Atroshckenko et al. investigate the use of deep neural networks to automatically recognize surgical actions and assess surgeon skill levels according to video recordings of robotic cardiac surgery simulations. 435 recordings from 19 surgeons performing three porcine wet lab tasks. A hybrid network architecture combining convolutional neural networks was applied to analyze temporal information. Researchers found that the action recognition network had an accuracy of 98% in distinguishing suturing and dissection. The skill assessment network showed an accuracy of 56%. Visualization analyses confirmed that the algorithm focused on features such as needles, sutures, and tissue regions. While action recognition is highly reliable, the skill assessment component would benefit from larger data sets.