Biology subjects

Luczak, A.

Publications and source records attributed to Luczak, A..

6 recordsLinked to original sources

bioRxiv2022

Combining Backpropagation with Equilibrium Propagation to improve an Actor-Critic RL framework

Backpropagation has been used to train neural networks for many years, allowing them to solve a wide variety of tasks like image classification, speech recognition, and reinforcement learning tasks. But the biological plausibility of backpropagation as a mechanism of neural learning has been questioned. Equilibrium Propagation (EP) has been proposed as a more biologically plausible alternative and achieves comparable accuracy on the CIFAR-10 image classification task. This study proposes the first EP-based reinforcement learning architecture: an actor-critic architecture with the actor network trained by EP. We show that this model can solve the basic control tasks often used as benchmarks for BP-based models. Interestingly, our trained model demonstrates more consistent high-reward behavior than a comparable model trained exclusively by backpropagation.

neuroscience↗

bioRxiv2020

Sensory experience selectively reorganizes the late component of evoked responses

In response to sensory stimulation, the cortex exhibits an early transient response followed by a late and slower activation pattern. Recent studies suggest that the early component represents features of the stimulus while the late component is associated with stimulus perception. Although very informative, these studies only focus on the amplitude of the evoked responses to study its relationship with sensory perception. In this work we expand upon the study of how patterns of evoked and spontaneous activity are modified by experience at mesoscale level using voltage and extracellular glutamate transient recordings over widespread regions of mice dorsal neocortex. We find that repeated tactile or auditory stimulation selectively modifies the spatiotemporal patterns of cortical activity, mainly of the late evoked response in anesthetized mice injected with amphetamine and also in awake mice. This modification lasted up to 60 minutes and results not only in an increase in amplitude of the late response after repeated stimulation, but also in an increase in the similarity between the spatiotemporal patterns of the late and the early evoked response patterns in anesthetized mice. This similarity increase occurs only for the evoked responses of the sensory modality that received the repeated stimulation. Thus, this selective long-lasting spatiotemporal modification of the cortical activity patterns might provide evidence that evoked responses are a cortex-wide phenomenon. This work opens new questions about how perception-related cortical activity changes with sensory experience across the cortex.

neuroscience↗

bioRxiv2020

Neurons learn by predicting future activity.

Understanding how the brain learns may lead to machines with human-like intellectual capacities. However, learning mechanisms in the brain are still not well understood. Here we demonstrate that the ability of a neuron to predict its future activity may provide an effective mechanism for learning in the brain. We show that comparing a neurons predicted activity with the actual activity provides a useful learning signal for modifying synaptic weights. Interestingly, this predictive learning rule can be derived from a metabolic principle, where neurons need to minimize their own synaptic activity (cost), while maximizing their impact on local blood supply by recruiting other neurons. This reveals an unexpected connection that learning in neural networks could result from simply maximizing the energy balance by each neuron. We show how this mathematically derived learning rule can provide a theoretical connection between diverse types of brain-inspired algorithms, such as: Hebbs rule, BCM theory, temporal difference learning and predictive coding. Thus, this may offer a step toward development of a general theory of neuronal learning. We validated this predictive learning rule in neural network simulations and in data recorded from awake animals. We found that in the sensory cortex it is indeed possible to predict a neurons activity [~]10-20ms into the future. Moreover, in response to stimuli, cortical neurons changed their firing rate to minimize surprise: i.e. the difference between actual and expected activity, as predicted by our model. Our results also suggest that spontaneous brain activity provides "training data" for neurons to learn to predict cortical dynamics. Thus, this work demonstrates that the ability of a neuron to predict its future inputs could be an important missing element to understand computation in the brain.

neuroscience↗

bioRxiv2020

Data-driven analysis of rat pup behavior reveals novel effects of maternal pre-conception exposure to nicotine

We present a deep neural network for data-driven analyses of infant rat behavior in an open field task. The network was applied to study the effect of maternal nicotine exposure prior to conception on offspring motor development. The neural network outperformed human expert designed animal locomotion measures in distinguishing rat pups born to nicotine exposed dams versus control dams. Notably, the network discovered novel movement alterations in posture, movement initiation and a stereotypy in "warm-up" behavior (the initiation of movement along specific dimensions) that were predictive of nicotine exposure. The results suggest that maternal preconception nicotine exposure delays and alters offspring motor development. In summary, we demonstrated that a deep neural network can automatically assess animal behavior with high accuracy, and that it offers a data-driven approach to investigating pharmacological effects on brain development. SignificanceRelating neuronal activity to behavior is crucial to understand brain function. Despite the staggering progress in monitoring brain activity, behavioral analyses still do not differ much from methods developed 30-50 years ago. The reason for that is the difficulty for automated video analyses to detect small differences in complex movements. Here we show that applying deep neuronal networks for automated video analyses can help to solve this problem. More importantly, knowledge extracted from the network allowed to identify subtle changes in multiple behavioral components, which were caused by maternal preconception nicotine exposure in rat pups. Thus, the examples presented here show how neuronal networks can guide the development of more accurate behavioral tests to assess symptoms of neurological disorders.

neuroscience↗

bioRxiv2020

Spatiotemporal structure of sensory-evoked and spontaneous activity revealed by mesoscale imaging in anesthetized and awake mice

Brain activity propagates across the cortex in diverse spatiotemporal patterns, both as a response to sensory stimulation and during spontaneous activity. Despite been extensively studied, the relationship between the characteristics of such patterns during spontaneous and evoked activity is not completely understood. To investigate this relationship, we compared visual, auditory, and tactile evoked activity patterns elicited with different stimulus strengths and spontaneous activity motifs in lightly anesthetized and awake mice using mesoscale wide-field voltage-sensitive dye and glutamate imaging respectively. The characteristics of cortical activity that we compared include amplitude, speed, direction, and complexity of propagation trajectories in spontaneous and evoked activity patterns. We found that the complexity of the propagation trajectories of spontaneous activity, quantified as their fractal dimension, is higher than the one from sensory evoked responses. Moreover, the speed and direction of propagation, are modulated by the amplitude during both, spontaneous and evoked activity. Finally, we found that spontaneous activity had similar amplitude and speed when compared to evoked activity elicited with low stimulus strengths. However, this similarity gradually decreased when the strength of stimuli eliciting evoked responses increased. Altogether, these findings are consistent with the fact that even primary sensory areas receive widespread inputs from other cortical regions, and that, during rest, the cortex tends to reactivate traces of complex, multi-sensory experiences that may have occurred in a range of different behavioural contexts.

neuroscience↗

bioRxiv2016

UP-DOWN cortical dynamics reflect state transitions in a bistable balanced network

In the idling brain, neuronal circuits often exhibit transitions between periods of sustained firing (UP state) and quiescence (DOWN state). Although these dynamics occur across multiple areas and behavioral conditions, the underlying mechanisms remain unclear. Here we analyze spontaneous population activity from the somatosensory cortex of urethane-anesthetized rats. We find that UP and DOWN periods are variable (i.e. non-rhythmic) and that the population rate shows no significant decay during UP periods. We build a network model of excitatory (E) and inhibitory (I) neurons that exhibits a new bistability between a quiescent state and a balanced state of arbitrarily low rate. Fluctuating inputs trigger state transitions. Adaptation in E cells paradoxically causes a marginal decay of E-rate but a marked decay of I-rate, a signature of balanced bistability that we validate experimentally. Our findings provide evidence of a bistable balanced network that exhibits non-rhythmic state transitions when the brain rests.

neuroscience↗