(a) Parametric plot of learning time vs energy while the consolidation threshold is varied. The threshold value runs from to 10 in steps of 0.5. For small maintenance costs, the threshold determines a trade-off between either a short learning time or a low energy (e.g. black curve). At higher maintenance costs, the most energy efficient threshold also leads to a short learning time. Average over 100 runs; parameter: . (b) Similar to the perceptron results in panel a, the effects of consolidation threshold on energy cost and learning time for training in a multi-layer network vary depending on the maintenance cost . Here, the threshold starts at 0.005 and is in increments of 0.005. When (black dots, each representing a unique consolidation threshold), there is a trade-off between shorter learning time and lower energy cost. When (red dots), the result is similar to the perceptron result with , where optimizing learning time or energy cost leads to a similar threshold. Parameters: , , required accuracy .