The fourth annual Imagine conference is a wrap and this year’s event set a new benchmark for edge AI. Companies including NVIDIA, Arm, AWS, Capgemini and Nestlé Purina discussed how their organizations are maximizing the technology, as industrial edge AI moves from R&D into production-scale deployments.
With almost 300 engaged attendees present at the Computer Science Museum in Mountain View, CA, and hundreds more joining the event virtually from around the globe, the one-day event featured thought-provoking discussions, groundbreaking innovations (including 17 hands-on edge AI demos from Edge Impulse partners), and a glimpse into the future of intelligent edge computing — proof positive that the field of edge AI is nearing maturity.
(Rewatch all the presentations from the event here.)
Key insights from Imagine thought leaders:
“What my team works on typically takes at least five years till it appears in the market. So the technology my team is defining today goes to our fantastic Silicon partners, and they put that into devices that we then need to make available to developers for about five years in the future. So we're constantly scanning the horizon of what compute needs will exist in the world in 10 years time.”
— Paul Williams, Senior Vice President and General Manager, IoT, Arm
“The promise of edge AI … it's this hyper convergence that we're seeing where silicon has progressed, hardware, connectivity, and then, of course, the models are the things that have really progressed significantly. So all of this together, this hyper convergence, is unlocking a really exciting new opportunity in this space.”
— Jeff Hebert, Executive VP, Global Head of Intelligent Products and Services, Capgemini
“We want to be able to treat patients at home as much as possible and follow them up and monitor them when they're at the hospital. To be able to do this, we'll need the best that sensor technology has to offer. Edge AI enables us to future proof the electronic medical record, to stay ahead and to be prepared for the new types of devices, new things that we can sense. These are not just for blood pressure and weight. I think the future will be ambient sensors that are non intrusive.”
— Eivind Holt, Tech Lead in Research and Innovation, DIPS
Another major highlight of this year’s Imagine was the unveiling of the published report with Manufacturing Dive that shares insights into the growing adoption of edge AI for industrial use cases and the need for more advanced tooling
How Edge AI is Solving Problems
AI is expanding the boundaries of what we can do at the edge, from technologies that transform our health and well being to solutions that increase factory output. So how are companies responding to edge AI and how are they using it to solve some of their biggest challenges?
Nestlé Purina
Nestlé Purina, the pet care company, uses AI to collect data on symptoms exhibited by pets, then maps the data back to potential illnesses or conditions, with the goal of catching these conditions sooner rather than later. One of the products created is a Smart Litter Box Monitor, which tracks the bathroom habits of cats. When an anomaly is detected, the owner is alerted.
UltraHuman
One of the leading ring makers has deployed the Ring AIR, a wearable which monitors sleep, movement, and recovery. With Edge Impulse’s new HR HRV model, UltraHuman was able to go from project start to mass production in two months. The company decreased their code size for that part of the algorithm by 40%, a major plus when squeezing more algorithms into wearable devices.
CarePredict
Specializing in home care and assisted living, CarePredict has created a fall detection device for the elderly. However, they needed to address the occurrence of too many false positives in their existing algorithm. With Edge Impulse’s edge AI capabilities, CarePredict's fall detection algorithm achieved a 65% reduction in false alarms compared to their previous algorithm.
GlobalSense
Known for building smart sensing mechanisms for IoT applications, one of the most exciting edge AI breakthroughs for the company is its intelligent sound detection. For example, a car listening to its environment and being able to detect a collision.
HP Poly
A subsidiary of HP, Poly released Voyager Free 60 (and 60+) earbuds which can now be controlled using voice commands. It is now possible to answer or ignore an incoming call by saying "Answer" or "Ignore" respectively. Using a keyword-detecting AI algorithm, this voice-controlled feature is the culmination of months of collaboration between Poly and Edge Impulse to train, build, and deploy, en masse, not just a novel function that responds to the two keywords, but works with eleven different languages.
Neurable
Imagine an AI-enabled EEG headset that can detect cognitive fatigue. Neurable has produced a wearable CNN referred to as “Fitbit for your brain.” This breakthrough could have major implications for medical professionals or pilots, for instance, people whose jobs depend on their ability to be alert and clear-minded.
Revolutionizing Industrial Edge AI: Groundbreaking Tools Unveiled at Imagine
At Imagine, Edge Impulse unveiled a suite of cutting-edge AI tools to transform industrial environments. These innovations address the unique challenges of industrial settings, where precision, speed, and reliability are paramount. Our new offerings include:
YOLO-Pro: Cutting-edge object detection for the production line
YOLO-Pro, based on the popular YOLO (You Only Look Once) algorithm, is the first model of its kind trained with extensive industrial datasets for optimal performance in real-world edge scenarios. Built from the ground up to outperform existing YOLO architectures, YOLO-Pro is also licensable for all businesses and optimized and available in various sizes to run efficiently on edge devices from CPUs to accelerators to GPUs.
Application Behavior: Tune models for real-world industrial performance
Build end-to-end applications that optimize models in the full context of a deployment against desired business outcomes. This solution empowers customers to optimize the entire machine learning application simultaneously, ensuring peak performance on edge AI devices. Supported applications include sound and speech detection, object tracking and counting, fall detection and many more.
Model Monitoring: Deploy in production confidently on fleets of devices
Continuously monitor and improve edge AI models at scale for optimal performance and rapid issue resolution:
- Deploy, analyze, and debug models on device with your IoT solution of choice
- Enrich and refine your models with data from the field
- Deliver seamless over-the-air (OTA) updates
- Monitor in-production model metrics
AI Actions: GenAI Tooling for Industrial Use Cases Accelerate machine learning development and enhance model performance by building high-quality datasets faster. Leverage large language models (LLMs) to create automated data pipelines for efficient data labeling and automated quality control. Harness the power of generative AI and NVIDIA Omniverse to create and enhance sensor, audio, and vision datasets with synthetic data. Integrate with established GenAI platforms including OpenAI, HuggingFace, ElevenLabs, and custom LLM sources.
These tools are not just incremental improvements; they’re solutions designed to dramatically enhance productivity, reduce downtime, and unlock new levels of operational intelligence in factories and production lines across the globe.
Looking Ahead
What’s the future of edge AI, and what do companies need to be successful? Edge Impulse co-founder and CEO, Zach Shelby, proposed the path forward with these three strategies:
- Empower developers with the tools to get the data they need to build great models, closing a large gap in the industry. This challenge can be solved with a mix of synthetic data sources, like Omniverse, Isaac Sim integrations, and generative AI to build millions of domain specific models.
- Continue with production optimization, the ability to squeeze more algorithms into smaller devices with the silicon and power available.
- Speed up deployment by being able to solve any problem, on any silicon, with any class of algorithm, with any data.
The excitement and energy from Imagine 2024 will continue to drive innovation in the world of edge AI. We’re already planning for next year’s conference, which promises to be even bigger and more impactful.
For those who couldn’t join us or for attendees looking to revisit the experience, recordings of the presentations are available for viewing.
And if Imagine 2024 left you inspired, sign up for the Enterprise Trial and get exclusive access to the latest tools and features to accelerate your edge AI projects.
See you next year!