Extended Data Fig. 7: Performance of CNNs on classification and regression tasks.
From: Engineered bacterial swarm patterns as spatial records of environmental inputs
a. Training and validation accuracy and loss for a CNN model which had three convolutional/max pooling blocks, trained on the dataset of images of the dual-input strain at various IPTG and arabinose conditions (same dataset as in Fig. 4f). b. Fine-tuning of three architectures pretrained on ImageNet; righthand panels represent models’ ability to identify the correct image class within its top three predicted classes. c. Learning curves of EfficientNet model trained on the dual-input pattern images with regression output. Loss = mean squared error; mean absolute error shown for further detail. d. Predictions of the trained model evaluated on images not seen in the original training dataset. Dotted lines represent location where predictions would match the true values. Error bars represent root mean squared error on the predictions for each given concentration.
