A tomograph is like a photographer who walks around an object and takes pictures from different sides. Usually, it snaps at regular intervals and uses the same amount of light for each shot. But if you need to shoot quickly and not waste extra film, you have to choose: which angles matter more and where to direct more light. That is exactly the problem the authors of the paper solve for X-ray computed tomography.

Instead of distributing the dose evenly, they propose reinforcement learning — a method where the program tries different options and remembers what works. The scanner learns from its mistakes: it first picks angles at random, then figures out that some directions reveal more details, and sends more X-ray photons there. This is especially useful when you have few images and need to save dose.

The authors tested the method in numerical experiments and compared it with the standard strategy. It turned out that the new approach produces sharper images and detects defects better, especially under tight constraints. Thus, the tomograph is no longer just a camera but becomes a smart observer that knows where to look.