scholarly journals Training deep neural density estimators to identify mechanistic models of neural dynamics

eLife ◽  
2020 ◽  
Vol 9 ◽  
Author(s):  
Pedro J Gonçalves ◽  
Jan-Matthis Lueckmann ◽  
Michael Deistler ◽  
Marcel Nonnenmacher ◽  
Kaan Öcal ◽  
...  
Keyword(s):  
Receptive Fields ◽  
Rhythmic Activity ◽  
Mechanistic Modeling ◽  
Neural Dynamics ◽  
Model Parameters ◽  
Initial Training ◽  
Neural Data ◽  
Compensation Mechanisms ◽  
Underlying Causes ◽  
Density Estimators

Mechanistic modeling in neuroscience aims to explain observed phenomena in terms of underlying causes. However, determining which model parameters agree with complex and stochastic neural data presents a significant challenge. We address this challenge with a machine learning tool which uses deep neural density estimators—trained using model simulations—to carry out Bayesian inference and retrieve the full space of parameters compatible with raw data or selected data features. Our method is scalable in parameters and data features and can rapidly analyze new data after initial training. We demonstrate the power and flexibility of our approach on receptive fields, ion channels, and Hodgkin–Huxley models. We also characterize the space of circuit configurations giving rise to rhythmic activity in the crustacean stomatogastric ganglion, and use these results to derive hypotheses for underlying compensation mechanisms. Our approach will help close the gap between data-driven and theory-driven models of neural dynamics.

scholarly journals Training deep neural density estimators to identify mechanistic models of neural dynamics

2019 ◽  
Author(s):  
Pedro J. Gonçalves ◽  
Jan-Matthis Lueckmann ◽  
Michael Deistler ◽  
Marcel Nonnenmacher ◽  
Kaan Öcal ◽  
...  
Keyword(s):  
Receptive Fields ◽  
Rhythmic Activity ◽  
Mechanistic Modeling ◽  
Neural Dynamics ◽  
Model Parameters ◽  
Initial Training ◽  
Neural Data ◽  
Compensation Mechanisms ◽  
Underlying Causes ◽  
Density Estimators

AbstractMechanistic modeling in neuroscience aims to explain observed phenomena in terms of underlying causes. However, determining which model parameters agree with complex and stochastic neural data presents a significant challenge. We address this challenge with a machine learning tool which uses deep neural density estimators— trained using model simulations— to carry out Bayesian inference and retrieve the full space of parameters compatible with raw data or selected data features. Our method is scalable in parameters and data features, and can rapidly analyze new data after initial training. We demonstrate the power and flexibility of our approach on receptive fields, ion channels, and Hodgkin–Huxley models. We also characterize the space of circuit configurations giving rise to rhythmic activity in the crustacean stomatogastric ganglion, and use these results to derive hypotheses for underlying compensation mechanisms. Our approach will help close the gap between data-driven and theory-driven models of neural dynamics.

scholarly journals An Integrated Approach of Mechanistic-Modeling and Machine-Learning for Thickness Optimization of Frozen Microwaveable Foods

Foods ◽  
2021 ◽  
Vol 10 (4) ◽  
pp. 763
Author(s):  
Ran Yang ◽  
Zhenbo Wang ◽  
Jiajia Chen
Keyword(s):  
Machine Learning ◽  
Learning Algorithm ◽  
Food Product ◽  
Integrated Approach ◽  
Mechanistic Modeling ◽  
Bayesian Optimization ◽  
Initial Training ◽  
Thickness Optimization ◽  
Heating Uniformity ◽  
Food Design

Mechanistic-modeling has been a useful tool to help food scientists in understanding complicated microwave-food interactions, but it cannot be directly used by the food developers for food design due to its resource-intensive characteristic. This study developed and validated an integrated approach that coupled mechanistic-modeling and machine-learning to achieve efficient food product design (thickness optimization) with better heating uniformity. The mechanistic-modeling that incorporated electromagnetics and heat transfer was previously developed and validated extensively and was used directly in this study. A Bayesian optimization machine-learning algorithm was developed and integrated with the mechanistic-modeling. The integrated approach was validated by comparing the optimization performance with a parametric sweep approach, which is solely based on mechanistic-modeling. The results showed that the integrated approach had the capability and robustness to optimize the thickness of different-shape products using different initial training datasets with higher efficiency (45.9% to 62.1% improvement) than the parametric sweep approach. Three rectangular-shape trays with one optimized thickness (1.56 cm) and two non-optimized thicknesses (1.20 and 2.00 cm) were 3-D printed and used in microwave heating experiments, which confirmed the feasibility of the integrated approach in thickness optimization. The integrated approach can be further developed and extended as a platform to efficiently design complicated microwavable foods with multiple-parameter optimization.

scholarly journals Boundary-Assisted Learning for Building Extraction from Optical Remote Sensing Imagery

2021 ◽  
Vol 13 (4) ◽  
pp. 760
Author(s):  
Sheng He ◽  
Wanshou Jiang
Keyword(s):  
Remote Sensing ◽  
Receptive Fields ◽  
Morphological Characteristics ◽  
Learning Task ◽  
Aerial Image ◽  
Model Parameters ◽  
Optical Remote Sensing ◽  
Building Extraction ◽  
Convolutional Network ◽  
Remote Sensing Imagery

Deep learning methods have been shown to significantly improve the performance of building extraction from optical remote sensing imagery. However, keeping the morphological characteristics, especially the boundaries, is still a challenge that requires further study. In this paper, we propose a novel fully convolutional network (FCN) for accurately extracting buildings, in which a boundary learning task is embedded to help maintain the boundaries of buildings. Specifically, in the training phase, our framework simultaneously learns the extraction of buildings and boundary detection and only outputs extraction results while testing. In addition, we introduce spatial variation fusion (SVF) to establish an association between the two tasks, thus coupling them and making them share the latent semantics and interact with each other. On the other hand, we utilize separable convolution with a larger kernel to enlarge the receptive fields while reducing the number of model parameters and adopt the convolutional block attention module (CBAM) to boost the network. The proposed framework was extensively evaluated on the WHU Building Dataset and the Inria Aerial Image Labeling Dataset. The experiments demonstrate that our method achieves state-of-the-art performance on building extraction. With the assistance of boundary learning, the boundary maintenance of buildings is ameliorated.

A mathematical model of cardiovascular dynamics for the diagnosis and prognosis of hemorrhagic shock

2021 ◽  
Author(s):  
Laura D’Orsi ◽  
Luciano Curcio ◽  
Fabio Cibella ◽  
Alessandro Borri ◽  
Lilach Gavish ◽  
...  
Keyword(s):  
Heart Rate ◽  
Cardiac Output ◽  
Arterial Pressure ◽  
Hemorrhagic Shock ◽  
Time Course ◽  
Differential Algebraic Equations ◽  
Model Parameters ◽  
Algebraic Equations ◽  
Cardiovascular Dynamics ◽  
Compensation Mechanisms

Abstract A variety of mathematical models of the cardiovascular system have been suggested over several years in order to describe the time-course of a series of physiological variables (i.e. heart rate, cardiac output, arterial pressure) relevant for the compensation mechanisms to perturbations, such as severe haemorrhage. The current study provides a simple but realistic mathematical description of cardiovascular dynamics that may be useful in the assessment and prognosis of hemorrhagic shock. The present work proposes a first version of a differential-algebraic equations model, the model dynamical ODE model for haemorrhage (dODEg). The model consists of 10 differential and 14 algebraic equations, incorporating 61 model parameters. This model is capable of replicating the changes in heart rate, mean arterial pressure and cardiac output after the onset of bleeding observed in four experimental animal preparations and fits well to the experimental data. By predicting the time-course of the physiological response after haemorrhage, the dODEg model presented here may be of significant value for the quantitative assessment of conventional or novel therapeutic regimens. The model may be applied to the prediction of survivability and to the determination of the urgency of evacuation towards definitive surgical treatment in the operational setting.

scholarly journals Inverse mechanistic modeling of transdermal drug delivery for fast identification of optimal model parameters

2020 ◽  
Author(s):  
Thijs Defraeye ◽  
Flora Bahrami ◽  
Rene M Rossi
Keyword(s):  
Drug Delivery ◽  
Inverse Modeling ◽  
Transdermal Drug Delivery ◽  
Drug Transport ◽  
Mechanistic Model ◽  
Mechanistic Modeling ◽  
Drug Uptake ◽  
Added Value ◽  
Model Parameters ◽  
Transdermal Drug Delivery Systems

Transdermal drug delivery systems are a key technology to administer drugs with a high first-pass effect in a non-invasive and controlled way. Physics-based modeling and simulation are on their way to become a cornerstone in the engineering of these healthcare devices since it provides a unique complementarity to experimental data and insights. Simulations enable to virtually probe the drug transport inside the skin at each point in time and space. However, the tedious experimental or numerical determination of material properties currently forms a bottleneck in the modeling workflow. We show that multiparameter inverse modeling to determine the drug diffusion and partition coefficients is a fast and reliable alternative. We demonstrate this strategy for transdermal delivery of fentanyl. We found that inverse modeling reduced the normalized root mean square deviation of the measured drug uptake flux from 26 to 9%, when compared to the experimental measurement of all skin properties. We found that this improved agreement with experiments was only possible if the diffusion in the reservoir holding the drug was smaller than the experimentally-measured diffusion coefficients suggested. For indirect inverse modeling, which systematically explores the entire parametric space, 30 000 simulations were required. By relying on direct inverse modeling, we reduced the number of simulations to be performed to only 300, so a factor 100 difference. The modeling approach's added value is that it can be calibrated once in-silico for all model parameters simultaneously by solely relying on a single measurement of the drug uptake flux evolution over time. We showed that this calibrated model could accurately be used to simulate transdermal patches with other drug doses. We showed that inverse modeling is a fast way to build up an accurate mechanistic model for drug delivery. This strategy opens the door to clinically-ready therapy that is tailored to patients.

scholarly journals Leave-One-Trial-Out (LOTO): A general approach to link single-trial parameters of cognitive models to neural data

2018 ◽  
Author(s):  
Sebastian Gluth ◽  
Nachshon Meiran
Keyword(s):  
Cognitive Model ◽  
Parameter Estimates ◽  
Model Parameters ◽  
Single Trial ◽  
Neural Data ◽  
Brain Signals ◽  
Complete Dataset ◽  
Advance Research ◽  
The Difference ◽  
Parameter Values

AbstractIt has become a key goal of model-based neuroscience to estimate trial-by-trial fluctuations of cognitive model parameters for linking these fluctuations to brain signals. However, previously developed methods were limited by being difficulty to implement, time-consuming, or model-specific. Here, we propose an easy, efficient and general approach to estimating trial-wise changes in parameters: Leave-One-Trial-Out (LOTO). The rationale behind LOTO is that the difference between the parameter estimates for the complete dataset and for the dataset with one omitted trial reflects the parameter value in the omitted trial. We show that LOTO is superior to estimating parameter values from single trials and compare it to previously proposed approaches. Furthermore, the method allows distinguishing true variability in a parameter from noise and from variability in other parameters. In our view, the practicability and generality of LOTO will advance research on tracking fluctuations in latent cognitive variables and linking them to neural data.

scholarly journals Patient-specific logic models of signaling pathways from screenings on cancer biopsies to prioritize personalized combination therapies

2018 ◽  
Author(s):  
Federica Eduati ◽  
Patricia Jaaks ◽  
Christoph A. Merten ◽  
Mathew J. Garnett ◽  
Julio Saez- Rodriguez
Keyword(s):  
Signaling Pathways ◽  
Large Scale ◽  
Dynamic Models ◽  
Ex Vivo ◽  
Response To Therapy ◽  
Mechanistic Modeling ◽  
Patient Specific ◽  
Model Parameters ◽  
Specific Response ◽  
Tumor Stages

AbstractMechanistic modeling of signaling pathways mediating patient-specific response to therapy can help to unveil resistance mechanisms and improve therapeutic strategies. Yet, creating such models for patients, in particular for solid malignancies, is challenging. A major hurdle to build these models is the limited material available, that precludes the generation of large-scale perturbation data. Here, we present an approach that couples ex vivo high-throughput screenings of cancer biopsies using microfluidics with logic-based modeling to generate patient-specific dynamic models of extrinsic and intrinsic apoptosis signaling pathways. We used the resulting models to investigate heterogeneity in pancreatic cancer patients, showing dissimilarities especially in the PI3K-Akt pathway. Variation in model parameters reflected well the different tumor stages. Finally, we used our dynamic models to efficaciously predict new personalized combinatorial treatments. Our results suggest our combination of microfluidic experiments and mathematical model can be a novel tool toward cancer precision medicine.

scholarly journals Significance of Model Parameter Variations in the pARD-RSC Model

2020 ◽  
Vol 4 (3) ◽  
pp. 109
Author(s):  
Armin Kech ◽  
Susanne Kugler ◽  
Tim Osswald
Keyword(s):  
Fiber Orientation ◽  
Short Fiber ◽  
Mechanistic Modeling ◽  
Model Parameters ◽  
Fiber Reinforced ◽  
Initial Alignment ◽  
Polybutylene Terephthalate ◽  
Fiber Volume ◽  
Input Parameters ◽  
Set Up

This study aims to evaluate how fiber orientation results are dependent on fluctuations in input parameters, such as the average fiber length, fiber volume content, and initial alignment of fibers. The range of parameters is restricted to deviations within one specific short fiber reinforced thermoplastic and is not set up to investigate the differences between materials. The evaluation was conducted by a virtual shear cell based on a mechanistic modeling approach. The fiber orientation prediction model discussed is the pARD-RSC (principal anisotropic rotary diffusion-reduced strain closure) model implemented as a user routine in AUTODESK MOLDFLOW INSIGHT® (AMI®). The material investigated was discontinuous short glass fiber reinforced PBT (polybutylene-terephthalate), which is often used for housings in various industries. It is shown that variation in the input parameters, although having an influence on the fiber orientation model parameters, only affects the final orientation moderately.

scholarly journals A Compact High-Quality Image Demosaicking Neural Network for Edge-Computing Devices

Sensors ◽  
2021 ◽  
Vol 21 (9) ◽  
pp. 3265
Author(s):  
Shuyu Wang ◽  
Mingxin Zhao ◽  
Runjiang Dou ◽  
Shuangming Yu ◽  
Liyuan Liu ◽  
...  
Keyword(s):  
Neural Network ◽  
Rapid Development ◽  
Receptive Fields ◽  
Image Features ◽  
Edge Computing ◽  
Model Parameters ◽  
Image Size ◽  
Feature Maps ◽  
Gaussian Smoothing ◽  
High Quality Image

Image demosaicking has been an essential and challenging problem among the most crucial steps of image processing behind image sensors. Due to the rapid development of intelligent processors based on deep learning, several demosaicking methods based on a convolutional neural network (CNN) have been proposed. However, it is difficult for their networks to run in real-time on edge computing devices with a large number of model parameters. This paper presents a compact demosaicking neural network based on the UNet++ structure. The network inserts densely connected layer blocks and adopts Gaussian smoothing layers instead of down-sampling operations before the backbone network. The densely connected blocks can extract mosaic image features efficiently by utilizing the correlation between feature maps. Furthermore, the block adopts depthwise separable convolutions to reduce the model parameters; the Gaussian smoothing layer can expand the receptive fields without down-sampling image size and discarding image information. The size constraints on the input and output images can also be relaxed, and the quality of demosaicked images is improved. Experiment results show that the proposed network can improve the running speed by 42% compared with the fastest CNN-based method and achieve comparable reconstruction quality as it on four mainstream datasets. Besides, when we carry out the inference processing on the demosaicked images on typical deep CNN networks, Mobilenet v1 and SSD, the accuracy can also achieve 85.83% (top 5) and 75.44% (mAP), which performs comparably to the existing methods. The proposed network has the highest computing efficiency and lowest parameter number through all methods, demonstrating that it is well suitable for applications on modern edge computing devices.

Sign in / Sign up

Export Citation Format

Share Document