Deep fake.pptxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

project

1. DEEP FAKE B y: Riwaz silwal Pawan bhattarai Sampada shrestha Slide 1 of 8

2. What are DeepFakes ? The phenomenon gained its name from a user of the platform Reddit, who went by the name “deepfakes” (deep learning + fakes). This person shared the first deepfakes by placing unknowing celebrities into adult video clips. This triggered widespread interest in the Reddit community and led to an explosion of fake content. The first targets of deepfakes were famous people, including actors (e.g., Emma Watson and Scarlett Johansson), singers (e.g., Katy Perry) and politicians (e.g., President Obama) Slide 2 of 8

3. How to deepfakes work? Deepfakes are commonly created using a specific deep network architecture known as autoencoder. Autoencoders are trained to recognize key characteristics of an input image to subsequently recreate it as their output. In this process, the network performs heavy data compression. Autoencoders consist of three subparts: - an encoder (recognizing key features of an input face) - a latent space (representing the face as a compressed version) - a decoder (reconstructing the input image with all detail) Slide 3 of 8

4. What is autoencoder and how is it used? An autoencoder is a type of neural network used for unsupervised learning. It consists of an encoder and a decoder, and its primary purpose is to learn an efficient representation (latent space) of the input data. The encoder compresses the input data into a lower-dimensional space, and the decoder reconstructs the input data from this compressed representation. In the context of creating deepfakes, which involve generating realistic-looking images or videos of one person's face onto another person's body, the use of two separate autoencoders is not efficient. This is because each autoencoder, when trained independently on different people, would learn unique features and representations specific to the individuals it was trained on. These representations are likely to be incompatible with each other, making it challenging to seamlessly combine them for the purpose of generating deepfakes. Slide 4 of 8

5. The trick Training Individual Autoencoders: Train an autoencoder for each person separately. Each autoencoder consists of an encoder and a decoder. The encoder in each autoencoder is responsible for compressing the facial features of the respective person into a latent space. Sharing Encoder Architecture: Design the encoder part of both autoencoders to have a similar architecture. This could involve using the same neural network structure or ensuring that the dimensions of the latent space are compatible. Creating Latent Space Representation: Use the encoder from the first person's autoencoder to encode an image of that person's face. This results in a compressed latent space representation. Generating Fake Image: Take the latent space representation obtained from the first person's encoder and input it into the decoder of the second person's autoencoder. Slide 5 of 8

6. The trick The shared-latent space assumption. The two heterogeneous images of x 1 and x 2 can be mapped into the same latent representation z by a coupling VAE for comparability, while the latent representations can be reconstructed into the original images, respectively, for completeness. Slide 6 of 8

7. Popular example we all have seen Popular Indian actress Rashmika Mandanna’s deepfake that went viral on social media not too long ago. Slide 7 of 8

8. Takeaway Deepfakes can be used in positive and negative ways to manipulate content for media, entertainment, marketing, and education. Deepfakes are not magic but are produced using techniques from Al that can generate fake content that is highly believable. Slide 8 of 8