atpaino
Joined 15 years ago
Karma
351
Submissions
38
Member since
Nov 26
Recent submissions
Comment · 9 years ago
Comment · 9 years ago
Yep, it definitely has room to improve. The work thus far has primarily been a proof-of-concept for the methodology used to generate training samples (i.e. starting with grammatically correct text and introducing errors). Next step is to try to include more high quality data, after which I may try out comment data from HN, etc. I think it would be interesting to see what the effect of somewhat noisier data like that would have on the model.
Comment · 9 years ago
Thanks for the links! I'll have to try out some of these. The data is definitely the limiting factor at this point.
Comment · 11 years ago
Nice post. However, I was kind of surprised to see the author list interpretability as one of the drawbacks of random forests:
> Another point that some might find a concern is that random forest models are black boxes that are very hard to interpret.
I generally agree that random forests are more difficult to interpret than linear models such as logistic regression, but I think they're still far more interpretable than more comparable non-linear models such as neural networks or SVMs with non-linear kernels. At the end of the day, a random forest is just a bunch of decision trees, each of which are very straightforward for humans to understand. Additionally, there are a number of straightforward methods available for assessing the importance of each individual feature in a random forest in aggregate ([1], section 15.4). Neural networks, on the other hand, result in models that are too cryptic to be meaningfully inspected by a human, and have more complex variable importance measures [2].
The relative interpretability of these non-linear models played a large factor in our decision to add random forests to our modeling stack at Sift Science, which you can read more about here: http://blog.siftscience.com/blog/2015/large-scale-decision-f...
[1]: http://statweb.stanford.edu/~tibs/ElemStatLearn/
[2]: http://www.massey.ac.nz/~mkjoy/pdf/Olden,Joy&DeathEM.pdf
Comment · 11 years ago
Author here, feel free to ask me any questions.
Comment · 13 years ago
It seems like this would be much more useful if the front and back of the card were OLED screens, or something similar, so that you could toggle the appearance of the card. I feel like that would solve most of the problems people are suggesting here - i.e., it could replace driver's license, photo id's, wouldn't be a problem at bars that hold on to your CC, etc. Personally, I'm going to hold off on getting this until there's some functionality similar to what I outlined above, because currently this could only replace 2 out of the 5 cards in my wallet.
> In the limit, you might even have a neural network generate increasingly harder types of grammatical corruptions, with the goal of "fooling" the corrector network.
Very interesting. I wonder how many constraints would need to be added to the corruptor model to ensure the corrupted sentence retains the same meaning as the original. Somewhat related to that, I've thought that a more basic curriculum learning setup could be deployed quite effectively here, and am hoping to try that out soon.