Everything started with a paper on adversarial machine learning. During the period that Hector was working at UCL, he learnt about different applications of machine learning for malware detection. The machine learning systems that he designed used to reach very good accuracy results.
in 2018, he discovered a paper published in Usenix Security that was proving how adversarial machine learning can defeat a malware detection system based on ML. This tool, called EvadeML, opened a new research area by changing aspects of the evaluation process. This mainly included adversarial scenarios.
Work after work, the idea of MLighter was starting to grow in Hector’s mind, including multiple discussions with researchers that are currently working on machine learning testing. His work on testing also helped to elaborate on how a machine learning system should be tested from different dimensions. That’s the reason MLighter is divided into model and code blind spots, it is important to consider both when we are performing the testing.
Then, a funding opportunity came from the CyberASAP program. This led to putting everything together and preparing a real product on this idea. With the help of the Knowledge Transfer Network and the training provided by CyberASAP, MLighter was taking form after its market research and validation process.
With a strong development effort and after passing multiple phases of the program, MLighter is ready for the demo day where we will show the results of all these years of research. If you want to attend, you have all the information at the following link: