trivialaugment

**TrivialAugment** is an **extremely simple augmentation strategy that applies a single randomly selected augmentation with a random magnitude to each image** — with zero hyperparameters, yet matching or outperforming RandAugment and AutoAugment. **How Does TrivialAugment Work?** - **Sample**: Pick one augmentation uniformly at random from the pool. - **Magnitude**: Sample a random magnitude uniformly from the valid range. - **Apply**: Apply the single augmentation to the image. - **That's It**: No $N$, no $M$, no policy, no search. Zero hyperparameters. - **Paper**: Müller & Hutter (2021). **Why It Matters** - **Zero Hyperparameters**: The simplest possible automated augmentation — no tuning at all. - **Competitive**: Matches or exceeds RandAugment and AutoAugment on ImageNet, CIFAR-10, CIFAR-100. - **Lesson**: Over-engineering augmentation policies may not be necessary — randomness works. **TrivialAugment** is **the laziest augmentation strategy that works** — randomly applying one augmentation at random strength, yet matching sophisticated learned policies.