The envisioned 'Wunda' API accepts input from multiple data sources (e.g. drones and wearable devices) and in multiple modalities (audio, imagery, and video) and returns an assessment about whether the data is AI-generated (a "DeepFake") or has otherwise been tampered with ("Data Poisoning").
The scientific basis of the Wunda is a set of custom, proprietary DNN classifiers that identify deepfake anomalies.
Promising initial data show that our proprietary classifiers perform certain tasks better than widely-used transformer technologies, under specific, widely-applicable conditions.
More work needs to be done to prove our hypothesis: that we are able to perform deepfake detection at the same levels of accuracy as leading deepfake-detection technologies, using fewer compute resources during training, and the same performance compute time.
If our hypothesis is correct, then we have identified a path to a market-leading* deepfake-detection technology. *(As effective, but less costly and resource-intensive, than the current market-leading deepfake detection software.)
We are seeking funding to finalize our proof of concept and bring our technology to market.
Potential clients include:
Governments, with an interest in discriminating against false information that could influence politics and national security.
Businesses whose business models rely on fraud prevention (e.g. banks and other financial institutions).
Businesses who can incorporate deepfake detection into their products to enhance user security and experience (e.g. email providers, video conference providers, and dating app companies).
