GAN
Definition
Generative Adversarial Network. A framework with two neural networks, a generator that creates synthetic data and a discriminator that tries to distinguish it from real data. They train together in an adversarial loop until the generator's output is difficult to separate from genuine content.
- Full name
- Generative Adversarial Network
- Components
- Generator and discriminator networks
- Training method
- Adversarial loop
- Use in forensics
- Basis for many deepfake and synthetic media pipelines
Common questions
How does the adversarial training loop actually improve the generator's output?+
The generator produces synthetic samples and the discriminator scores them as real or fake; the generator is updated to better fool the discriminator while the discriminator is updated to catch the generator's current weaknesses, and repeating this cycle pushes the generator's output progressively closer to the real data distribution.
Why has GAN-based media detection become harder over successive model generations?+
Because the discriminator in training already learns to suppress detectable artifacts in the generator's output, later GAN versions leave fewer of the visual and statistical tells that earlier detection methods relied on, forcing forensic detectors to keep updating to newer artifact classes or model-specific fingerprints.
Related terms
- Diffusion Model
- A generative neural network architecture (Ho et al., 2020; Stable Diffusion, Rombach et al., 2022) that learns to reverse a noise-addition process...
- Encoder-Decoder
- A neural architecture where an encoder compresses an input into a compact latent representation and a decoder reconstructs an output image from...
- Latent Space
- The compressed, lower-dimensional representation of data learned by a neural network's internal layers. Generative models sample from or navigate this space to...
- NeRF (Neural Radiance Field)
- A neural representation that encodes a 3-D scene as a continuous volumetric function, allowing novel viewpoints to be rendered. In talking-head systems,...
- Voice Conversion
- A signal-processing or deep-learning technique that transforms the vocal characteristics of a source speaker's utterance to match a target speaker, while preserving...