textgenrnn

hy. To do this entirely anonymously, the next step would be to publish on a public blockchain or anonymous service so that the owner's device (that generated the emails originally) can uploaded a signed statement that proves they were the phone pwned and who the offending app was. A similar idea seems baked into a couple of crypto initiatives https://coincentral.com/sentinel-protocol/ but fundamentally we're talking about an anonymous reputation system modeled after how swarms operate to gossip risk. It would be necessarily stochastic in nature because you'd be depending on the 3rd parties to send emails a bit at a time, but if you get a deluge of phones all reporting the same app, you can assume fairly confidently that app has been compromised. Punishment (Brand reputation, sanctions by app store) for being compromised would encourage better security. This could (and would need to be) operated at the hardware level and orchestrated by the OS, and OS provider. This is the kind of thing apple and google could do as part of their privacy initiatives around "differential privacy" https://venturebeat.com/2019/12/21/ai-has-a-privacy-problem-... What, you think that deep learning chip on your phone is there to make cute avatars?

nerally the embeddings work best on text which follows real-world word usage, which memes do not follow. (in this case, it's better to let the network train the embeddings from scratch) 3) A 512-cell LSTM might be too big for a word-level model of that size; since the text follows rules, a 256-cell Bidirectional might work better.

.

erature from my 128-cell LSTM network (https://github.com/minimaxir/textgenrnn/blob/master/outputs/...), which is several orders smaller (OP uses a 512-cell network), and many magnitudes lower amount of training data. (I am curious how long OP spent training the network.)