AI in the woods

Important terms


Latent features

The human eye can recognize different types of moss, or stages of their growth. But if we were asked what exactly we pay attention to when differentiating between one and the other, we might not be able to tell.

Most modern machinelearning systems are trained to learn latent features in their data. It is not always clear what the features are and how they came about, but we know they are useful to the machine to interpret the world.

Yum! Nice straw! All living beings have a way to recognise foods they can eat, using features of objects encoded by their perceptual system.

Classification is one of the main tasks performed by AI systems. When the problem involves just two classes, e.g. 'edible' vs 'inedible', we talk of 'binary classification'.

The basis of any artificial brain is the ability to use so-called features to interpret stimuli from the real world. A feature is a property that helps distinguishing between objects. So next time you're in the woods collecting leaves, ask yourself: which features would I use to tell a leaf from another?