Hey, Sorry, There was a small mess up yesterday hence you received only a part of the actual newsletter. Here’s the rest!
TextFooler
MIT’s new system, “TextFooler,” a general framework that can successfully attack natural language processing (NLP) systems — the types of systems that let us interact with our Siri and Alexa voice assistants — and “fool” them into making the wrong predictions (adverserial attacks). TextFooler works in two parts: altering a given text, and then using that text to test two different language tasks to see if the system can successfully trick machine-learning models.
Github - TextFooler - A Model for Natural Language Attack on Text Classification and Inference - https://github.com/jind11/TextFooler
This is the source code for the paper: Jin, Di, et al. "Is BERT Really Robust? Natural Language Attack on Text Classification and Entailment." arXiv preprint arXiv:1907.11932 (2019).
Meena’s Cost
Speaking of NLP, It’s quite unlikely that you’ve not heard about Meena - The new conversational chatbot from Google, which is a 2.6 billion parameter end-to-end trained neural conversational model. Meena seems impressive with What Google has demonstrated. But do you know at What cost? (Hint: $1,400,000 of compute time to train this)
Pandas 1.0
Pandas, a project started by Wes McKinney has come a long way through to hit its 1.0 version. Yes, Pandas v1.0 was released recently with new Deprecation Policy (something you should care about if you’ve got alot of pandas code in PROD)
What’s new in Pandas 1.0.0 - https://pandas.pydata.org/docs/whatsnew/v1.0.0.html
Pandas 1.0 is here - https://www.anaconda.com/pandas-1-0-is-here/
It’s always great to see an open-source project thriving and succeeding with an awesome community and support. Let’s just wish, this happens with most open-source initiatives!
If you’ve enjoyed our newsletter, Please take a moment to share it with your friends and encourage them to signup!
Thanks!
~AbdulMajed