Applications are the foundation of your business. As IT professionals, you need to provide the right infrastructure for these apps to perform as expected and deliver the best experience possible. You know how to do this, you’re good at it. This is your job after all. But when it comes to new types of applications such as Artificial Intelligence (AI) and Machine Learning (ML), getting IT resources aligned with the need of your internal stakeholder can be challenging.
At Cisco Live Barcelona earlier this year, we showed multiple use cases where AI and ML can make a genuine game-changing difference for the enterprise. In fact, we showed demos in multiple verticals and use cases, including retail, natural language processing (NLP), and image classification.
Brick and mortar retail stores face intense competition with online retailers. In many cases, online retailers have been able to leverage AI and ML to track the likely behavior of customers, and even potential customers. Yet, a traditional physical store may not employ AI/ML. At Cisco Live, for which we partnered with Intel and Vispera, we showed a shelf full of products, just like a traditional brick and mortar store. With a camera and Vispera software running on Cisco HyperFlex, the system was able to identify what products were on the shelf, automatically track inventory based on customers taking and putting products back on the shelf, and monitor customer behavior. This type of automation eliminates manual counting of inventory and can even lead to better prediction of product demand.
The whole AI/ML retail solution was running on 2-node HyperFlex Edge mounted on the wall, ensuring that it could be easily deployed, even in tight physical environments.
Natural Language Processing
One of the most exciting developments in AI/ML has been in Natural Language Processing (NLP). Google has created the pre-trained Bidirectional Encoder Representations from Transformers (BERT) model, which can process natural language much better than many previous models. At Cisco Live, we showed BERT running on a Cisco UCS C480 ML with 8 x NVIDIA V100 GPUs in a Jupyter Notebook that a typical data science team would use. Then, we fed the text from the Wikipedia page on Cisco UCS and asked BERT to find the answers. As you can see in the screenshot below, BERT was able to accurately answer multiple questions.
This type of natural language processing can help doctors go through years of patient medical records, process financial reporting reports faster, and can even be part of interactive chatbots to answer customer questions. The use cases are simply endless.
As part of a fun demonstration, Cisco created a deep learning model to recognize Cisco logos. In fact, we challenged customers to take photos of Cisco logos around the World of Solutions at Cisco Live to see if there were any Cisco logos that the model couldn’t identify. Rather than simply running the model on a server, we ran it on an NVIDIA Jetson Nano to highlight that the data pipeline can be extended all the way to the edge. In this case, the model was running in a very small form factor board.
We want to thank Groupware Technology, an IT solutions provider, for helping Cisco to develop this demo. In fact, Groupware used data augmentation techniques, such as rotating the logo, changing colors, etc., to create a training data set of 54,000 images. This data set then went through 200,000 steps of training to create the model that was placed in the NVIDIA Jetson Nano. While the model recognized many Cisco logos, even those found on the Cisco Live backpack, there was just one Cisco logo that the model could not recognize. Of course, that Cisco logo will be added to our training set to improve the model.
While this logo recognition is done merely for fun, it highlights the value of having a data pipeline extended all the way out to the edge, enabling new applications that were not possible before.
Supporting Data Pipeline with Kubeflow
Many customers were surprised to hear that Cisco is the number two contributor to the open-source project Kubeflow, right behind Google. Kubeflow integrates deep learning frameworks, such as TensorFlow, with Kubernetes and makes it easier to move your AI/ML workloads between the cloud and on-prem. At the show, we had an opportunity to chat about how Cisco HyperFlex can leverage Kubeflow to extend the data pipeline from the data center to the edge.
Even More AI/ML at Cisco Live Barcelona
There were many more AI/ML discussions at Cisco Live Barcelona. In case you missed some of the breakout sessions, here are two more on AI and ML that you might find of interest:
- Operationalize Machine Learning: Bridging the gap between Data Scientist and IT – BRKINI-2348
- How to optimizer your K8s infrastructure for AI/ML deployment with a few clicks – PSOCLD-2982
Are You Leveraging AI and ML in Your Enterprise?
As an IT organization, how are you helping your enterprise to leverage AI/ML? Do you have the same challenges as many Cisco customers when it comes to working with your data science team?
Feel free to reach out to your Cisco account team and let us help you to accelerate your AI/ML deployment.