For an inside look at how Blume Global is solving problems with data science, we spoke with a Deep Learning Architect, Yen Tien Wu. He shared with us about the differences between machine learning and deep learning, as well as the new technologies he is utilizing to transform the supply chain industry.
Can you explain to us what your role is as a deep learning architect?
Yen: Essentially, I use deep learning to power our products. One of my current projects utilizes object detection (also known as RetinaNet) to recognize port terminal congestion and estimate waiting time.
What’s the difference between machine learning and deep learning?
Yen: Deep learning is based on machine learning and utilizes the specific architecture of neural networks to accomplish tasks with images and languages. Machine learning is limited to working with data that contains numbers and class. However, detecting features like a face, objects, or understanding the difference between words like king or queen are done through deep learning. It’s impossible to use machine learning to create a model to recognize these things since they are not a number.
I recently attended a lunch and learn where you demoed one of your projects utilizing object detection, can you explain it?
Yen: In this project, I’ve utilized object detection to help track port congestion. This technology utilizes artificial intelligence to recognize the congestion of port terminals and assigns levels as a number based on how congested it is. In our app, we’re able to show customers almost a Google map of how much time a truck driver can expect to wait at the port terminal and gives customers real-time updates. In the future, we’re hoping to detect specific container and the chassis to track where they are exactly in the port terminal.
How does this kind of technology benefit the supply chain?
Yen: We can use deep learning to turn something into data. Through this technology, we can detect and count the containers coming in through port terminal. Our target is to cover all ports around the world. We have also built a deep learning model to read articles and identify what it is – whether a receipt or shipping data.
Instead, of relying on manual processes to record container numbers, we can reduce errors and increase efficiency. We can also utilize deep learning to build a model to determine the proof of delivery, simplifying the overall process.
How might this kind of technology be utilized in the future within the supply chain?
Yen: Currently, we use ML to predict approximate times but it’s hard to know exactly where containers are at every point of the shipment journey. Ideally, in the future there will be more cameras we can utilize to detect where the container is at all times and to detect container numbers. We may even be able to tap into surveillance cameras on the highways, to track where the container is at every point of the shipment journey.
To learn more about how Blume Global is powering the supply chain with data-driven technologies, make sure to visit our Solutions page here.