Sequence learning via biophysically realistic learning rules (Cone and Shouval 2021)

 Download zip file 
Help downloading and running models
Accession:266774
This work proposes a substrate for learned sequential representations, via a network model that can robustly learn and recall discrete sequences of variable order and duration. The model consists of a network of spiking leaky-integrate-and-fire model neurons placed in a modular architecture designed to resemble cortical microcolumns. Learning is performed via a biophysically realistic learning rule based on “eligibility traces”, which hold a history of synaptic activity before being converted into changes in synaptic strength upon neuromodulator activation. Before training, the network responds to incoming stimuli, and contains no memory of any particular sequence. After training, presentation of only the first element in that sequence is sufficient for the network to recall an entire learned representation of the sequence. An extended version of the model also demonstrates the ability to successfully learn and recall non-Markovian sequences.
Reference:
1 . Cone I, Shouval HZ (2021) Learning precise spatiotemporal sequences via biophysically realistic learning rules in a modular, spiking network. Elife [PubMed]
Model Information (Click on a link to find other models with that property)
Model Type:
Brain Region(s)/Organism: Visual cortex;
Cell Type(s): Abstract integrate-and-fire leaky neuron;
Channel(s):
Gap Junctions:
Receptor(s):
Gene(s):
Transmitter(s):
Simulation Environment: MATLAB;
Model Concept(s): Sequence learning; Eligibility traces;
Implementer(s): Cone, Ian [iancone at rice dot edu]; 
March 18, 2021: Initial code upload (Markovian FF and Non-Markovian)

January 24, 2023: -Changes made to Markovian code: (upload of Markovian A2A)
		*Original Markovian code (Markovian FF) incorrectly restricted Messenger to Timer intercolumnar 
		connections were such that they were limited to be feed-forward (i.e. ordinal to the presented sequence)
		*Updated Markovian code (Markovian A2A) correctly allows for any (all-to-all or A2A) Messenger to Timer intercolumnar
		connections. To allow for this, the following changes were made:
			*Feed-forward threshold changed from 20Hz to 30Hz
			*T^d,ff_max changed from .00345 to .0045 
			*eta^p_ff changed from 20 x 3500 ms^-1 to 8.8 x 3500 ms^-1
			*eta^d_ff changed from 15 x 3500 ms^-1 to 10 x 3500 ms^-1
			*eta_ff changed from 0.25 to 0.4
		*Supplementary Table 1 has been updated include values for both Markovian FF and Markovian A2A
	      -Updates made to Supplementary Tables:
		*Original Supplementary Tables contained incorrectly scaled or incorrectly stated parameters.
		Please see[1] for more details. The Supplementary Tables included with this upload have been 
		updated to rectify these errors.

The above errors were found during a replication study (Zajzon et al. 2023[1]), which also proposed the solution of 
raising the feed-forward threshold (implemented in Markovian A2A). We thank the authors of that study for their contributions to these changes.

Additionally, Zajzon et al. 2023[1] produced a well-documented and effecient implementation of this model in NEST, so we encourage those 
interested in using the model for their own purposes to also consider their implementation as a potential code base[2].


1.Zajzon, B., Duarte, R. & Morrison, A. Towards reproducible models of sequence learning: replication and analysis of a modular spiking network with reward-based learning. bioRxiv 2023.01.18.524604 (2023) doi:10.1101/2023.01.18.524604.
2.Zajzon, B. Towards reproducible models of sequence learning: replication and analysis of a modular spiking network with reward-based learning. https://github.com/zbarni/re_modular_seqlearn (2022).

Loading data, please wait...