Mathematical and connectionist models of human memory: a comparison

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Psychology

Background:

• Convolution-based models explain human memory data but require predefined mathematical operations.
• Connectionist models face challenges with single-trial learning and exhibit catastrophic interference in multiple list learning.

Purpose of the Study:

• To compare the efficacy of convolution-based models and DARNET in accounting for human memory data.
• To evaluate DARNET's ability to model single-trial learning and associative recall.

Main Methods:

• Utilized DARNET, a connectionist model, to learn associations between vectors into memory traces.
• Compared DARNET's performance against established convolution-based models using similar higher-level architectures.

Main Results:

• DARNET successfully accounts for a wide range of human memory data.
• DARNET demonstrates comparable performance to convolution-based models in explaining experimental data.
• DARNET effectively models single-trial learning and associative recall.

Conclusions:

• DARNET offers a viable connectionist alternative for modeling human memory, particularly for single-trial learning.
• The associative mechanism in DARNET matches the explanatory power of convolution-based models.
• Further investigation into model limitations is warranted.