NVIDIA GTC 2021 was packed with breakthroughs in AI, data center, graphics, accelerated computing, intelligent networking, and more. Every year, GTC congregates NVIDIA’s major announcements and points out the most relevant innovations for the primary industries. This time, we were part of the event as speakers, introducing MaskCam, a smart camera based around NVIDIAs Jetson Nano. We also participated in different sessions during the five-day conference.
GTC serves different audiences. Your context and business interest determine the focus of the talks that benefit you the most. We focused on some interesting AI applications and results rather than centering so much on the hardware aspect of the conference (which is central and high-level). So, please help us by adding some exciting happenings that we might have skipped in this selection.
Forecasting Improvement in Retail
Paul Hendricks shared some of the best practices of using NVIDIA’s RAPIDS Data Science Library (2021). Retailers like Walmart and Dominos have already seen an increase in their forecasting accuracy and reduced logistics costs related to inventory management by accelerating the Machine Learning training process.
According to Paul, accelerated Machine Learning models in RAPIDS provide flexibility to perform faster hyperparameter optimization, and therefore perform more experiments in less time achieving significant accuracy improvements.
RAPIDS library seems to be a promising tool to power our demand forecasting, price optimization, and out-of-stock prediction solutions when talking about millions of item-by-store combinations.
Increasing efficiency with video analytics in Transport & Logistics
Christiaan Hen and Tim Toerber showcased how video analytics improve the operational efficiency of airports. Tracking and object detection tools enhanced situational awareness and increased operational efficiencies while providing a safer work environment for airlines and improving overall airport operations.
They showed several use cases that are relevant for the industry, such as:
- Provide real-time turnaround alerts
- Calculate the time an airplane is holding for gates
- Identify vehicular movement hot spots
- Deliver real-time safety alerts
- Support decision making for improved flow control
This talk is a perfect example of how AI is disrupting the transport and logistics industry, providing real-time data to optimize processes. Kudos to SEA-TAC and Assaia teams for the great talk and demo.
AI Trust for the Financial Sector
Trust is important, and regulatory requirements are high for the Financial Sector as they have always been—even before AI existed. Now with many applications of Machine Learning, this is taken to another level. To demonstrate the trustworthiness of AI in an industry that has been historically more conservative than others seems to be a deal-breaker.
Jochen Papenbrock shared some techniques and conclusions on how to create a trustworthy model, given some examples from the Financial Sector:
- Post-modeling layers helps interpret and explain the model and data
- Synthetic generation of stress datasets to challenge the model
- These are often computationally intensive techniques that require a GPU-based platform
MaskCam: an Open Source Smart Camera to Detect Face Masks
We have been proudly sharing our MaskCam process since the release of its open-source code last March. This project was a prototype reference design for a Jetson Nano-based smart camera system to detect face mask usage in real-time.
This time, Braulio Ríos from Tryolabs and Evan Juras of the BDTI team went further on the development process from the adaptations made through hardware details to our experience working with the NVIDIA ecosystem.
We are glad to keep showcasing this solution funded by NVIDIA, allowing an absolute AI on the edge tool —MaskCam has real-time Computer Vision on the edge and IoT capabilities for remote communication.
🍷 Dinner with Strangers
This year, we participated in several Dinners With Strangers, getting to know awesome people working in cutting-edge AI applications. Indeed, we gained exciting industry takeaways from each of these meetings. Still, the most important takeaway is that NVIDIA teams and the experts who hosted these events were open to obtaining feedback, understanding user’s needs, and profoundly discussing their projects.
🍷 Art and AI
Amidst the current excitement and controversy around NFT’s, NVIDIA introduced the AI Art Gallery as a highlight of GTC 2021. This gallery exposed artists using AI as a tool for their pieces: visual art, music, and poetry. From art sessions such as a live reading of generated poetry to a reversion of AI-generated soul music, this was an amazing approach to AI artist’s processes.
We were part of the Dinner with Strangers, led by Pindar Van Arman and Daniel Ambrosi, about Digital to Physical Art. These artists, who also showed their work at the GTC AI Art Gallery, shared their thoughts about why they still choose to bring their art to the physical aspect even though the initial creation is 100% digital
This session brought several exchanges from random thoughts about Sophia’s Art to the creations of Somnium Space and its goals.
🍷 Omniverse in Manufacturing
Omniverse Enterprise was one of the big announcements of GTC 2021. This collaborative platform will be available in the next few months and allows teams of any scale to simulate and transform complex design workflows in real-time.
At this session, Mike Geyer, Hue Davis, and Jeff Kember from NVIDIA’s team briefly introduced the Omniverse Platform and shared that the primary inspiration for this tool was for autonomous vehicles and robotics.
They showed how this platform could be used in manufacturing to prototype the final product, meaning a massive compression for time to market. Certainly, this is a fantastic tool for industrial and mechanical engineers. They demonstrated “live” some use cases—prototyping a coffee machine, a Volvo car, and a Robot’s arm.
🍷 NLP for Healthcare
Christopher Parisien, a Senior Deep Learning Applied Scientist at NVIDIA, hosted this event. Even though this demonstration was directed at the Healthcare industry, these learnings applied to many other industries or had vertical applications.
He shared these NVIDIA resources for speech recognition and natural language processing:
- Transfer Learning Toolkit: This toolkit reduces costs associated with large-scale data collection and labeling and eliminates the burden of training for AI/ML models ground up.
- NeMo: This is a toolkit for creating Conversational AI applications. It has extendable collections of pre-built modules and ready-to-use models for ASR, NLP, and TTS.
- Jarvis: This GPU-accelerated framework builds conversational AI applications. It includes pre-trained models and services for highly optimized deployments.
Cheers for GTC 2021!
This event was up to our expectations—great speakers, Demos, networking spaces, and general functioning of an online event. We were looking forward to unlocking more exciting projects around NVIDIA’s hardware capabilities. Please leave us your GTC 2021 favorite moments in the comment section. We will be happy to dig in a little bit more! See you next year.
Like what you read?
Subscribe to our newsletter and get updates on Deep Learning, NLP, Computer Vision & Python.