Next Meetup

#17: Fair and Private NLP
Where: Tyro FinTech Hub, 5/155 Clarence St, Sydney NSW 2000 When: Monday December 17th from 6pm; presentations from 6:30pm. Join us to hear about the intersection between privacy, ethics and NLP. Hope to see you there! -- Team NLP Sydney TALK: Mind the GAP - Kellie Webster Despite large gains on academic benchmarks, neural network models still make mistakes which can be both embarrassing and have real social impact in the industry setting. Errors in coreference resolution are particularly egregious since the correct resolution is easy for humans to label, hard for systems to learn, and has the potential to impact a great variety of high-visibility applications including fact extraction and translation. In this talk, I will present a new coreference benchmark Google AI Language has released, GAP, which is sampled to provide diverse coverage of reference problems seen in-the-wild while being balanced in a number of dimensions including pronoun gender. goo.gl/language/gap-coreference TALK: Generalised Differential Privacy for Text Document Processing - Mark Dras In common with many fields of research that work with large quantities of human-derived data, how to keep attributes of data private -- for instance, for a text, the author gender, native language or identity -- against inference mechanisms built with machine learning is a topic of interest in NLP. Most existing work proposing techniques for improving text-related privacy has approached it on an empirical basis, giving no guarantees in terms of privacy, and has focussed on preserving privacy of texts in a learner's training set: a goal there is to produce author-obscuring word embeddings or other text representations that can be made available in place of the original text. Another kind of scenario is one where there is an adversary who has access to original raw data -- and is consequently able to build strong inference mechanisms -- and the goal is to keep private some property of a new piece of text, with some guarantees attached to this. One well-known and robust framework that does give guarantees is differential privacy, but its reliance on the notion of adjacency between datasets has complicated its application to text document privacy. However, generalised differential privacy permits the application of differential privacy to arbitrary datasets endowed with a metric and has been demonstrated on problems involving the release of individual data vectors. In this talk I will outline how generalised differential privacy can apply to the specific NLP task of author obfuscation. ABOUT KELLIE Kellie obtained her PhD in computational linguistics from the University of Sydney with a thesis exploring how cognitive insights may be applied to automatic coreference resolution. Since then, she has been working at Google on a range of discourse modeling problems. She is most proud of her work on diversity and inclusion initiatives, including establishing the Google AI mentorship program which aims to strengthen the pipeline into AI research for promising undergraduate students from groups traditionally underrepresented in technological careers. ABOUT MARK Mark Dras is an associate professor in the Department of Computing at Macquarie University in Sydney, Australia. He obtained his PhD from Macquarie University and the Microsoft Research Institute in Australia; following that he was a research fellow at the University of Pennsylvania's Institute for Research in Cognitive Science, before returning to Macquarie. His research is mostly in the area of Natural Language Processing, where he has worked on a range of problems currently including native language identification, image captioning and privacy in text processing. He is an executive board member of the Asian chapter of the Association for Computational Linguistics (ACL), the main scientific and professional society for people working on computational problems involving human language.

Tyro FinTech Hub - Events Space

Level 5, 155 Clarence Street · Sydney

    Past Meetups (16)

    What we're about

    Sydney community for developers and applied researchers working on natural language processing (NLP). Our core interests include applications, crowdsourcing and interactive NLP. We'll be holding events every few months, with a mix of networking and technical talks.

    Goals:
    1) provide a venue for individuals and companies to share their triumphs and challenges with each other,
    2) share opportunities with those who are keen to make a career in NLP yet stay in Sydney, and
    3) help Sydney become a global leader in NLP.

    We welcome different levels of experience, from those curious about the field, those building or applying NLP in their work, researchers, and experts.

    #nlproc

    Members (564)

    Photos (123)