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....
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2023-05-01
|
| Series: | Frontiers in Computational Neuroscience |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fncom.2023.1136010/full |
| _version_ | 1797821452567707648 |
|---|---|
| author | Pau Vilimelis Aceituno Pau Vilimelis Aceituno Matilde Tristany Farinha Reinhard Loidl Benjamin F. Grewe Benjamin F. Grewe |
| author_facet | Pau Vilimelis Aceituno Pau Vilimelis Aceituno Matilde Tristany Farinha Reinhard Loidl Benjamin F. Grewe Benjamin F. Grewe |
| author_sort | Pau Vilimelis Aceituno |
| collection | DOAJ |
| description | 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. Similar hierarchical structures routinely emerge in artificial neural networks (ANNs) trained for object recognition tasks, suggesting that similar structures may underlie biological neural networks. However, the classical ANN training algorithm, backpropagation, is considered biologically implausible, and thus alternative biologically plausible training methods have been developed such as Equilibrium Propagation, Deep Feedback Control, Supervised Predictive Coding, and Dendritic Error Backpropagation. Several of those models propose that local errors are calculated for each neuron by comparing apical and somatic activities. Notwithstanding, from a neuroscience perspective, it is not clear how a neuron could compare compartmental signals. Here, we propose a solution to this problem in that we let the apical feedback signal change the postsynaptic firing rate and combine this with a differential Hebbian update, a rate-based version of classical spiking time-dependent plasticity (STDP). We prove that weight updates of this form minimize two alternative loss functions that we prove to be equivalent to the error-based losses used in machine learning: the inference latency and the amount of top-down feedback necessary. Moreover, we show that the use of differential Hebbian updates works similarly well in other feedback-based deep learning frameworks such as Predictive Coding or Equilibrium Propagation. Finally, our work removes a key requirement of biologically plausible models for deep learning and proposes a learning mechanism that would explain how temporal Hebbian learning rules can implement supervised hierarchical learning. |
| first_indexed | 2024-03-13T09:52:54Z |
| format | Article |
| id | doaj.art-c9f9ddbe5c3648268f12097888aa869a |
| institution | Directory Open Access Journal |
| issn | 1662-5188 |
| language | English |
| last_indexed | 2024-03-13T09:52:54Z |
| publishDate | 2023-05-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Computational Neuroscience |
| spelling | doaj.art-c9f9ddbe5c3648268f12097888aa869a2023-05-24T06:08:56ZengFrontiers Media S.A.Frontiers in Computational Neuroscience1662-51882023-05-011710.3389/fncom.2023.11360101136010Learning cortical hierarchies with temporal Hebbian updatesPau Vilimelis Aceituno0Pau Vilimelis Aceituno1Matilde Tristany Farinha2Reinhard Loidl3Benjamin F. Grewe4Benjamin F. Grewe5Institute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, SwitzerlandETH AI Center, ETH Zurich, Zurich, SwitzerlandInstitute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, SwitzerlandInstitute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, SwitzerlandInstitute of Neuroinformatics, University of Zurich and ETH Zurich, Zurich, SwitzerlandETH AI Center, ETH Zurich, Zurich, SwitzerlandA 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. Similar hierarchical structures routinely emerge in artificial neural networks (ANNs) trained for object recognition tasks, suggesting that similar structures may underlie biological neural networks. However, the classical ANN training algorithm, backpropagation, is considered biologically implausible, and thus alternative biologically plausible training methods have been developed such as Equilibrium Propagation, Deep Feedback Control, Supervised Predictive Coding, and Dendritic Error Backpropagation. Several of those models propose that local errors are calculated for each neuron by comparing apical and somatic activities. Notwithstanding, from a neuroscience perspective, it is not clear how a neuron could compare compartmental signals. Here, we propose a solution to this problem in that we let the apical feedback signal change the postsynaptic firing rate and combine this with a differential Hebbian update, a rate-based version of classical spiking time-dependent plasticity (STDP). We prove that weight updates of this form minimize two alternative loss functions that we prove to be equivalent to the error-based losses used in machine learning: the inference latency and the amount of top-down feedback necessary. Moreover, we show that the use of differential Hebbian updates works similarly well in other feedback-based deep learning frameworks such as Predictive Coding or Equilibrium Propagation. Finally, our work removes a key requirement of biologically plausible models for deep learning and proposes a learning mechanism that would explain how temporal Hebbian learning rules can implement supervised hierarchical learning.https://www.frontiersin.org/articles/10.3389/fncom.2023.1136010/fullcortical hierarchiesdeep learningcredit assignmentsynaptic plasticitybackpropagationspiking time-dependent plasticity |
| spellingShingle | Pau Vilimelis Aceituno Pau Vilimelis Aceituno Matilde Tristany Farinha Reinhard Loidl Benjamin F. Grewe Benjamin F. Grewe Learning cortical hierarchies with temporal Hebbian updates Frontiers in Computational Neuroscience cortical hierarchies deep learning credit assignment synaptic plasticity backpropagation spiking time-dependent plasticity |
| title | Learning cortical hierarchies with temporal Hebbian updates |
| title_full | Learning cortical hierarchies with temporal Hebbian updates |
| title_fullStr | Learning cortical hierarchies with temporal Hebbian updates |
| title_full_unstemmed | Learning cortical hierarchies with temporal Hebbian updates |
| title_short | Learning cortical hierarchies with temporal Hebbian updates |
| title_sort | learning cortical hierarchies with temporal hebbian updates |
| topic | cortical hierarchies deep learning credit assignment synaptic plasticity backpropagation spiking time-dependent plasticity |
| url | https://www.frontiersin.org/articles/10.3389/fncom.2023.1136010/full |
| work_keys_str_mv | AT pauvilimelisaceituno learningcorticalhierarchieswithtemporalhebbianupdates AT pauvilimelisaceituno learningcorticalhierarchieswithtemporalhebbianupdates AT matildetristanyfarinha learningcorticalhierarchieswithtemporalhebbianupdates AT reinhardloidl learningcorticalhierarchieswithtemporalhebbianupdates AT benjaminfgrewe learningcorticalhierarchieswithtemporalhebbianupdates AT benjaminfgrewe learningcorticalhierarchieswithtemporalhebbianupdates |