Recall performance in CA3.

Left: Each panel on the left column shows an example of the retrieval quality in a CA3 model as labelled on the right hand side of the figure, when recall cue with cue quality Ccue = 0.4 is provided to EC. The horizontal axes represent the position of the pattern in the sequence. Each colored line shows the correlation between the retrieved and the correct patterns in a sequence. The RCN (first row) is highly sensitive to noise, whereas the LCN (last row) seems to keep the same input information while retrieving the stored sequence. DDN(α ≤ 0.9) shows the best performance and retrieves the previously stored sequence from the noisy input cue. Right: Summary of recall performance in CA3 for the entire range of noise levels. The pattern completion plot for different network models. Data-points above the diagonal indicate good sequence completion, whereas data below the diagonal indicate that information in the input is lost. Data on the diagonal show that the network maintains the information in the input cue along the sequence. Overall, DDN(α ≤ 0.9) performs best, even with highly corrupted cues.