From blickets to synapses: inferring temporal causal networks by observation

Area of Science:

• Computational Neuroscience
• Developmental Psychology
• Cognitive Science

Background:

• Human infants rapidly learn causal relationships from limited observations.
• Existing computational models offer explanations for infant causal inference.
• Understanding the mechanisms of early causal learning is crucial for developmental science.

Purpose of the Study:

• To propose a novel spiking neuronal network (SNN) model for learning temporal and causal event dependencies.
• To demonstrate how biologically realistic synaptic plasticity rules can explain cognitive causal assumptions.
• To present a mechanism for transmitting neural circuit functionality via synaptic pattern copying.

Main Methods:

• Developed a spiking neuronal network implementing spike-time dependent plasticity, long-term depression, and heterosynaptic competition.
• Incorporated transmission delays within the network to model temporal relationships.
• Utilized the network as an emulator for state inference and tested its ability to replicate known cognitive biases.

Main Results:

• The SNN successfully learned a dynamical model of event relationships, mimicking Rescorla-Wagner-like learning.
• Biologically plausible synaptic plasticity rules accounted for behavioral data on backwards blocking and screening-off phenomena.
• The model demonstrated the capacity to copy synaptic connectivity patterns from observed spontaneous neural activity.

Conclusions:

• Spiking neuronal networks with biologically realistic plasticity can model infant causal learning.
• This framework provides a unified explanation for temporal, causal, and cognitive aspects of event learning.
• The proposed mechanism offers a novel approach for understanding and potentially transferring neural circuit functionality.