Not all solutions are created equal: An analytical dissociation of functional and representational similarity in deep linear neural networks
Why two neural networks can work the same way but think differently
Two neural networks can solve the same task using completely different internal representations—or use similar representations while computing things in fundamentally different ways. The researchers found that representational similarity (shared internal codes) and functional similarity (solving the same problem) are independent properties, and that networks built to withstand weight changes are actually forced to use task-specific, less flexible representations.
When neuroscientists compare brain activity patterns across animals or people, they assume similar patterns mean similar computations. This work shows that assumption breaks down—networks can have aligned representations without aligned function. The findings change how researchers should interpret comparisons of neural codes in brains and artificial networks, suggesting they need to measure function directly rather than assuming it from representation alone.