Computational modeling of pair-association memory in inferior temporal cortex.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Cognitive Science

Background:

• The inferior temporal (IT) cortex in monkeys exhibits unique neuronal activities during visual stimulus-association tasks.
• Existing neural network models do not fully account for these observed IT cortex activities, highlighting a gap in understanding brain memory mechanisms.

Purpose of the Study:

• To elucidate the computational principles underlying visual stimulus-stimulus association in the brain.
• To construct a biologically plausible neural network model capable of learning and performing delayed pair-association tasks.
• To explain the distinctive neuronal activities observed in the IT cortex.

Main Methods:

• Developed a two-network model (N1: association network, N2: trainer network) to simulate pair-association memory formation.
• N2 receives N1 output and external input, providing a learning signal to guide N1's state transitions.
• Computer simulations were used to test the model's performance on a delayed pair-association task.

Main Results:

• The model successfully learned and performed the delayed pair-association task, distinguishing the correct target.
• The model's simulated neuronal activity closely matched empirical data from the IT cortex.
• A trajectory attractor was formed in N1, connecting cue-coding to target-coding states, explaining memory trace formation.

Conclusions:

• The proposed model provides a viable computational principle for visual association memory.
• It is hypothesized that N1 corresponds to area TE and N2 to the rhinal cortex, offering insights into brain structures involved in learning and memory.
• The model explains existing physiological findings and generates testable predictions for future research on learning and memory.