Maxis x WWCode KL: Deep Learning & Computer Vision


Join us in the evening to hear from our speakers on evolution of image compression tech and tips & tricks in training deep neural networks on noisy labels!

Tentative Schedule

19:00 Arrival & light refreshments
19:20 Welcome
19:30 The Past, Present and Future of Image Compression by Roza
19:50 Sharing by Maxis AI Stakeholders: Kai Hui & May Ching
20:10 Training Deep Neural Networks on Noisy Labels by Jan
20:30 Q & A
20:45 Networking
21:00 Adjourn

Speakers & Topics

1. Fairoza Amira Binti Hamzah (Roza) is currently pursuing Ph.D in the Information Science and Control Engineering of Nagaoka University of Technology, Japan and expected to graduate at the end of March 2019. She will be working as Assistant Professor in the Kyoto College Graduate Studies of Informatics in Kyoto, Japan from April 2019. She has internship experiences in University of Warwick, UK, Panasonic Co. Ltd., Japan and Google Inc., Japan.

The Past, Present and Future of Image Compression: Prompt advances in multidimensional image data and its resolution have caused the development study of the most suitable compression method for them to be very crucial. The prevalent JPEG compression standard has becoming unsuitable for current image type, thus making researchers to move towards in utilizing deep learning architecture for image compression. Furthermore, various image compression architectures such as the BPG, JPEG 2000 etc., are being studied to cater the needs of current data.

2.1 Wu Kai Hui is a mechanical engineer by training. Kai Hui spent her university years eating, sleeping, and breathing the laws of thermodynamics, material sciences and robotics. Equipped with all things hardware, she plunged herself into the software world and by complete chance entered the realm of Deep Learning. From navigating Jupyter Notebook to working on the state-of-the-art neural networks, she tries her hand on everything and hopes to one day combine robotics with her newfound skill in AI to create THE next big thing.

The Failures, Successes and Pitfalls of Taking Your First Steps in Deep Learning: Deep Learning is the next BIG thing. Everyone is scrambling to learn it as fast as they can. But if you don’t have a degree in Data Science how can you deliver results? How do you convince your company to take a gamble and invest in your project? This is the perfect after-dinner story of Drones, Object Detection, Cell Towers, including what happens after your project has garnered stakeholders interest and how to scale such initiatives throughout the company.

2.2 Ng May Ching built a robot as an engineering student in university, and is currently the head of IT in Maxis, where she leads a team to develop, deliver and develop, test and operate all the application platforms in Maxis, from go-to-market product and channel development to innovation and centers of excellences, from apps to insight to cloud solutions. She has recently rediscovered that neural networks from engineering school days have become popular again along due to google cat videos, and cloud, big data and digital proliferation, and is excited to build the capability to develop new AI business solutions which can work 24/7 whilst we sleep.

AI on our Deep Minds: Today we have an explosion of data points we collect about our customers - our customers look at 1.3 million apps on the mobile phone in a day, 5 GB of data browsing records every second and millions of “footprints” - customer interactions - this is a good problem to have to better analyze the data to provide personalized and unmatched customer experience and uncover new business models, but how do we do this? Learn about how we ventured into the world of AI in Maxis, why it matters and how we can unlock the possibilities of AI powered business solutions.

3. Dr. Jan Sauer is a data scientist currently working at the Center of Applied Data Science (CADS), where he designs and delivers customized data science solutions to corporate clients throughout Southeast Asia. He was previously a scientist at the German Cancer Research Center, where he developed a data-driven approach to identifying personalized drug treatment plans for colorectal cancer patients. His contributions to this project consisted of the design and implementation of a deep learning platform to identify complex, multicellular structures in images as well as assess the patient-specific impact of drugs.

Training Deep Neural Networks on Noisy Labels: High-content image analysis is a powerful and well-established method employed in many industries, including biotechnology, manufacturing, robotics, and surveillance. Nonetheless, the automatic segmentation of complex images is often very difficult due to their heterogeneity. Deep learning has emerged as a promising approach to solving this problem thanks to the incredible power of deep neural networks (DNNs). The downside of DNNs is that they require a large amount of labeled training data, an impossibly time-consuming task. In this talk, Jan presents an example of the robustness of deep neural networks towards training label corruption and show how this can be used to design a semi-supervised algorithm for the segmentation of microscope images of multicellular structures that doesn’t require any manual annotation of training data.

  • Venue Address was changed to "Lvl 25, Menara Maxis-next to KLC". Orig#415359 2019-03-05 02:25:05
More updates
Wed Mar 6, 2019
7:00 PM - 9:00 PM MYT
Add to Calendar
Everyone FULL
Ladies Only FULL
Reserved SOLD OUT RM99,999.00
Venue Address
Lvl 25, Menara Maxis-next to KLC Malaysia
#WWCodeKL Women Who Code KL