📄 中文摘要
研究人员展示了一种简单的最近邻分类方法，能够在仅仅看到几张照片后识别新对象，而无需经过数周的训练。通过对数据准备方式进行小幅调整，如特征的中心化和缩放，这种传统方法在常见测试中往往能取得更好的结果。该方法在多个设置中表现出意外的有效性，在某些情况下超越了早期的复杂方法。因此，对于希望实现快速学习的机器，少量示例的学习（即少样本学习）可能不需要复杂的技巧。
SimpleShot：通过微小调整让计算机从少量图片中学习新事物
📄 English Summary
SimpleShot: Revisiting Nearest-Neighbor Classification for Few-Shot Learning
Researchers have demonstrated that a simple nearest-neighbor classification method can identify new objects after seeing only a few photos, eliminating the need for weeks of training. By making minor adjustments to data preparation, such as centering and scaling features, this traditional approach often yields better results on common tests. It has proven surprisingly effective across various setups, outperforming earlier complex methods in some cases. Thus, for machines aiming to learn quickly from minimal examples—known as few-shot learning—complex tricks may not be necessary.