Deep Learning Meet-up #18

##### Share, Exchange and Learn ######

Please join our Facebook group for knowledge exchange & updates:
https://www.facebook.com/groups/deeplearninghk

& our github page for materials in previous tech updates section:
https://github.com/deep-learning-hk/agenda

=======================================================================

Open topics:
== 1. "Weak Supervision" by Kin Tse ==

• We will go through the article about Weak Supervision by Stanford AI Lab (http://ai.stanford.edu/blog/weak-supervision/). We will introduce Weak Supervision and then we can have some discussion around this topic.
• What is Weak Supervision?
  Weak supervision is a methodology to cheaply leverage subject matter expertise to programmatically label millions of data points. (excerpted from https://towardsdatascience.com/a-technique-for-building-nlp-classifiers-efficiently-with-transfer-learning-and-weak-supervision-a8e2f21ca9c8)

== 2. "Use Facebook StarSpace algorithm to do user embedding" by LP Cheung ==

• StarSpace is an algorithm proposed by Facebook for entity embedding. It is generic enough to embed whatever you want. An interesting application is user embedding which empowers the downstream tasks, such as demographic attributes prediction. In HK01, we use StarSpace to do article and user embedding.

I will go through the concepts and how to use it.

reference:
[1]: the original paper: https://arxiv.org/abs/1709.03856?fbclid=IwAR32Q0AerKcih_pAN_QIBl7-bY7T-FmjeGKKK-HV0ReI-EyaPWt8JsZf17Q

[2]: my medium post: https://medium.com/@lpcheung/starspace-mining-and-embedding-user-interests-28081937f95

Agenda:
8:00pm - 8:30pm - Intro and Tech updates from everyone
8:30pm - 9:00pm - Open Topic 1 & Topic
9:00pm - 9:30pm - Open Topic 2
9:30pm - 10:00pm - Free Coding!

== Special thanks for Carol's sponsorship for the venue ==