Avatar

Chief technology officer for Cisco Australia and New Zealand, Carl Solder, interviews a Cisco engineering fellow, JP Vassuer, on the theme of innovation and how he creates elegant engineering solutions. 

Cisco has an established reputation for driving innovation. From Cisco’s inception in the mid 1980’s – when it introduced its first multi-protocol router – to today, Cisco’s engineering teams have been responsible for producing over 25,000 patents.

Given Cisco’s diverse set of customers, it is in a unique position to learn about customer business problems and technical challenges, and it is from these learnings that our engineers’ minds get working to help solve those problems and challenges.

Over the last 10 years, the reliance on the network for both business and consumers has become much stronger. The world has seen an increase in use of the internet through the rise in social media use, the number of connected devices across the globe, and a greater reliance on those connected devices. This has forced companies to adapt their business to meet these changing dynamics. In this time, we have seen the emergence of new business and technology challenges appearing more rapidly and frequently, challenging the status quo of those established businesses. To help our customers meet the challenges that this dynamic and rapid changing business landscape throws at them, Cisco has picked up its pace of innovation to help its customers meet those challenges head on.

That begs a question. How does Cisco drive innovation? One way to find out more is by providing some insight into the mind of one of Cisco’s most prolific innovators, JP Vasseur. JP is a Cisco fellow and to put that position into perspective Cisco fellows represent 21 individuals out of the 30,000+ strong engineering force in Cisco – essentially the pinnacle of technical talent within Cisco. JP holds over 650 patents and has a current focus on developments in the artificial intelligence and machine learning space. Let’s chat to JP to learn a little more about himself and the innovation he and his team are helping to drive for Cisco’s customers.

Q: Welcome JP – can you tell us a little bit about who you are, where you are from and what drives you?

A: Hello Carl – who am I? I have been in the tech industry for over 30 years. I have worked in multiple countries, in multiple jobs and multiple industries. In this time I have been able to gain a very diverse background with many experiences that I can draw from that help me solve problems. I am an innovator at heart, but I also like to think of myself as a disruptor and someone who loves disruptive innovation. To me, being an innovator is about solving problems. Specifically targeting those problems that have not yet been solved and sometimes thought as non-solvable. I am also an engineer at heart and my engineering mindset allows me to bring some creativity to solving those tough problems. And it allows me to always innovate with customers.

Q: You currently have a focus on artificial intelligence and machine learning (AI/ML) – when did you start getting into this space and what sparked your interest to work in this field?

A: The spark happened about 10 years ago. I was working on a problem and had the realisation that traditional methods I had used previously for solving my problems were not working. I had to look outside of this to find a way to help solve my problem. That was how I learned about some new ML developments that I thought could help me. This got me started in this field.

Right now, there is a lot of chatter in the industry about AI and ML. What I have realised over that time is when ML/AI can be used to help solve problems, and when it’s not the right tool to solve problems. I am focused on using AI/ML to solve the right problems.

Q: Can you tell our readers what is the difference between AI and ML? Are they the same thing, are they complementary or are they two different things that solve different problems?

A: Artificial intelligence aims to imitate human intelligence by using algorithms and logic to simulate the way the human mind might work. Machine learning is a subset of artificial intelligence and is focused on improving its learning through mathematic and statistical modelling.

Q: Can you tell me some examples of AI solutions that Cisco has released?

A: Yes, there are quite a few.

The first example is Cisco AI network analytics. This uses advanced machine learning to identify critical issues in your WiFi network. We use AI driven baselining to define normal usage patterns by considering a number of parameters. The system dynamically learns a baseline with high granularity and then automatically identifies anomalies in the network, thus drastically reducing the noise of the many alarms potentially driven by static rules. We then drive anomaly detection to determine the root cause of problems to help ease troubleshooting. We have over 2,000 customers using this technology today.

The second example is Cisco AI endpoint analytics. It aims to help identify device categories. As the number and types of devices in a network keeps growing, for example with the internet of things; AI/ML helps reduce the number of unknown devices. It uses a combination of deep packet inspection, machine learning and integrations with Cisco and third-party products (like Cisco cyber vision, Cisco identity services engine and service now) to identify, categorise and label all endpoints. Moreover, we are also using ML algorithm to check that there are no spoofing attacks: when a device claims to be of a given type, ML algorithms trained to recognize the behaviour of devices of that type are used to detect spoofing attacks.

Another example of AI is the innovation we are driving with our Webex solutions. Using AI and ML techniques, we can target and cancel out background noise on a Webex call in real time. Noise that could be generated, for example, from a pet dog running into the room barking, or indeed the noise from your kids playing in the background. Built into this functionality is real-time voice to text translation as well. All are very cool developments that we are very proud of here at Cisco.

Q: Are there any exciting innovations you are working on now that we can expect to see soon?

A: Yes. I cannot tell you too much as it’s still highly confidential. As we have been releasing our AI solutions to date, our customers have been telling us how the power of AI is helping their networks become more operationally efficient. They are now challenging us to solve even harder problems in their network – problems that no existing product or solution can help with today. What I can tell you is that we want to evolve the power of AI and take it to the next level, doing something that no one else in the industry is doing. We are working on a project right now that we believe will change the game. We are excited at the promise for what this technology can do and how it can help our customers. And I can give you a teaser … we want to give the ability for networks to learn. More very soon.

Q: How far along is Cisco into our AI/ML development journey and what does the future hold for our AI/ML innovations?

A: My team and I have been working on ML for the last 10 years and we managed to produce ML products at high scale (this the hard part!). I think we have passed the first wave of ML innovations which was primarily about troubleshooting and anomaly detection. And there is so much more to do. As I said earlier, I like to solve problems that have not been solved yet. That means thinking outside the box even more than we have done before to engineer elegant and innovative ways for solving those unsolved problems. We want to change the game and take AI/ML to the next level and stay pragmatic and driven problems to be solved.

“I like to solve problems that have not been solved yet. That means thinking outside the box even more than we have done before to engineer elegant and innovative ways for solving those unsolved problems.”

Q: Are there any anecdotes you have heard from customers who have adopted your AI solutions so far?

A: I have two stories that come to mind.

The first story relates to us learning about the impact of human psychology when using our AI/ML solutions. We developed a system that looked for anomalies that would indicate an attack. Our system required the administrator to provide feedback (via a thumbs up or down) that helped our AI engine determine if the attack was legitimate. This customer told us our software was not working. We found out that the administrator was not providing the feedback needed by the AI engine just in case they gave the wrong feedback. There was a psychological effect in play that we did not consider. That helped us refine our software to take into consideration this psychological effect.

The second story made my day. It’s related to a current development that I cannot talk about too much, but I want to share some early success that we are having. We have been working with a customer who is an alpha customer for our latest AI development. When we first explained to them the power of this AI, they simply did not believe it was possible. After implementing and trying out our solution, not only did the AI software prove to the customer that it could do what we claim but also helped to “save the day” for a problem that happened in their operational environment. We are excited at the potential that this new development can have and cannot wait until it’s finished so we can show it to our customers.

Q: Lastly, do you have any message for anyone who would like to learn more about how to get into AI development?

A: Yes. There are many mistakes a budding AI developer might make. Many start by jumping straight in by taking the data set and immediately start using off the shelf AI libraries to do some analysis. Instead of this, start with taking a course on machine learning to learn the basics. There are many great courses out there. First do spend time on cleaning up your data and understanding the dataset. Do not try to implement in your lab next – think about “scale” since the largest issue will be to make it work at scale. Then figure out what you want to do with that data and what you want the algorithm to answer for you. As you learn more you will get to understand the multiple AI algorithms options that will work best for achieving your stated goal.