Biologically-inspired neuronal adaptation improves learning in neural networks.

Since humans still outperform artificial neural networks on many tasks, drawing inspiration from the brain may help to improve current machine learning algorithms. Contrastive Hebbian learning (CHL) and equilibrium propagation (EP) are biologically plausible algorithms that update weights using only local information (without explicitly calculating gradients) and still achieve performance comparable to conventional backpropagation. In this study, we augmented CHL and EP with Adjusted Adaptation, inspired by the adaptation effect observed in neurons, in which a neuron's response to a given stimulus is adjusted after a short time. We add this adaptation feature to multilayer perceptrons and convolutional neural networks trained on MNIST and CIFAR-10. Surprisingly, adaptation improved the performance of these networks. We discuss the biological inspiration for this idea and investigate why Neuronal Adaptation could be an important brain mechanism to improve the stability and accuracy of learning.

Combining backpropagation with Equilibrium Propagation to improve an Actor-Critic reinforcement learning framework.

Backpropagation (BP) 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 BP 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 BP.

Neurons learn by predicting future activity.

Understanding how the brain learns may lead to machines with human-like intellectual capacities. It was previously proposed that the brain may operate on the principle of predictive coding. However, it is still not well understood how a predictive system could be implemented in the brain. Here we demonstrate that the ability of a single neuron to predict its future activity may provide an effective learning mechanism. 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. We show how this mathematically derived learning rule can provide a theoretical connection between diverse types of brain-inspired algorithms, thus, offering a step toward development of a general theory of neuronal learning. We tested this predictive learning rule in neural network simulations and in data recorded from awake animals. Our results also suggest that spontaneous brain activity provides "training data" for neurons to learn to predict cortical dynamics. Thus, the ability of a single neuron to minimize surprise: i.e. the difference between actual and expected activity, could be an important missing element to understand computation in the brain.

Predictive Neuronal Adaptation as a Basis for Consciousness.

Being able to correctly predict the future and to adjust own actions accordingly can offer a great survival advantage. In fact, this could be the main reason why brains evolved. Consciousness, the most mysterious feature of brain activity, also seems to be related to predicting the future and detecting surprise: a mismatch between actual and predicted situation. Similarly at a single neuron level, predicting future activity and adapting synaptic inputs accordingly was shown to be the best strategy to maximize the metabolic energy for a neuron. Following on these ideas, here we examined if surprise minimization by single neurons could be a basis for consciousness. First, we showed in simulations that as a neural network learns a new task, then the surprise within neurons (defined as the difference between actual and expected activity) changes similarly to the consciousness of skills in humans. Moreover, implementing adaptation of neuronal activity to minimize surprise at fast time scales (tens of milliseconds) resulted in improved network performance. This improvement is likely because adapting activity based on the internal predictive model allows each neuron to make a more "educated" response to stimuli. Based on those results, we propose that the neuronal predictive adaptation to minimize surprise could be a basic building block of conscious processing. Such adaptation allows neurons to exchange information about own predictions and thus to build more complex predictive models. To be precise, we provide an equation to quantify consciousness as the amount of surprise minus the size of the adaptation error. Since neuronal adaptation can be studied experimentally, this can allow testing directly our hypothesis. Specifically, we postulate that any substance affecting neuronal adaptation will also affect consciousness. Interestingly, our predictive adaptation hypothesis is consistent with multiple ideas presented previously in diverse theories of consciousness, such as global workspace theory, integrated information, attention schema theory, and predictive processing framework. In summary, we present a theoretical, computational, and experimental support for the hypothesis that neuronal adaptation is a possible biological mechanism of conscious processing, and we discuss how this could provide a step toward a unified theory of consciousness.