By: Luba Gloukhova, Founding Chair, Deep Learning World

In anticipation of his upcoming conference co-presentation, CNNs:  A Game Changer for Manufacturing?, at Deep Learning World in Las Vegas, June 3-7, 2018, we asked Abbas Chokor, Staff Data Scientist at Seagate Technology, a few questions about his work in deep learning.

Q: In your work with deep learning, what do you model (i.e., what is the dependent variable, the behavior or outcome your models predict)?

A: As the volumes, variety, and velocity of produced data from our production lines increase, our organization is committed to support its internal stakeholders all over Seagate to make the factory smarter.  Our engineers and operators are challenged every day by TB’s of data from thousands machines and drives that require visual inspection and analysis.

With the rise of deep learning as a promising solution in the area of object detection and image recognition, Convolution Neural Networks (CNN) are unlocking the value of our data in many applications. For instance, one of our models is able to accurately detect and find physical and magnetic defects in our products. Another model is boosting the security of our facilities through automated facial recognition as a new way to validate employee access.

Q: How does deep learning deliver value at your organization – what is one specific way in which model outputs actively drive decisions or operations?

A: Seagate produces and ships daily hundreds of thousands of drives. What is easy and quick for a CNN is hard and dull for humans. Deep learning models are able to detect defects in our products in very early stages and substitute human biased visual inspection. They are also able to replace the manual check-in/check-out of operators and visitors by an automated video surveillance system. As such, deep learning is efficiently shortening our production cycle’s time, improving the quality and reliability of our products, and therefore reducing costs.

Q: Can you describe a quantitative result, such as the performance of your model or the ROI of the model deployment initiative?

A: By having sufficient amount of data and picking the right solution for the right problem, we are able to make the business case for deep learning. From reducing manual person hours to enhancing the quality of our products, such models are bringing a lot of value for our business. Results showed CNN’s increasing the accuracy of prediction by 15 to 20%, compared to traditional machine learning models. Such increase in the performance is equivalent to substantial cost savings.

Q: What surprising discovery or insight have you unearthed in your data?

A: CNN proved to be an effective method to boost the performance of deployed machine learning models to unprecedented accuracies. However, in some cases, converging to the right architecture is time consuming and requires a lot of training. Thus, some light “preprocessing” before feeding the data in a CNN is very helpful. Moreover, instead of feeding all the data in one deep neural network, creating an ensemble of “small” neural networks could deliver the same results and avoid overfitting.

Q: What excites you most about the field of deep learning today?

A: As we move Seagate to a World Class Data Science company, deep learning is proving to be a game changer for our operations. Yet, many challenges are ascending after deploying some of our models in production. How to transfer the knowledge between models? In other words, can machines teach other machines? Such concerns are paving the way for many exciting opportunities, especially within evolutionary algorithms. Automating the development of deep learning algorithms is an interesting aspect that will gain a lot of momentum in the next few years.

Q: Sneak preview: Please tell us a take-away that you will provide during your talk at Deep Learning World.

A: There have been many lessons that we learned from our practices at Seagate. By leveraging traditional machine learning and deep learning algorithms, we were able to develop and deploy effective models in our factories. Investing efforts in training and educating our stakeholders up front was a crucial point to get the most value out of our data and change the culture within our company. Performing some preprocessing and ensembling neural networks from different subsystems simplifies the model development and deployment. In this session, we will share with you how we are deploying models on the edge with under $50 per system.


Don’t miss Abbas’ conference co-presentation, CNNs:  A Game Changer for Manufacturing?, on Wednesday, June 6, 2018 from 1:25 to 2:15 PM at Deep Learning World in Las Vegas, June 3-7, 2018. Click here to register to attend.

By: Luba Gloukhova, Founding Chair, Deep Learning World

Luba Gloukhova facilitates and accelerates advanced research projects at a major R&D hub of the Silicon Valley. She supports Stanford GSB faculty by conceiving and generating innovative solutions that drive their cutting edge research. Luba also serves as the founding chair of Deep Learning World, the premier conference covering the commercial deployment of deep learning.