Combating Deepfakes: Multi-LSTM and Blockchain as Proof of Authenticity for Digital Media
| Autor: | Steven Delaney, Christopher Chun Ki Chan, Vimal Kumar, Munkhjargal Gochoo |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Closed captioning
0209 industrial biotechnology Information retrieval Smart contract business.industry Computer science Digital content 02 engineering and technology Digital media Metadata 020901 industrial engineering & automation 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing The Internet Misinformation business |
| Zdroj: | 2020 IEEE / ITU International Conference on Artificial Intelligence for Good (AI4G). |
| DOI: | 10.1109/ai4g50087.2020.9311067 |
| Popis: | Malicious use of deep learning algorithms has allowed the proliferation of high realism fake digital content such as text, images, and videos, to exist on the internet as readily available and accessible consumable content. False information provided through algorithmically modified footage, images, audios, and videos (known as deepfakes), coupled with the virality of social networks, may cause major social unrest. The emergence of misinformation from fabricated digital content suggests the necessity for anti-disinformation methods such as deepfake detection algorithms or immutable metadata in order to verify the validity of digital content. Permissioned blockchain, notably Hyperledger Fabric 2.0, coupled with LSTMs for audio/video/descriptive captioning is a step towards providing a feasible tool for combating deepfake media. Original content would require the original artist attestation of untampered data. The smart contract combines a varied multiple LSTM networks into a process that allows for the tracing and tracking of a digital content's historical provenance. The result is a theoretical framework that enables proof of authenticity (PoA) for digital media using a decentralized blockchain using multiple LSTMs as a deep encoder for creating unique discriminative features; which is then compressed and hashed into a transaction. Our work assumes we trust the video at the point of reception. Our contribution is a decentralized blockchain framework of deep discriminative digital media to combat deepfakes. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|