Machine learning is impacting countless industries, from the recent discovery of a black hole to improving healthcare, we are just scratching the surface. The retail industry is a prime example. Retailers and manufacturers are racing to figure out how they can employ machine learning to target specific consumers, monitor trends, and discover new pricing models.
While retailers and manufacturers are doubling down on new ways to target and sell to consumers, Jia Rui Ong, a two-time Nanodegree program graduate, and his team are employing machine learning to help you, the consumer, find the best price for the clothing you desire.
We recently had a chance to sit down with Jia Rui Ong and his team at Yux to discuss their product, as well as, our newly updated Machine Learning Nanodegree program.
Could you tell us a bit about your company, Yux?
Yux is a Chrome extension app that helps consumers find clothing at a cheaper price. More specifically, if a consumer finds a piece of clothing that they like, say a t-shirt, but they are looking for a better deal, our app can assist them in finding a suitable alternative. It’s fairly simple, using our app, the consumer takes a snapshot or “crop” of the item they are looking to purchase, and our app searches across hundreds of brands, in just seconds, to return the most similar items at competitive prices.
How did this all start? Do you and your founders have a background in computer science?
Yux was started by Jie Xun, Yao Yang and I. We are a team of three Computer Science seniors from Nanyang Technological University, in Singapore, who are passionate about machine learning.
As university students, we have to shop on a budget. We decided to come together and create something when we realized we were encountering a common problem: searching for and finally finding an article of clothing, shirts in particular, that we liked, only to realize that the price was unaffordable. Faced with this problem, we wanted to build a solution. What we came up with was an online visual search engine that would help others like us find affordable and stylish alternatives to shirts that they really like.
How did you find Udacity and what motivated you to enroll?
Jie Xun and I found Udacity a couple of years ago. We were searching for courses beyond what our university was offering. In particular, we were interested in deep learning. In looking at online courses, we came across the Deep Learning Nanodegree and Artificial Intelligence Nanodegree programs. For us, it was the first online course that brought together cutting edge concepts like CNNs (Convolutional Neural Networks), RNNs (Recurrent Neural Networks) and GANs (Generative Adversarial Networks) in a project-based setting. We were impressed by the world-class ensemble of instructors including Ian Goodfellow, Andrew Trask and Siraj Raval, and decided to enroll.
How have your Udacity skills played a role in developing your app?
The Nanodegree program experience provided us with a broader understanding about how machine learning skills can actually be applied in a practical sense. More specifically, the training and projects that involved convolutional neural networks and transfer learning were crucial for building Yux. These skills were the foundation for creating the image classifier that our app utilizes. Transfer learning is an important skill, as the lack of training data is still a significant limitation for most real-world applications of machine learning today.
We just announced the launch of our new Machine Learning Nanodegree program in collaboration with Amazon Web Services. We added two new projects focused on deployment skills, are these relevant to your company?
Absolutely; deployment of machine learning models to production is an important part of our work. When we started building Yux, there weren’t any resources to learn how to deploy machine learning models, so we taught ourselves through the build process. Training a CNN is just half of the work. In order to make predictions on the fly, we had to expose the TensorFlow model on an AWS Lambda endpoint. Additionally, we had to hook it up to a persistent database, and also ensure the security of our deployment stack. So, yes, deployment skills are crucial.
When did you begin building the app and what are some of your biggest takeaways in starting your own company?
We started working on the core image search engine in 2017 and launched our Chrome Extension App in Fall 2018. In hindsight, one of our biggest takeaways is the importance of launching early, even if your product is not yet perfect. Being more or less first to market is a great position to start from; now, we can iterate and improve upon the app as its live.
Another important aspect of the product that we learned early on, is that users would love the ability to utilize this app with Youtube. This has informed our training data collection process and helped broaden our focus on how we can help consumers.
What do you think of the future? What do you hope to achieve/where do you hope to go next with the app?
Our team believes Artificial Intelligence will play an important role in the fashion industry for years to come. We see an opportunity in taking Yux beyond just a visual search engine to a personal AI fashion stylist that brings together the physical data about your wardrobe and the online data of your fashion interests to offer the most personalized online shopping experience out there.
Do you have any advice you would give to others that might be considering an app idea?
Don’t be afraid to move forward with an idea, even if you might not have the skills to build it today. There are many opportunities, in our case, Udacity, that will teach you practical skills necessary to get your first product to market. From there, you can iterate until it meets industry standards.
Congratulations, Jia Rui Ong and team! If you would like to learn the latest machine learning skills, check out the Machine Learning Nanodegree program which includes two new projects, focused on deployment skills with Amazon SageMaker.