{
"title": "Network-level performance gains from adding astrocyte dynamics to neuron-only baselines",
"data_points": [
{
"doi": "10.1155/2012/476324",
"study": "Porto-Pazos / Alvarellos-González et al. 2012",
"value": "all six astrocyte-inspired algorithms improved validation accuracy over the ANN; benefit scaled with network complexity",
"method": "Six neuron-glia algorithms compared against a plain ANN baseline; 50 independent runs; validation accuracy and MSE as metrics",
"metric": "classification efficacy (ANGN vs. ANN)",
"study_system": "Multi-layer artificial neural networks (ANN) with artificial astrocyte elements (ANGN) trained on a 6-bit multiplexer and Iris classification",
"value_source_section": "5. Discussion and Conclusions",
"value_source_sentence": "ANGN implemented using these six algorithms (thus including artificial astrocytes that simulate the potentiation of the connections and penalize the lack of activity) improved the ANN that did not include artificial astrocytes."
},
{
"doi": "10.1371/journal.pone.0029445",
"study": "Wade et al. 2011",
"value": "firing rate of neuron N4 synchronised to SIC for t < 115 s; phase-synchronisation frequency ~0.1 Hz in AM mode",
"method": "Four-neuron clusters N1–N4 with SIC-driven postsynaptic depolarisation; firing-rate synchronisation measured over 0–200 s",
"metric": "firing-rate synchronisation between remote neurons driven by astrocytic SIC",
"study_system": "Bidirectionally coupled astrocyte-neuron (AN) cell-level model with STDP learning and extrasynaptic NMDAR-mediated slow inward currents (SICs)",
"value_source_section": "Spatially distributed learning signals",
"value_source_sentence": "The firing rate of N4 is synchronized to SIC below 115 s but thereafter the firing rate of N4 increases due to presynaptic activity alone, as shown also in Figure 6E."
},
{
"doi": "10.3389/fncel.2021.631485",
"study": "Gordleeva et al. 2021",
"value": "≤ 2 s delay of astrocytic Ca²⁺ elevation following synchronous neuronal discharge",
"method": "Two-dimensional image patterns encoded as spiking inputs; astrocyte Ca²⁺ threshold [Ca²⁺]thr activates gliotransmitter feedback; retention measured by recall firing-rate fidelity",
"metric": "astrocytic Ca²⁺ activation delay (working-memory timescale)",
"study_system": "Spiking Izhikevich neuron network with astrocytic Ca²⁺ gating of synaptic strength; working-memory retrieval task (1-item and multi-item)",
"value_source_section": "4.1. Single-Item WM",
"value_source_sentence": "In accordance with the experimental data (Bindocci et al., 2017), we tuned the model parameters in such a way that the onset of calcium elevation in the astrocytes induced by synchronous neuronal discharge had a delay of ≤ 2 s."
},
{
"doi": "10.1016/j.isci.2023.108241",
"study": "Naghieh et al. 2023",
"value": "~20 % improvement in WM performance; SR maintains function up to 60 % random synaptic damage",
"method": "Four damage-mode experiments; SR enabled vs. disabled; metrics: recall accuracy, firing-frequency restoration, robustness thresholds",
"metric": "working-memory robustness improvement from astrocyte self-repair",
"study_system": "Spiking neuronal network with astrocyte self-repair (SR) mechanism tested under random and pattern-specific synaptic damage",
"value_source_section": "Discussion",
"value_source_sentence": "Simulation results suggested that the addition of the SR mechanism increases the resilience of the network up to 60% synaptic impairment, which is a 20% improvement compared to the same network without SR capability."
},
{
"doi": "10.3389/fnins.2020.603796",
"study": "Rastogi et al. 2020",
"value": "+9 % (MNIST) and +5 % (F-MNIST) test accuracy after A-STDP-based repair",
"method": "Synaptic-fault injection (50-90%) then re-learning with/without A-STDP; test classification accuracy as function of epochs",
"metric": "classification accuracy recovery by A-STDP in 90%-faulty network",
"study_system": "Unsupervised spiking neural network trained on MNIST and F-MNIST with astrocyte-modulated STDP (A-STDP) for fault tolerance",
"value_source_section": "3. Results",
"value_source_sentence": "Interestingly, A-STDP is able to repair faults even in a 90% faulty network and improve the testing accuracy by almost 9% (5%) for the MNIST (F-MNIST) dataset."
},
{
"doi": "10.3390/e25050745",
"study": "Stasenko et al. 2023",
"value": "astrocytic modulation prevents stimulation-induced hyperexcitation and restores image representation lost in the raster diagram",
"method": "Astrocyte modulation on vs. off; analysis of raster-diagram image recovery and spike regularity",
"metric": "prevention of stimulation-induced hyperexcitation and bursting, enabling image recovery",
"study_system": "Spiking neural network (excitatory + inhibitory) with astrocytic slow modulation of synaptic strength; 2-D image stimulation patterns",
"value_source_section": "Abstract",
"value_source_sentence": "We found that astrocytic modulation prevented stimulation-induced SNN hyperexcitation and non-periodic bursting activity."
}
],
"description": "Six independent computational studies that compared an identical neuronal architecture with and without astrocyte coupling on a cognitive or learning benchmark. Across all six, adding astrocyte Ca²⁺ / gliotransmitter dynamics improved task-relevant metrics. Verbatim quotes are taken from each paper's Results or Discussion.",
"comparison_id": "astrocyte_contribution_to_network_performance",
"source_papers": [
"10.1155/2012/476324",
"10.1371/journal.pone.0029445",
"10.3389/fncel.2021.631485",
"10.1016/j.isci.2023.108241",
"10.3389/fnins.2020.603796",
"10.3390/e25050745"
],
"integrative_claim": "Across six architectures — feed-forward ANNs (Porto-Pazos 2012), bidirectionally coupled AN models (Wade 2011), Izhikevich working-memory SNNs (Gordleeva 2021), fault-tolerant SNNs with self-repair (Naghieh 2023; Rastogi 2020), and information-encoding SNNs (Stasenko 2023) — adding astrocyte Ca²⁺ / gliotransmitter dynamics consistently improves task-relevant performance. The common functional contribution is an additional slow (~0.1–10 s) timescale that extends the network's time-scale repertoire beyond purely neuronal (~ms) dynamics."
}