How best to evaluate synthesized images has been a longstanding problem in image-to-image translation, and to date remains largely unresolved. This paper proposes a novel approach that combines signals of image quality between paired source and transformation to predict the latter's similarity with a hypothetical ground truth. We trained a Multi-Method Fusion (MMF) model via an ensemble of gradient-boosted regressors using Image Quality Assessment (IQA) metrics to predict Deep Image Structure and Texture Similarity (DISTS), enabling models to be ranked without the need for ground truth data. Analysis revealed the task to be feature-constrained, introducing a trade-off at inference between metric computation time and prediction accuracy. The MMF model we present offers an efficient way to automate the evaluation of synthesized images, and by extension the image-to-image translation models that generated them.

The thalamus is the primary gateway that relays sensory information to the cerebral cortex. While a single recipient cortical cell receives the convergence of many principal relay cells of the thalamus, each thalamic cell in turn integrates a dense and distributed synaptic feedback from the cortex. During sensory processing, the influence of this functional loop remains largely ignored. Using dynamic-clamp techniques in thalamic slices in vitro, we combined theoretical and experimental approaches to implement a realistic hybrid retino-thalamo-cortical pathway mixing biological cells and simulated circuits. The synaptic bombardment of cortical origin was mimicked through the injection of a stochastic mixture of excitatory and inhibitory conductances, resulting in a gradable correlation level of afferent activity shared by thalamic cells. The study of the impact of the simulated cortical input on the global retinocortical signal transfer efficiency revealed a novel control mechanism resulting from the collective resonance of all thalamic relay neurons. We show here that the transfer efficiency of sensory input transmission depends on three key features: i) the number of thalamocortical cells involved in the many-to-one convergence from thalamus to cortex, ii) the statistics of the corticothalamic synaptic bombardment and iii) the level of correlation imposed between converging thalamic relay cells. In particular, our results demonstrate counterintuitively that the retinocortical signal transfer efficiency increases when the level of correlation across thalamic cells decreases. This suggests that the transfer efficiency of relay cells could be selectively amplified when they become simultaneously desynchronized by the cortical feedback. When applied to the intact brain, this network regulation mechanism could direct an attentional focus to specific thalamic subassemblies and select the appropriate input lines to the cortex according to the descending influence of cortically-defined "priors".

L'identification du codage neuronal dans le thalamus et le cortex cerebral, et en particulier dans l'aire visuelle primaire, se heurte a la complexite du reseau neuronal qui repose sur une diversite etonnante des neurones, sur les plans morphologique, biochimique et electrique, et de leurs connexions synaptiques. A cela s'ajoute une importante diversite des proprietes fonctionnelles de ces neurones refletant en grande partie la forte recurrence des connexions synaptiques au sein des reseaux corticaux ainsi que la boucle cortico-thalamo-corticale. En d'autres termes, le calcul global effectue dans le reseau thalamo-cortical influence, via des milliers de connexions synaptiques excitatrices et inhibitrices, la specificite de la reponse de chaque neurone. Dans une premiere partie, nous avons developpe un modele de bombardement synaptique contextuel reproduisant la dynamique de milliers de synapses excitatrices et inhibitrices convergeant vers un neurone cortical avec l'avantage de pouvoir parametrer le niveau de synchronisation des synapses afferentes. Nous montrons que le niveau de synchronisation synaptique est relie au taux de correlation de l'activite neuronale sous-liminaire dans le cortex visuel du chat, avec d'une part un regime ou le codage neuronal est tres redondant pour des stimulations artificielles classiquement utilisees du type reseau de luminance sinusoidale, et d'autre part un regime ou le codage neuronal est beaucoup plus riche presentant moins de correlation neuronale pour des stimulations naturelles. Ces resultats indiquent que le taux de correlation de l'activite neuronale sous-liminaire est un indicateur fonctionnel du regime de codage dans lequel est engage le cortex cerebral. Dans une seconde partie, nous avons etendu l'exploration du codage neuronal au thalamus, passerelle principale qui transmet les informations sensorielles en provenance de la peripherie vers le cortex cerebral. Le thalamus recoit un fort retour cortico-thalamique qui resulte du calcul global effectue par les aires corticales. Nous avons etudie son influence en modelisant une voie retino-thalamo-corticale mixant neurones artificiels et neurones biologiques in vitro dans laquelle un bombardement synaptique d'origine corticale est mime via l'injection de conductances stochastiques excitatrices et inhibitrices en clamp dynamique. Cette approche confere l'avantage de pouvoir controller individuellement chacun des neurones thalamiques dans la voie artificielle. Nous montrons qu'un processus de facilitation stochastique a l'echelle de la population s'adjoint au gain cellulaire classique pour controler le transfert de l'information sensorielle de la retine au cortex visuel primaire. Ce processus de facilitation stochastique, qui n'aurait pas pu etre discerne a l'echelle de la cellule individuelle, est gouverne par le taux de correlation inter-neuronale de l'activite neuronale dans le thalamus. A l'inverse des conceptions classiques, -un fort taux de decorrelation- optimise le transfert sensoriel de la retine au cortex en favorisant la synchronisation des afferences synaptiques. Nous suggerons qu'une decorrelation induite par les aires corticales pourrait augmenter l'efficacite du transfert pour certaines assemblees cellulaires dans le thalamus, constituant ainsi un mecanisme attentionnel a l'echelle des circuits thalamo-corticaux. En parallele, nous avons developpe une methode d'extraction des fluctuations des conductances synaptiques des neurones a partir d'enregistrements intracellulaires unitaires. Cette methode devrait permettre de raffiner nos connaissances sur la nature des contextes synaptiques dans lesquels sont immerges les neurones avec des retombees potentielles sur le developpement de nouveaux modeles de bombardements synaptiques. En conclusion, nos travaux confirment l'hypothese d'un codage neuronal base sur la synchronisation synaptique conditionnee par le niveau de correlation de l'activite neuronale. Nos travaux sont coherents avec de nombreuses etudes sur les processus attentionnels et suggerent que des mecanismes de correlation et decorrelation actives, ainsi que des activites oscillatoires, pourraient reguler le transfert de l'information entre les organes sensoriels et les aires corticales.

The thalamic output during different behavioral states is strictly controlled by the firing modes of thalamocortical neurons. During sleep, their hyperpolarized membrane potential allows activation of the T-type calcium channels, promoting rhythmic high-frequency burst firing that reduces sensory information transfer. In contrast, in the waking state thalamic neurons mostly exhibit action potentials at low frequency (i.e., tonic firing), enabling the reliable transfer of incoming sensory inputs to cortex. Because of their nearly complete inactivation at the depolarized potentials that are experienced during the wake state, T-channels are not believed to modulate tonic action potential discharges. Here, we demonstrate using mice brain slices that activation of T-channels in thalamocortical neurons maintained in the depolarized/wake-like state is critical for the reliable expression of tonic firing, securing their excitability over changes in membrane potential that occur in the depolarized state. Our results establish a novel mechanism for the integration of sensory information by thalamocortical neurons and point to an unexpected role for T-channels in the early stage of information processing.

Variations of excitatory and inhibitory conductances determine the membrane potential (V(m)) activity of neurons, as well as their spike responses, and are thus of primary importance. Methods to estimate these conductances require clamping the cell at several different levels of V(m), thus making it impossible to estimate conductances from "single trial" V(m) recordings. We present here a new method that allows extracting estimates of the full time course of excitatory and inhibitory conductances from single-trial V(m) recordings. This method is based on oversampling of the V(m). We test the method numerically using models of increasing complexity. Finally, the method is evaluated using controlled conductance injection in cortical neurons in vitro using the dynamic-clamp technique. This conductance extraction method should be very useful for future in vivo applications.

Various types of neural-based signals, such as EEG, local field potentials and intracellular synaptic potentials, integrate multiple sources of activity distributed across large assemblies. They have in common a power-law frequency-scaling structure at high frequencies, but it is still unclear whether this scaling property is dominated by intrinsic neuronal properties or by network activity. The latter case is particularly interesting because if frequency-scaling reflects the network state it could be used to characterize the functional impact of the connectivity. In intracellularly recorded neurons of cat primary visual cortex in vivo, the power spectral density of Vm activity displays a power-law structure at high frequencies with a fractional scaling exponent. We show that this exponent is not constant, but depends on the visual statistics used to drive the network. To investigate the determinants of this frequency-scaling, we considered a generic recurrent model of cortex receiving a retinotopically organized external input. Similarly to the in vivo case, our in computo simulations show that the scaling exponent reflects the correlation level imposed in the input. This systematic dependence was also replicated at the single cell level, by controlling independently, in a parametric way, the strength and the temporal decay of the pairwise correlation between presynaptic inputs. This last model was implemented in vitro by imposing the correlation control in artificial presynaptic spike trains through dynamic-clamp techniques. These in vitro manipulations induced a modulation of the scaling exponent, similar to that observed in vivo and predicted in computo. We conclude that the frequency-scaling exponent of the Vm reflects stimulus-driven correlations in the cortical network activity. Therefore, we propose that the scaling exponent could be used to read-out the “effective” connectivity responsible for the dynamical signature of the population signals measured at different integration levels, from Vm to LFP, EEG and fMRI.

The power spectra of many nonlinear systems can display power-law frequency scaling with a fractional exponent. Such fractal behavior has been reported using various observables of neural activity from spike train statistics [4] to population dynamics [5]. However the origin of such scaling invariance remains controversial [6]. The analysis of the subthreshold membrane activity (Vm) in single neurons can help to dissect out the possible origins of the power-law. In the anesthetized and paralyzed cat, we recorded the intracellular response of V1 neurons to visual stimuli of various spatio-temporal statistics. We observed that, for each cell, the fractal exponent changes in a rather systematic way according to the stimulus. Such a dependency implies that local network dynamics are involved in the genesis of this power-law. We designed a model which derives the fractal exponent modulation from the topology of the functional connectivity. Using Cluster Point Process theory [2], we showed analytically that this effect depends on synchrony within local assemblies and that the emergence of power-law is further determined by the scaling property of network interactions. We led numerical simulations to confirm the power-law change at the Vm level. To check if those variations can be influenced by the cell intrinsic properties, this paradigm has been investigated in vitro using dynamic-clamp injections of generated conductances that reproduce fractal behaviour.