This project explores a combined architecture with associative memory, predictive-coding components, and attention-based connections.
The experiments examine individual components and the behavior of the combined model. The development notes identify difficulties with stable inference, memory retrieval, and adaptation between pathways.
Some recorded assessments disagree about how well the components work. Several comparisons also show the combined architecture underperforming a simpler baseline. The useful next step is to resolve those comparisons and isolate the contribution of each component.
The project remains exploratory. A trainable combination of these mechanisms is not, by itself, evidence that the combination is useful.