi Tae-Young, Jaimes Rafael, Kim Kyung Won, Lee Jinseok