NUS
 
ISS
 

Deep Learning Masterclass on Computer Vision

This Masterclass will give participants an update on the latest advances in Deep Learning from the industry perspective and more significantly provides a practical jumpstart into Deep Learning using Deep Neural Networks in computer vision applications.

Key Takeaways

  • Gain a practical understanding about Deep Learning, Convolutional Neural Network and Network Architectures.
  • Learn & Apply Convolutional Neural Network to image classification problems
  • Acquire competencies in using TensorFlow framework and building image classifier together with pre-processing pipeline


Pre-requisite
 

  • Basic Python programming (Python3.5)
  • Understanding of Machine Learning & Neural Network concepts
  • Bring your own laptop (optional)



Enquiries

Please contact Ms. Maybelline NEO Manlin at tel: 65167646 or email to issnmm@nus.edu.sg for more details.

 

Date / Time / Venue
  • 29 Jan 2018, Monday
  • 9:00am - 5:00pm 
  • Institute of Systems Science
    25 Heng Mui Keng Terrace
    Singapore 119615
View Map
Fee: S$214
(Fees are inclusive of 7% GST)
Registration ended
on Monday, 22 Jan 2018 

Programme

9:00am

Conducted by: David and Jawad

  • Introduction to Computer Visions

    Introducing Computer Vision, different applications of CV, approach with hand-crafted feature extractors.

  • History of Deep Learning

    Inspired by biological neurons, brief on neural network and define deep neural network. The first paper that toys around the idea and the experiment will be mentioned.

  • Introduction to Convolutional Neural Networks (CNN)

    Overview on multi-layer perceptron and origins of convolutional neural network. Concepts such as Local Connectivity, Spatial arrangement, constraints on strides and use of zero padding are introduced.

  • Network Architectures

    Sharing network architectures of CNN, namely LeNet, AlexNet, ResNet, etc. Building a ConvNet, e.g.: Input, Conv, ReLU, Pool, Fully-connected layers.

  • ImageNet Benchmarks

    Illustrate the accuracy of network architectures on Imagenet dataset.

  • Visualizing the black box

    Making deep learning more transparent by visualizing the convolutional units. Sneak peek into a neuron to better understand workings of neural networks.

  • Recent advancements (Generative Adversarial Networks-GANs)

    Latest architecture, applications and how well it fare against previous approaches.

  • Practical tips on training Deep Neural Networks (DNN) models

    Sharing tips on model training and general rule of thumb on setting various parameters.

  • Deep Learning Frameworks & Libraries

    Big tech companies, e.g. Google, Facebook, Amazon, etc have their own deep learning framework. Learn the pros and cons as a starter.

  • Environment set-up
Setup Python environment for workshop.  
12:00 pm Lunch Break
1:00pm

Conducted by: Weimin and Zane

  • Motivations of TensorFlow
    • What is TensorFlow and why do we use it?
  • A brief introduction to TensorFlow
    • Building a simple one-hidden-layer neural network using TensorFlow
  • Build deep image classifier using Convolutional Neural Networks
    • Convolutional, pooling and Fully-Connected layers, lose functions, training and evaluation
  • Techniques to improve model accuracy
    • Batch normalisation
    • Dropout
    • L2 regularisation
  • Image augmentations
  • Using pre-trained model for transfer learning (Optional)
  • Discussions

Code walkthrough and how TensorFlow API works. Participants are expected to apply what they learn in the walkthrough session and produce an accurate model on the other dataset provided.

5:00pm Thanks & Goodbye

Programme may be subjected to changes.

Instructors

David Low-high resMr. David LowView Biography
Close

David Low is the Co-founder and Chief Data Scientist at Pand.ai, building AI-powered chatbot to disrupt and shape the booming conversational commerce space with Deep Natural Language Processing. Pand.ai is the only AI chatbot startup accepted into Nvidia Inception Program in the region and currently serves two Fortune Global 500 companies in the financial sector. David represented Singapore and National University of Singapore (NUS) in the Data Science Game'16 held in France and clinched top spot among Asian and American teams.

Throughout his career, he has engaged in data science projects ranging from Manufacturing, Telco, E-commerce to Insurance industry. Some of his works, including sales forecast modeling and influencer detection, had won him awards in several competitions and were featured on IDA's website and NUS publication. Earlier in his career, David was involved in research collaborations with Carnegie Mellon University (CMU) and Massachusetts Institute of Technology (MIT) on separate projects funded by National Research Foundation and SMART. As a pastime activity, he competed on Kaggle and achieved Top 0.2% worldwide ranking. He is occasionally invited to speak at local and overseas data science conference/events such as Analytics Leaders Summit, Taiwan Fintech Convention, DBS Machine Learning workshop and Deep Learning Summit 2017.

David Low-high res Mr. Zane LimView Biography
Close

Zane is a data science and machine learning practitioner and an AI enthusiast. He is currently working as a senior data scientist, leading a sub-team of data scientists at Go-Jek, Indonesia's first and largest unicorn technology firm. His job involves end-to-end predictive modelling application, from building data pipelines to machine learning modelling to engineering real-time deployment.

Off-work, Zane has participated in various data science hackathons and won good standings in a few of them, with the highlight being one of the Singaporean representatives at the worldwide Data Science Games competition held in Paris. He is an avid Kaggler and a mentor of Udacity's Artificial Intelligence nanodegree programme.

David Low-high res Mr Mohamed JawadView Biography
Close

Jawad is a Data Scientist with 4 years experience in Industry and Research. Some of his work includes simulation modelling for NUS School of Design and Environment, and anomaly detection for financial institutions. construction workers. His current role at Go-Jek involves solving high impact business problems through data science, machine learning and plain old software engineering.

Jawad believes in lifelong learning and spends his spare time participating in Hackathons/Kaggle and keeping up to date with emerging technologies such as deep learning.

Zane Lim Mr. Wang WeiminView Biography
Close

Weimin has 5 years of experience in Data Science and Machine Learning and holds a Master's degree in Statistics from NUS. Throughout his career, Weimin has accomplished various research and industrial projects, including Drug Discoveries using Machine Learning, which is a collaborative work between Merck and Stanford. He has also published papers, as well as represented Singapore and NUS in the 2016 Data Science Game held in Paris, where his team clinched top rankings globally. He has been invited as keynote speaker for various university workshops, industrial sharings, conferences and tech meetups. During his spare time, he is also a Kaggler Master, and has won various competitions in the top 1% spots.

Weimin is currently the Data Science Lead at Go-Jek, and he is responsible for several Deep Learning initiatives such as Real-time Time Series Forecast, Fraud Detection and Food Recommendation. He is passionate about creating business impact using AI, with current focuses on Deep Learning as well as TensorFlow in production. In his free time, he also loves to share his knowledge by publishing blogs.

 


More than one Google Analytics scripts are registered. Please verify your pages and templates.
A+
A-
Scrolltop
More than one Google Analytics scripts are registered. Please verify your pages and templates.