Backpropagation

…It develops backpropagation and off-policy Value Matching objectives, addresses trapping in intermediate metastable states, and compares committor, rate and equilibrium estimates on benchmark systems.

We introduce a statistical physics framework for learning in neural architectures composed of single or interconnected asymmetric attractor networks. These systems can exhibit a manifold of global fixed points capable of implementing sophisticated input-output mappings, which we characterize analytically. Learning from extensive datasets is achieved through the stabilization of fixed points via a fully distributed and local learning process, implemented at the single-neuron level. This simple mechanism yields performance comparable to that of conventional feedforward deep neural networks train

Backpropagation has been highly successful in training artificial neural networks for deep learning. In contrast, the brain's credit assignment mechanism remains elusive, and implementing effective credit assignment algorithms for multilayer networks in a biologically plausible manner has been challenging. Due to the locality constraint of synaptic plasticity, synaptic weights should only be updated using signals available at presynaptic and postsynaptic neurons, without accessing information from unconnected neurons. How a biological neuron encodes and transmits feedforward predictions and er

…the predominant learning algorithm, backpropagation, has long been considered biologically implausible [4]. One reason is the weight-transport problem - or the problem of how a global error signal can be accurately transmitted across the network to minimize the error resulting from each layer’s parameters. Previous solutions to the weight-transpo

Correlative Hebbian learning rules are broadly supported by experimental evidence, but do not prescribe a mechanism to coordinate plasticity across multiple circuit layers during target-directed learning. Recent experimental and theoretical work suggests that top-down input to the apical dendrites of pyramidal cells may provide a solution to this deep credit assignment problem. Experiments have shown that dendritic-spike-dependent plasticity is promoted by top-down excitatory inputs carrying information about learning targets, and it is suppressed by lateral feedback inhibition. However, exist

Humans maintain a repertoire of memories, which are continuously acquired, updated, and expressed as appropriate for the current context to meet behavioral demands. This continual learning remains a major challenge for artificial neural networks and has been suggested to require purpose-built mechanisms, such as memory replay or carefully designed learning rules. Sensorimotor learning provides a particularly useful testbed for theories of continual learning, as previous experiments have revealed a rich phenomenology of the complex and often intriguing dependence of motor adaptation on both rec

Training recurrent neural networks (RNNs) to perform neuroscience tasks can be challenging. Unlike in machine learning where any architectural modification of an RNN (e.g. GRU or LSTM) is acceptable if it facilitates training, the RNN models trained as models of brain dynamics are subject to plausibility constraints that fundamentally exclude the usual machine learning hacks. The “vanilla” RNNs commonly used in computational neuroscience find themselves plagued by ill-conditioned loss surfaces that complicate training and significantly hinder our capacity to investigate the brain dynamics unde

Training deep neural networks typically relies on backpropagating high-dimensional error signals---a computationally intensive and biologically implausible process. However, since most tasks involve low-dimensional outputs, we propose that low-dimensional error signals may suffice for effective learning. To test this hypothesis, we introduce a novel local learning rule based on Feedback Alignment that leverages indirect, low-dimensional error feedback to train large networks. Our method decouples the backward pass from the forward pass, enabling precise control over error signal dimensionality

…we analyze one such SNN trained with backpropagation based on a

Learning in neuronal circuits requires coordinated modifications of synaptic strengths across hierarchical networks to refine computation and improve behavioral outcomes. This coordination requires solving the credit assignment problem. Yet, how neuronal circuits achieve credit assignment remains a central unsolved question in systems neuroscience [1]. Several modeling studies have suggested putative biological mechanisms for back-propagating error signals through multi-layer networks. These functionally motivated models assume distinct neuronal compartments representing local error signals t

Studying and understanding the code of large neural populations hinges on learning accurate models of population activity. Statistical models that are based on sparse nonlinear Random Projections (RP) of the population proved to be highly accurate, efficient, and scalable. Interestingly, this class of models has a straightforward biologically-plausible implementation as a shallow neural circuit. Here we extend the RP models, and learn them by optimizing or ``reshaping" the random projections themselves. The reshaping of projections is akin to changing synaptic connections in the neural circuit

…machine learning algorithms, such as backpropagation [1], on difficult tasks that the brain is capable of solving. It has been shown that one of the obstacles to backpropagation-like algorithm implementation in the brain, the ’weight transport’ problem, can be overcome with initially random feedbac

…learning rules that approximate backpropagation [1-9]. Despite achieving impressive performance quantified by accuracy, these studies have not covered the breadth of solution characteristics found by these rules. In this work, we leverage established theoretical tools from deep learning to investigate the robustness of solutions, and gain insights into generalization properti

Gradient-based optimization powered by the backpropagation algorithm proved to be the pivotal method in the training of non-spiking artificial neural networks. At the same time, spiking neural networks hold the promise for efficient processing of real-world sensory data by communicating using discrete events in continuous time. We derive the backpropagation algorithm for a recurrent network of spiking (leaky integrate-and-fire) neurons with hard thresholds and show that the backward dynamics amount to an event-based backpropagation of errors through time. Our derivation uses the jump condition