What we're about
Upcoming events (2)
Date: 11th/Nov/2019 Location: Google Office, Pyrmont Sponsor: Google (Food, Venue and Drinks) 5:45pm: Registration 6:30pm: Opening speech, and housekeeping words by Richard Xu and Matthew Ranocchiari from Google 6:32pm: "Efficient Diversified Mini-Batch Selection using Variable High-layer Features" and "Realistic Image Generation using Region-phrase Attention" using TensorFlow Abstract: Two recent works will be presented in detail. Their theories and the corresponding TensorFlow coding will be shown for both of them to illustrate how we achieve it. These works are part of the proceedings to appear in the Asian Conference on Machine Learning this November. Firstly, Stochastic Gradient Descent (SGD) has been widely adopted in training Deep Neural networks of various structures. In this talk, we show methods instead to use variable higher-layer features that are updated at each iteration when the parameter changes. To avoid the high computation cost, several contributions have been made to speed up the computation of DPP sampling, including (1) using hierarchical sampling to break down a single DPP sampling with large Gram-matrix into many DPP samplings of much smaller Gram-matrix and (2) using Markov k-DPP to encourage diversity across iterations. Empirical results show a much more diversified mini-batch in each iteration in addition to a much-improved convergence compared with the previous approach. The talk for this work will be 20 minutes. Secondly, The Generative Adversarial Network (GAN) has achieved remarkable progress in generating synthetic images from text, especially since the use of the attention mechanism. In this talk, we proposed a novel method in which we introduced an additional set of natural attentions between object-grid regions and word phrases. A set of auxiliary bounding boxes defines the object-grid region. They serve as superior location indicators to where the alignment and attention should be drawn with the word phrases. We perform experiments on the Microsoft Common Objects in Context (MSCOCO) dataset and prove that our proposed approach is capable of generating more realistic images compared with the current state-of-the-art algorithms. The talk for this work will be 10 minutes. Speaker intro: Ms. Erica Huang graduated from Honors (first class) from the University of Sydney in 2016. She is currently a Ph.D. student at the University of Technology Sydney (UTS) supervised by Prof Richard Xu. She is specializing in probabilistic Deep Learning Generation. She is also a lecturer in AI and Machine Learning Masterclass at UTS. 7:00pm: Just how HOT is Deep Learning? by Richard Xu Abstract: In this talk, Richard will show the audience some of the current and interesting statistics on how hot Deep Learning is, both from a scientific research and application point of view. Speaker intro: Richard is an Associate Professor in Machine Learning, and the director of Machine Learning and Data Analytics Lab, Global Big Data Technologies Centre, University of Technology Sydney. He has been a first-line researcher in ML, Data Analytics and Deep Learning. He has published more than fifty peer-reviewed publications and has co-authored with the world’s best statistical machine learning researchers in Oxford and Cambridge. Richard wrote more than 1500 slides of Statistics, Probability, and ML course for Ph.D. students around the world, and he also published one of the world's most popular Mandarin-speaking machine learning MOOCs. He leads a group of 30 talented scholars, data scientists, Ph.D. students, engineers, and communicators to apply their research and engineering skills in industry insight analytics. 7:15pm Announcements and closing remarks by Google 7:25pm – 8:30 Social time We encourage people to stay and to socialize. We strive to make the meetup a real "meetup" event. So please stay and enjoy the night with hundreds of Deep Learners!
Whether you are a brand new TF user, or have been using it for more than 10 month since it was first released, we have something for you. We will demo the basics on how to use TensorFlow and we are also inviting someone from Google to interact with the audience.