When people ask me what artificial intelligence (AI) is, I tell them it is ultimately the next step in the evolution of automation. In the past, the way we automated things was very deterministic. We wrote scripts that basically did the same thing day in and day out, and you really didn’t have to think about it.
What we’re building with AI today is more about cognitive reasoning. These solutions are starting to act more like humans and change their behavior over time. It’s almost like hiring a new intern. First, you must understand their skills. After that, you expect them to get better over time as they learn. We’re really moving away from these machine learning (ML), deterministic algorithms to deep learning algorithms that learn things as you interact with them and use them more.
ML has been around for decades in the form of things like logistic regression and decision tree algorithms. With deep learning, we’re starting to use much more complicated models with many more layers, many more weights. The ability to continuously learn from a lot more data is what brought things like ChatGPT to life.
Supervised vs. unsupervised learning
With supervised learning, models like ChatGPT are trained on large collections of text data, such as books, articles, and web pages, to predict the next word in a sentence. It’s similar to what Juniper is doing with video conferencing app data. We’re building models that can actually use this labeled data to analyze video conferencing performance and predict when issues will arise. Unsupervised learning is what we did with location, where no labeled data was available. We needed to learn the path-loss model for every mobile device in an area to make a better location estimate.
AI model training
Almost all deep learning models have some sort of training schedule assigned to them. For example, let’s say I have a bunch of video conferencing app data that I have to join with my network feature data. With anomaly detection, we’re basically training models every day on the last three months of data. By using historical data, the model can predict anomalies that will likely occur within the next 10 minutes.
Neural networks vs. deep learning
A neural network is a general term for any network of interconnected nodes that processes data. The original neural network model was basically building up layers of weights that could be trained to predict something. Deep learning refers to neural networks with multiple hidden layers and complex architectures designed to tackle advanced machine learning tasks. It is commonly used in applications that require high computational power and extensive training data to achieve state-of-the-art performance. Think of them as very large neural network models with 175 billion different weights inside of them. So, the subtle difference between neural networks and deep learning is around the size and number of weights we have to train.
Generative AI and large language models
Generative AI (or GenAI) and large language models (LLMs) are basically synonyms for the same concept. LLMs usually apply to some sort of language, such as looking at the previous 500 words in a string to predict the next word.
You can technically take these same attention-based transformer models, these GenAI models, and apply them to other use cases. For example, we can use them to truly translate text to search SQL, elastic search. This will enable customers to explore their network data in databases and replace some of their BI tools.
As announced at last year’s Mobility Field Day, Juniper is integrating GenAI and LLMs into our Marvis® Virtual Network Assistant (VNA). The first phase of this is focused on making it much easier for customers and support teams to get specific answers out of hundreds of thousands of public documents.
It’s just one of the many ways Juniper is harnessing the power of AI to help IT enterprises be more efficient and productive every day. To learn more, be sure to check out all the engaging presentations from Juniper’s AI-Native NOW virtual event, which are now available on demand.
