Engineers have demonstrated a simple computational approach for improving the way artificial intelligence classifiers, such as neural networks, can be trained based on limited amounts of sensor data.