In recent years, machine-learning has taken the world by storm and is at the heart of many breakthroughs across a wide-variety of fields from computer vision, to biology, to autonomous vehicles, to robotics. In contrast, the application of deep-learning to cryptanalysis is still nascent with only a handful of successful research in the space. This begs the question whether or not deep-learning is bound to revolutionize cryptanalysis or if it’s just a fashion that will go out of style in a few years?
To start answering this question, this keynote looks at which cryptanalysis use cases are a good fit for deep-learning and those that are unlikely to be successful. Then it reflects on how some of the cutting edge deep-learning techniques, such as self-supervised learning open the door to tackle use cases that were previously out of reach. Finally it discusses which deep-learning techniques and tooling beyond training neural networks to attack crypto can creatively be applied to cryptanalysis to push the boundary even further.
