Devin Soni
Devin Soni
There are many possible attacks that have not yet been implemented. Some of these include: - Phrase-level attacks - Invert part-of-speech order - Change tense - Replacing words with homonyms,...
Currently this assumes words are separated by spaces. If this is not the case, or if an earlier attack removes a space, several words may be treated as a single...
Right now we assume no feedback between adversary and classifier. What if the adversary has access to the labels? What if the adversary has access to the raw probabilities? What...
Such as the various ReLU variants This entails adding the raw functions for them as well as adding entries for them in the serialization/deserialization functions
Currently, the testing just makes sure that the algorithm runs without error on a toy problem. It would be nice to do something more akin to unit testing, but I'm...
Of course, more algorithms are always great. Some suggestions: * Coordinate descent * Ant colony optimization * Differential evolution * Cuckoo search * Cross-entropy method
It would be good to find all of the instances where the algorithms may be unstable and handle these cases appropriately (such as overflow). Some cases are handled, but there...
Where applicable, it would be nice to implement both the continuous and discrete versions of algorithms.