Translation invariance property

1. What is meant by translational invariance?
2. What is translation invariance in AI?
3. How do we achieve translation invariance?
4. Where is translation invariance used?

What is meant by translational invariance?

Translational invariance implies that, at least in one direction, the object is infinite: for any given point p, the set of points with the same properties due to the translational symmetry form the infinite discrete set p + na | n ∈ Z = p + Z a.

What is translation invariance in AI?

Translation invariance means that the system produces exactly the same response, regardless of how its input is shifted. For example, a face-detector might report "FACE FOUND" for all three images in the top row.

How do we achieve translation invariance?

Also, translational invariance can be achieved by applying a pooling operation, meaning that a region or object can become invariant to (small) translations after the application of pooling. Translational invariance is a highly desirable property in many tasks such as object recognition and audio recognition.

Where is translation invariance used?

As we replace the output with the max in case of max-pooling, hence even if we change the input slightly, it won't affect the values of most of the pooled outputs. Translational Invariance is a useful property where the exact location of the object is not required.