Invariant odor recognition with ON–OFF neural ensembles

Invariant stimulus recognition is a challenging pattern-recognition problem that must be dealt with by all sensory systems. Since neural responses evoked by a stimulus are perturbed in a multitude of ways, how can this computational capability be achieved? We examine this issue in the locust olfactory system. We find that locusts trained in an appetitive-conditioning assay robustly recognize the trained odorant independent of variations in stimulus durations, dynamics, or history, or changes in background and ambient conditions. However, individual- and population-level neural responses vary unpredictably with many of these variations. Our results indicate that linear statistical decoding schemes, which assign positive weights to ON neurons and negative weights to OFF neurons, resolve this apparent confound between neural variability and behavioral stability. Furthermore, simplification of the decoder using only ternary weights ({+1, 0, −1}) (i.e., an “ON-minus-OFF” approach) does not compromise performance, thereby striking a fine balance between simplicity and robustness.

FosGFP expression does not capture a sensory learning-related engram in superficial layers of mouse barrel cortex

Immediate-early gene (IEG) expression has been used to identify small neural ensembles linked to a particular experience, based on the principle that a selective subset of activated neurons will encode specific memories or behavioral responses. The majority of these studies have focused on “engrams” in higher-order brain areas where more abstract or convergent sensory information is represented, such as the hippocampus, prefrontal cortex, or amygdala. In primary sensory cortex, IEG expression can label neurons that are responsive to specific sensory stimuli, but experience-dependent shaping of neural ensembles marked by IEG expression has not been demonstrated. Here, we use a fosGFP transgenic mouse to longitudinally monitor in vivo expression of the activity-dependent gene c-fos in superficial layers (L2/3) of primary somatosensory cortex (S1) during a whisker-dependent learning task. We find that sensory association training does not detectably alter fosGFP expression in L2/3 neurons. Although training broadly enhances thalamocortical synaptic strength in pyramidal neurons, we find that synapses onto fosGFP+ neurons are not selectively increased by training; rather, synaptic strengthening is concentrated in fosGFP− neurons. Taken together, these data indicate that expression of the IEG reporter fosGFP does not facilitate identification of a learning-specific engram in L2/3 in barrel cortex during whisker-dependent sensory association learning.

Scene Preferences, Aesthetic Appeal, and Curiosity

A gradient of µ-opioid receptors extends from early sensory areas of the cerebral cortex to associative cortex, with the greatest density of receptors in the most anterior associative regions. In 2006, Biederman and Vessel proposed that the hedonic value of perceptual and cognitive experience is a function of activation of this gradient. A desire for opioid activity provided by this gradient renders us infovores, always seeking novel but richly interpretable experiences. Richly interpretable experiences engage the opioid-dense anterior regions of the gradient, while novel experiences engage neural ensembles that have yet to undergo adaptation. Support for this proposal derives from the greater activity elicited in opioid-rich parahippocampal cortex for preferred over nonpreferred scenes, with neural network modeling of visual aesthetic responses suggesting that representations in later stages are more predictive of aesthetic responses, and psychopharmacological experiments that support the potential involvement of endogenous opioids.

Third party stabilization of unstable coordination in systems of coupled oscillators

The Haken-Kelso-Bunz (HKB) system of equations is a well-developed model for dyadic rhythmic coordination in biological systems. It captures ubiquitous empirical observations of bistability – the coexistence of in-phase and antiphase motion – in neural, behavioral, and social coordination. Recent work by Zhang and colleagues has generalized HKB to many oscillators to account for new empirical phenomena observed in multiagent interaction. Utilising this generalization, the present work examines how the coordination dynamics of a pair of oscillators can be augmented by virtue of their coupling to a third oscillator. We show that stable antiphase coordination emerges in pairs of oscillators even when their coupling parameters would have prohibited such coordination in their dyadic relation. We envision two lines of application for this theoretical work. In the social sciences, our model points toward the development of intervention strategies to support coordination behavior in heterogeneous groups (for instance in gerontology, when younger and older individuals interact). In neuroscience, our model will advance our understanding of how the direct functional connection of mesoscale or microscale neural ensembles might be switched by their changing coupling to other neural ensembles. Our findings illuminate a crucial property of complex systems: how the whole is different than the system’s parts.

Malaria Parasite Detection using Efficient Neural Ensembles

Caused by the bite of the Anopheles mosquito infected with the parasite of genus Plasmodium, malaria has remained a major burden towards healthcare for years with an approximate 400,000 deaths reported globally every year. The traditional diagnosis process for malaria involves an examination of the blood smear slide under the microscope. This process is not only time consuming but also requires pathologists to be highly skilled in their work. Timely diagnosis and availability of robust diagnostic facilities and skilled laboratory technicians are very much vital to reduce the mortality rate. This study aims to build a robust system by applying deep learning techniques such as transfer learning and snapshot ensembling to automate the detection of the parasite in the thin blood smear images. All the models were evaluated against the following metrics - F1 score, Accuracy, Precision, Recall, Mathews Correlation Coefficient (MCC), Area Under the Receiver Operating Characteristics (AUC-ROC) and the Area under the Precision Recall curve (AUC-PR). The snapshot ensembling model created by combining the snapshots of the EfficientNet-B0 pre-trained model outperformed every other model achieving a f1 score - 99.37%, precision - 99.52% and recall - 99.23%. The results show the potential of model ensembles which combine the predictive power of multiple weal models to create a single efficient model that is better equipped to handle the real world data. The GradCAM experiment displayed the gradient activation maps of the last convolution layer to visually explicate where and what a model sees in an image to classify them into a particular class. The models in this study correctly activate the stained parasitic region of interest in the thin blood smear images. Such visuals make the model more transparent, explainable, and trustworthy which are very much essential for deploying AI based models in the healthcare network.

Malaria Parasite Detection using Efficient Neural Ensembles

Caused by the bite of the Anopheles mosquito infected with the parasite of genus Plasmodium, malaria has remained a major burden towards healthcare for years with an approximate 400,000 deaths reported globally every year. The traditional diagnosis process for malaria involves an examination of the blood smear slide under the microscope. This process is not only time consuming but also requires pathologists to be highly skilled in their work. Timely diagnosis and availability of robust diagnostic facilities and skilled laboratory technicians are very much vital to reduce the mortality rate. This study aims to build a robust system by applying deep learning techniques such as transfer learning and snapshot ensembling to automate the detection of the parasite in the thin blood smear images. All the models were evaluated against the following metrics - F1 score, Accuracy, Precision, Recall, Mathews Correlation Coefficient (MCC), Area Under the Receiver Operating Characteristics (AUC-ROC) and the Area under the Precision Recall curve (AUC-PR). The snapshot ensembling model created by combining the snapshots of the EfficientNet-B0 pre-trained model outperformed every other model achieving a f1 score - 99.37%, precision - 99.52% and recall - 99.23%. The results show the potential of model ensembles which combine the predictive power of multiple weal models to create a single efficient model that is better equipped to handle the real world data. The GradCAM experiment displayed the gradient activation maps of the last convolution layer to visually explicate where and what a model sees in an image to classify them into a particular class. The models in this study correctly activate the stained parasitic region of interest in the thin blood smear images. Such visuals make the model more transparent, explainable, and trustworthy which are very much essential for deploying AI based models in the healthcare network.

Prefrontal neural ensembles encode an internal model of visual sequences and their violations

Theories of predictive coding hypothesize that cortical networks learn internal models of environmental regularities to generate expectations that are constantly compared with sensory inputs. The prefrontal cortex (PFC) is thought to be critical for predictive coding. Here, we show how prefrontal neuronal ensembles encode a detailed internal model of sequences of visual events and their violations. We recorded PFC ensembles in a visual local-global sequence paradigm probing low and higher-order predictions and mismatches. PFC ensembles formed distributed, overlapping representations for all aspects of the dynamically unfolding sequences, including information about image identity as well as abstract information about ordinal position, anticipated sequence pattern, mismatches to local and global structure, and model updates. Model and mismatch signals were mixed in the same ensembles, suggesting a revision of predictive processing models that consider segregated processing. We conclude that overlapping prefrontal ensembles may collectively encode all aspects of an ongoing visual experience, including anticipation, perception, and surprise.

Zebrafish capable of generating future state prediction error show improved active avoidance behavior in virtual reality

Animals make decisions under the principle of reward value maximization and surprise minimization. It is still unclear how these principles are represented in the brain and are reflected in behavior. We addressed this question using a closed-loop virtual reality system to train adult zebrafish for active avoidance. Analysis of the neural activity of the dorsal pallium during training revealed neural ensembles assigning rules to the colors of the surrounding walls. Additionally, one third of fish generated another ensemble that becomes activated only when the real perceived scenery shows discrepancy from the predicted favorable scenery. The fish with the latter ensemble escape more efficiently than the fish with the former ensembles alone, even though both fish have successfully learned to escape, consistent with the hypothesis that the latter ensemble guides zebrafish to take action to minimize this prediction error. Our results suggest that zebrafish can use both principles of goal-directed behavior, but with different behavioral consequences depending on the repertoire of the adopted principles.

BIFURCATING NEURONS IN THE ANTERIOR THALAMIC NUCLEI

The anterior thalamic nuclei (ATN) form a nodal point within a hippocampal-cingulate-diencephalic memory system. ATN projections to different brain structures are conventionally viewed as distinct, but ATN neurons may send collaterals to multiple structures. The anteromedial subregion (AM) is the primary source of efferents to the medial prefrontal cortex (mPFC). Using a dual-retrograde neurotracer strategy, we discovered bifurcating AM neurons for tracers placed in the mPFC when paired with other regions. A semi-quantitative analysis found a high proportion of AM neurons (~36%) showed collateral projections when the mPFC was paired with dorsal subiculum (dSub); 20% were evident for mPFC paired with caudal retrosplenial cortex (cRSC); and 6% was found for mPFC and ventral hippocampal formation (vHF). About 10% of bifurcating AM neurons was also identified when the mPFC was not included, that is, for cRSC with dSub, and cRSC with vHF. Similar percentages of bifurcating neurons were also found within the anterior region of the adjacent nucleus reuniens (Re). The high frequency of bifurcating neurons suggests a new perspective for ATN function. These neurons would facilitate direct coordination among distal neural ensembles to support episodic memory and may explain why the ATN is a critical region for diencephalic amnesia.