Tversky Neural Networks

Post by c32c33429009ed6 » Sun, Aug 17, 2025, 12:52 PM UTC

sdenton4 wrote: post #10

abeppu wrote: post #7

Earlier quoted context omitted.

I think the case for interpretability could have been made better, but in Figure 3 I think if you look at the middle "prototype" rows from the traditional vs Tversky layers, and scroll so you can't see the rows above, I think you could pick out mostly which Tversky prototype corresponds to each digit, but not which traditional/linear prototype corresponds to each digit. So I do think that's more interpretable in two…

It's wishful thinking. Neutral networks need to be over parameterized to find good solutions, meaning there is a surface of solutions. The optimization procedure tries to walk towards that surface as quickly as possible, and tend to find a low-energy point on the surface of solutions. In particular, a low energy solution isn't sparse, and therefore isn't interpretable.

Post by tpoacher » Sun, Aug 17, 2025, 1:18 PM UTC

bobmarleybiceps wrote: post #6

dkdcio wrote: post #5

> Another useful property of the model is interpretability. Is this true? my understanding is the hard part about interpreting neural networks is that there are many many neurons, with many many interconnections, not that the activation function itself is not explainable. even with an explainable classifier, how do you explain trillions of them with deep layers of nested connections

I've decided 100% of papers saying their modification of a neural network is interpretable are exaggerating.