the challenges to deep learning just keep on coming. here’s an important technical paper by Sengupta and Friston showing that accuracy doesn’t always entail robustness. (corroborating some of the issues on my own critical appraisal of #deeplearning.) https://arxiv.org/abs/1804.11313
-
And unsurprisingly even very simple programs struggle with perturbations. Simple Turing machines for example don’t always recover from simple perturbations. Indicates more complex structures won’t fair any better.
Don’t think any simple machine is going to be robust across sufficiently complex range of circumstances. That’s why humans mix bottom-up and top-down info; System I & System II (reflex and deliberative), etc
-
-
And... I think we need to take seriously the idea that there’s no clear line between data and algorithm. I often find practitioners failing to test and adjust assumptions accordingly.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.