schedule May 15th 11:45 AM - 12:15 PM place Wesley Theatre people 197 Interested

You must have heard it a few times that AI has beaten human in image recognition. Is that true? Have you seen it yourself? I am going to demonstrate Cyclops, an image recognition we built to recognise car model far better than any human.

From here on, this talk will take you through our journey, how it's all began, why we built the early version of Cyclops and what was the outcome. Furthermore, how we used this technology to dramatically improve consumer experience and built many consumers facing products which we thought was not possible before.

I will then dive down into technical details, starting from how we built Cyclops 1.0 with Tensorflow and how we overcame the training complexity with transfer learning. However, transfer learning comes with a limitation of directional invariance in which I will show what it is and how we overcame it with our novel solution.

Next, I will show you that building a car recognition as complex as Cyclops 2.0 requires a more superior model and modification of our existing transfer learning technique. I will also take you to see problems we faced with low coverage when we are going deeper and how we solved them. I will then investigate how a distributed training can speed up the training process to make this practical.

 
 

Outline/Structure of the Talk

Start with live demo and continue with slides. I like my slides to be very visual, less text and avoiding jargon as much as possible to make it easier for audience to understand and more engaging. My talk will end with question and answer.

1. Live Demo. Who’s better, AI or Human?

2. History

Where does it all begin? Why did we build Cyclops 1.0? What was the problem we tried to solve? How did Cyclops provide solution? What was the end result?

3. Application

What cool consumer facing product we built with Cyclops tech? How does it improved consumer experience dramatically?

4. Technical/Implementation

How did we build Cyclops? Using Tensorflow framework, Convolution Neural Network (CNN) architecture with Inception V3 & V4 models. How does CNN work? How did we minimise training complexity with transfer learning? How does transfer learning work? What problem comes with transfer learning and how dd we solve it? How does distributed training speed up training process even more. What problems we were facing when building Cyclops 2.0 car recognition and how did we solve it?

5. Question and Answer

Learning Outcome

Audience will gain an understanding of image recognition system and how can it be used to build creative products and improve consumer experience. The will also learn how to reduce training complexity with transfer learning, and distributed training and how to overcome problems common problems when building an image classification as complex as Cyclops 2.0

Target Audience

Developer, Tech Lead, anyone with an interest in Machine Learning and Image Recognition

Prerequisites for Attendees

Basic understanding of Computer Science and Machine Learning

schedule Submitted 1 year ago

Public Feedback

comment Suggest improvements to the Speaker