Accelerating the Consumer Internet

The past year is certainly one for the books, dominated by a once-in-a-century pandemic that most of us have never experienced at this scale. 

While the brunt of the toll has been met by the frontline workers tasked with saving lives and deploying vaccines, we’ve all been hit hard by social distancing and stay-at-home orders. 

Most of us haven’t been able to go into the office or get together with others. This stark shift in day-to-day routine has caused us to increase our use of technology on so many levels. Here are a few examples: 

• Connecting with coworkers and friends on video conferencing platforms like Zoom, Microsoft Teams, and Google Meet.
• Massive purchasing on e-commerce platforms, such as Amazon, Shopify, eBay, for what we might otherwise have bought in store.
• Using delivery services like DoorDash, UberEats, GrubHub, Instacart to order food, groceries, and beverages in the face of limited in-person dining options. 
• Streaming movies, TV, videos, and music live and on-demand via Netflix, Disney+, HBO Max, and Spotify, with theaters being closed or limited in attendance.
• Connecting with friends and family more than ever on Facebook, LinkedIn, Twitter, TikTok, Clubhouse, and other social media platforms.

The massive growth in new users and engagement levels have been off the chart. To keep up with user demand, the technology behind it has also had to scale at an unprecedented rate. 

Yet, your favorite platforms are able to do more with less, thanks to accelerated AI tools. Using NVIDIA GPUs has allowed the best data science, machine learning, and artificial intelligence teams to handle the largest amounts of data, train on them in the lowest time possible, and do it all at the lowest possible cost.

Tailoring Your Next Streaming Recommendation

Netflix has become one of the most ubiquitous video platforms ever created. It’s now the primary way that most people stream shows and movies in the comfort of their own homes.

The underlying technology that determines what someone will watch next is called a recommender system, the visible queue of Netflix movie titles and corresponding thumbnails that pop up using complex and compute-intensive algorithms based on a user’s historical viewing history.  

This home screen is driven by an incredibly powerful recommendation engine that analyzes user preferences, ratings and viewer history in order to provide the most relevant and high-quality content to users.

Companies such as Netflix have a robust inference platform that enables them to serve these useful recommendations at scale with high efficiency. 

Recommender systems often use a tree-based model approach. This approach helps discrete user choices similar to a decision tree diagram. NVIDIA's Python Data Science suite, known as RAPIDS, includes a powerful inferencing solution called Forest Inferencing Library (FIL) that is purpose-built to accelerate these kinds of decision tree-based recommender systems on GPUs. 

Accelerating these models allows for additional model complexity which increases recommender accuracy while at the same time often lowering infrastructure costs. The more accurate the recommender model, the more likely a user will watch the most desirable content which improves customer satisfaction and retention.  

Enhancing Video-Powered Social Network

TikTok is one of the biggest hits in the past year, a social network featuring a video-driven feed driving super-high engagement. 

The secret behind TikTok's success lies in its ability to perform computer vision recognition on every video uploaded onto its platform. When a machine recognizes what’s happening in every video, it allows for the most targeted and relevant content to appear in front of users which in turn creates a higher customer affinity to the platform.

However, computer vision is a very compute-intensive task and it can be an effectively impossible task without the right computing tools. TikTok leveraged NVIDIA T4 GPUs to perform computer vision at scale driving their business to unprecedented levels through user growth and increased engagement.

AI Insights from Consumer Internet Leaders

To share insights into the latest AI breakthroughs, NVIDIA's GTC 2021 will bring together data science and machine learning experts, including from some of the world's leading Consumer Internet companies.

Check out the latest developments in topics such as accelerated analytics, deep learning, natural language processing, computer vision featuring experts from across the industry, including:

• LinkedIn: Hear from Weiwei Guo and Huiji Gao, natural language processing leaders, about how they built a framework for deep natural language processing called DeText.
• Spotify: Catch Joseph Cauteruccio and Marc Romeyn, machine learning engineers, on how they were able to accelerate their model evaluation process from hours to minutes using RAPIDS and NVIDIA GPUs.
• GrubHub: See how data scientist Alex Egg was able to build a better-personalized recommendation framework for their delivery platform.
• SalesForce: Discover how Nitish Keskar, senior research scientist, and Shashank Harinath, principal machine learning engineer, used NVIDIA's Triton Inference server to enable large-scale NLP models like BERT and GPT2. 
• Square: Understand machine learning engineer Stephane Fotso's journey to transform their product recommendation system with deep learning and BERT.
• Postmates: Watch how Lance Lin and Frank Yang, engineering leads, used machine learning to make an improved personalization and discovery interface for their customers.

In all, there are more than 1,400 sessions to explore across data science, machine learning, deep learning, recommender systems, computer vision, natural language processing, and more topics. GTC will also host a variety of networking opportunities, special events, and other interactive activities.

Registration is free, and GTC will kick off on April 12 with a keynote by NVIDIA CEO and founder Jensen Huang, and run through April 16. I encourage you to attend to learn how AI is transforming the consumer internet.

Register now for free.