Applying machine-learning methods directly to the compressed waveforms—without time-consuming image reconstruction—enables real-time ‘image-free’ vision-based cell sorting 1.