3 ms·
>My point is that "more data" here simply represents the mental effort that has already been exerted in pre-AI/DL era, which we're now capitalizing on while we
by sebzim4500 4y ago
>My point is that "more data" here simply represents the mental effort that has already been exerted in pre-AI/DL era, which we're now capitalizing on while we can
Not really. It's not simply that modern architectures are not adding additional inductive biases, they are actively throwing away the inductive bias that used to be used by everyone. For example, it was taken for granted that you should use CNNs to give you translation invariance, but apparently now visual transformers can match that performance with the same amount of compute.
- 0xBABAD00C 4y agoI don't think what you're saying contradicts what I'm saying. My baseline / reference point wasn't CNNs.
- jerpint 4y agoThere are bit caveats here: Vision transformers outperform CNNs in HUGE data regimes. On small datasets, CNNs still shine. Also, if you take a CNN with modern tricks, they can be on par with vision transformers, e.g convnext Transformers really dominate when you scale the amount of data to infinity
- alexmolas 4y agoYes, that's what I was saying: scale > inductive bias. Apparently, scaling is all you need ;)