Loidl, R. | Search bioRxiv

bioRxiv2024

Hierarchical Target Learning in the Mammalian Neocortex: A Pyramidal Neuron Perspective

1Computational neuroscience currently debates two competing hypotheses to explain hierarchical learning in the neocortex: deep learning inspired approximations of the backpropagation algorithm, where neurons adjust synapses to minimize an error, and target learning algorithms, where neurons learn by reducing the feedback needed to achieve a desired target activity. While both hypotheses have been supported by theoretical studies, there is currently no empirical test that compares them directly. Here we provide such tests in the mouse neocortex by evaluating both hypotheses against experimental data at the single cell level and at the population level. At the single cell level, we conduct in vitro experiments that clarify the relationship between algorithmic learning signals and synaptic plasticity within individual pyramidal neurons. At the population level, we analyze in vivo calcium imaging data in the lateral visual cortex. By combining in vivo and in vitro data we reveal a critical discrepancy between neocortical hierarchical learning and canonical machine learning.

neuroscience↗

bioRxiv2023

Learning Cortical Hierarchies with Temporal Hebbian Updates.

A key driver of mammalian intelligence is the ability to represent incoming sensory information across multiple abstraction levels. For example, in the visual ventral stream, incoming signals are first represented as low-level edge filters and then transformed into high-level object representations. These same hierarchical structures routinely emerge in artificial neural networks (ANNs) trained for image/object recognition tasks, suggesting that a similar process might underlie biological neural networks. However, the classical ANN training algorithm, backpropagation, is considered biologically implausible, and thus several alternative biologically plausible methods have been developed. For instance, several cortical-inspired ANNs in which the apical dendrite of a pyramidal neuron encodes top-down prediction signals have been proposed. In this case, akin to theories of predictive coding, a prediction error can be calculated locally inside each neuron for updating its incoming weights. Notwithstanding, from a neuroscience perspective, it is unclear whether neurons could compare their apical vs. somatic spiking activities to compute prediction errors. Here, we propose a solution to this problem by adapting the framework of the apical-somatic prediction error to the temporal domain. In particular, we show that if the apical feedback signal changes the postsynaptic firing rate, we can use differential Hebbian updates, a rate-based version of the classical spiking time-dependent plasticity (STDP) updates. To the best of our knowledge, this is the first time a cortical-like deep ANN has been trained using such time-based learning rules. Overall, our work removes a key requirement of biologically plausible models for deep learning that does not align with plasticity rules observed in biology and proposes a learning mechanism that would explain how the timing of neuronal activity can allow supervised hierarchical learning.

neuroscience↗