Characterisation of nonlinear receptive fields of visual neurons by convolutional neural network

Ukita, Jumpei; Yoshida, Takashi; Ohki, Kenichi

doi:10.1038/s41598-019-40535-4

Cited by 24 publications

(20 citation statements)

References 59 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We visualized the preferred features of neurons in CNNs trained with the modified database using the DeepDream response-maximization approach [78]. This analysis indicated that neurons in early layers developed circular receptive fields (RFs) reminiscent of retinal ganglion and LGN cells; in deeper layers, they resembled oriented RFs akin to both simple and complex cells (S5 Fig), as previously reported, e.g., [59,[79][80][81][82]. Paralleling stimulation protocols typically used in experimental recordings in visual areas, we then examined CNNs responses to oriented gratings.…”

Section: Visual Stability During Saccadic Eye Movementsmentioning

confidence: 69%

Motor-related signals support localization invariance for stable visual perception

Benucci

2022

PLoS Comput Biol

View full text Add to dashboard Cite

Our ability to perceive a stable visual world in the presence of continuous movements of the body, head, and eyes has puzzled researchers in the neuroscience field for a long time. We reformulated this problem in the context of hierarchical convolutional neural networks (CNN)—whose architectures have been inspired by the hierarchical signal processing of the mammalian visual system—and examined perceptual stability as an optimization process that identifies image-defining features for accurate image classification in the presence of movements. Movement signals, multiplexed with visual inputs along overlapping convolutional layers, aided classification invariance of shifted images by making the classification faster to learn and more robust relative to input noise. Classification invariance was reflected in activity manifolds associated with image categories emerging in late CNN layers and with network units acquiring movement-associated activity modulations as observed experimentally during saccadic eye movements. Our findings provide a computational framework that unifies a multitude of biological observations on perceptual stability under optimality principles for image classification in artificial neural networks.

show abstract

Section: Visual Stability During Saccadic Eye Movementsmentioning

confidence: 69%

Motor-related signals support localization invariance for stable visual perception

Benucci

2022

PLoS Comput Biol

View full text Add to dashboard Cite

show abstract

“…An important milestone towards this goal is to achieve general system identification models that can predict the response of large populations of neurons to arbitrary visual inputs. In recent years, deep neural networks have set new standards in predicting responses in the visual system (Yamins et al, 2014; Vintch et al, 2015; Antolík et al, 2016; Cadena et al, 2017; Batty et al, 2016; Kindel et al, 2017; Klindt et al, 2017; Zhang et al, 2018; Ecker et al, 2018; Sinz et al, 2018) and the ability to yield novel response characterizations (Walker et al, 2019; Bashivan et al, 2019; Ponce et al, 2019; Kindel et al, 2019; Ukita et al, 2019).…”

Section: Introductionmentioning

confidence: 99%

Generalization in data-driven models of primary visual cortex

Lurz

Bashiri

Willeke

et al. 2020

Preprint

View full text Add to dashboard Cite

Deep neural networks (DNN) have set new standards at predicting responses of neural populations to visual input. Most such DNNs consist of a convolutional network (core) shared across all neurons which learns a representation of neural computation in visual cortex and a neuron-specific readout that linearly combines the relevant features in this representation. The goal of this paper is to test whether such a representation is indeed generally characteristic for visual cortex, i.e. generalizes between animals of a species, and what factors contribute to obtaining such a generalizing core. To push all non-linear computations into the core where the generalizing cortical features should be learned, we devise a novel readout that reduces the number of parameters per neuron in the readout by up to two orders of magnitude compared to the previous state-of-the-art. It does so by taking advantage of retinotopy and learns a Gaussian distribution over the neuron’s receptive field position. With this new readout we train our network on neural responses from mouse primary visual cortex (V1) and obtain a gain in performance of 7% compared to the previous state-of-the-art network. We then investigate whether the convolutional core indeed captures general cortical features by using the core in transfer learning to a different animal. When transferring a core trained on thousands of neurons from various animals and scans we exceed the performance of training directly on that animal by 12%, and outperform a commonly used VGG16 core pre-trained on imagenet by 33%. In addition, transfer learning with our data-driven core is more data-efficient than direct training, achieving the same performance with only 40% of the data. Our model with its novel readout thus sets a new state-of-the-art for neural response prediction in mouse visual cortex from natural images, generalizes between animals, and captures better characteristic cortical features than current task-driven pre-training approaches such as VGG16.

show abstract

“…By calculating the area under the curve (AUC) of the receiver operating characteristic curve using 5-fold cross-validation (DuBrava et al, 2017; Le et al, 2019b; Le, Ho & Ou, 2017; Ukita, Yoshida & Ohki, 2019), we performed feature selection and hyperparameters optimization to determine the models. Calculation of AUC was performed using an R library named ‘ROCR’ (version 1.0.7).…”

Section: Methodsmentioning

confidence: 99%

Machine learning analysis to identify the association between risk factors and onset of nosocomial diarrhea: a retrospective cohort study

Kurisu

Yoshiuchi

Ogino

et al. 2019

PeerJ

View full text Add to dashboard Cite

BackgroundAlthough several risk factors for nosocomial diarrhea have been identified, the detail of association between these factors and onset of nosocomial diarrhea, such as degree of importance or temporal pattern of influence, remains unclear. We aimed to determine the association between risk factors and onset of nosocomial diarrhea using machine learning algorithms.MethodsWe retrospectively collected data of patients with acute cerebral infarction. Seven variables, including age, sex, modified Rankin Scale (mRS) score, and number of days of antibiotics, tube feeding, proton pump inhibitors, and histamine 2-receptor antagonist use, were used in the analysis. We split the data into a training dataset and independant test dataset. Based on the training dataset, we developed a random forest, support vector machine (SVM), and radial basis function (RBF) network model. By calculating an area under the curve (AUC) of the receiver operating characteristic curve using 5-fold cross-validation, we performed feature selection and hyperparameter optimization in each model. According to their final performances, we selected the optimal model and also validated it in the independent test dataset. Based on the selected model, we visualized the variable importance and the association between each variable and the outcome using partial dependence plots.ResultsTwo-hundred and eighteen patients were included. In the cross-validation within the training dataset, the random forest model achieved an AUC of 0.944, which was higher than in the SVM and RBF network models. The random forest model also achieved an AUC of 0.832 in the independent test dataset. Tube feeding use days, mRS score, antibiotic use days, age and sex were strongly associated with the onset of nosocomial diarrhea, in this order. Tube feeding use had an inverse U-shaped association with the outcome. The mRS score and age had a convex downward and increasing association, while antibiotic use had a convex upward association with the outcome.ConclusionWe revealed the degree of importance and temporal pattern of the influence of several risk factors for nosocomial diarrhea, which could help clinicians manage nosocomial diarrhea.

show abstract

Characterisation of nonlinear receptive fields of visual neurons by convolutional neural network

Cited by 24 publications

References 59 publications

Motor-related signals support localization invariance for stable visual perception

Motor-related signals support localization invariance for stable visual perception

Generalization in data-driven models of primary visual cortex

Machine learning analysis to identify the association between risk factors and onset of nosocomial diarrhea: a retrospective cohort study

Contact Info

Product

Resources

About