| Term | Definition |
|---|---|
| Generative Adversarial Network (GAN) | The system consists of a competitive setup between two neural networks, with one network generating counterfeit data and the other attempting to recognize the counterfeit. Over time, the system develops its ability to generate real data, such as highly convincing pictures or audio. |
| Gradient Clipping | A mechanism used during training to keep changes in the model's weights from becoming too large, hence stabilizing the learning process and preventing the model from crashing. |
| Gradient Descent | A method used in machine learning to improve a model by slowly adjusting its settings to reduce errors until it reaches an optimum outcome. |