With Deepfake technology contributing to propaganda, identify fraud and sexual harassment, more work needs do be done to identify what videos are real and fake. Thankfully, computer scientists at UC Riverside are working on that very tech, and they've already been able to identify altered faces in deepfake footage through a deep neural network.
Riverside's technology is still new, but the promising tech is already able to identify 99% of manipulated videos. We don’t know how advanced this can get or if it will eventually be sold to security systems, but this is a great step forward. Anything that can protect the public from cybercriminals is always a good thing.
Uncovering deepfakes with Expression Manipulation Detection
The aforementioned neural network is called the Expression Manipulation Detection, or EMD. According to the UC Riverside scientists, EMD will divide the tasks into two components. One branch focuses on everything about facial expressions while the other branch is an encoder-decoder that’s responsible for manipulation detection and localization.
“Multi-task learning can leverage prominent features learned by facial expression recognition systems to benefit the training of conventional manipulation detection systems. Such an approach achieves impressive performance in facial expression manipulation detection.”
Amit Roy-Chowdhury via UC Riverside News
It’s a fairly unique system and one that can be revolutionary once these scientists can perfect the technology. Being able to program an AI through a neural network that can identify these deepfakes crimes is something that can only benefit humanity.
Read More: US Army planning $22 billion spend on Microsoft HoloLens glasses for soldiers
The EMD still has some kinks
Despite some of the advanced processing, the technology is fairly new and has proven to be a bit troublesome. It’s the downside of unproven technology so it looks like the EMD still has some kinks to work out.
“What makes the deepfake research area more challenging is the competition between the creation and detection and prevention of deepfakes which will become increasingly fierce in the future. With more advances in generative models, deepfakes will be easier to synthesize and harder to distinguish from real.”
Amit Roy-Chowdhury via UC Riverside News
Hopefully, the folks at UC Riverside will be able to master the technology and get this to the public. Identity fraud is still a scam that many people go through and technology can help out a lot in fighting this.