Just last week we published our first “Tech Innovators” content Emerging Technologies: Tech Innovators in Edge AI (full report available to subscribing Gartner clients). In this research we profile 12 innovations where technology providers are advancing edge artificial intelligence (AI). This is a result of our emerging technologies and trends case-based research. Typically we reach out to well over a hundred vendors requesting participation. We pre-qualify them and then conduct a series of interviews with the tech provider on their capabilities and marquis implementations. From these efforts we produce a body of content including “Tech Innovators.”
With the Edge-AI case-based research we reached out to over 150 tech providers. As a result we examined over 30 tech providers and over 100 adopters. Here are some of our major findings.
- The manufacturing, media and services, communications, and retail are early adopter industries with edge artificial intelligence (AI).
- Model optimization advancements are critical to broad and accelerated expansion of edge AI.
- Advances in edge AI also accelerate the adoption of IoT and IoT-enabled products and services.
- Until 2025, edge AI will remain embryonic due to the lack of knowledge and experience related to empowering the “edges”(e.g., embedded, device and local servers).
Here are the 2020 Tech Innovators in Edge AI
Anagog, Atos, Chooch AI, Deci, Deeplite, Kneron, Latent AI, Matroid, Octonion, ONE Tech, Reality AI and Tact.ai (click to enlarge graphic)
Let’s take a quick look into a couple of them.
Deeplite Offers a Software Solution to an Edge AI Hardware Problem
Deeplite is an innovator because its automated AI model optimizer (Neutrino) enables robust DNNs to run on highly resource constrained edge devices with high performance and accuracy. The user provides pretrained PyTorch or TensorFlow models, desired model size and target constraints like acceptable accuracy. Then, using these constraints, Neutrino outputs optimized convolutional neural networks (CNNs) in ONNX, PyTorch or TensorFlow.
Customers can export models optimized with Neutrino to third-party compilers such as CUDA, OpenVino, and TensorRT. That said, for highly constrained computing devices, Deeplite also provides a platform-aware optimizer for mixed precision techniques such as 2-bit or 1-bit networks.
Optimization technology, like Deeplite’s, enables greater AI performance at the edge. This optimization significantly expands the number potential business use cases and so elevates the overall business value and market potential for edge AI.
Kneron Leverages System-on-Chip Innovation to Deliver More AI to the Edge
Kneron is innovating with robust neural processing at the very edge with low power AI “systems on a chip.” This innovation improves the autonomy of edge devices’ by decreasing reliance on cloud computing for advanced analytics and control. By supporting mainstream AI models, Kneron’s technology benefits a wide range of edge use cases across smart home devices, IP cameras, laptops, and driver monitoring systems.
Kneron provides a reconfigurable AI engine that allows Kneron SoCs and NPUs to deconstruct mainstream AI models into basic building blocks for reassembly to support real-time voice and image recognition. Through the ability to build AI models based on needs and resource constraints, Kneron SoCs and NPUs future-proof devices to accommodate future mainstream AI models.
Like Deeplite, Kneron enables greater AI workloads in disconnected environments at the very edge. This significantly increases the real-time business use cases where edge AI can deliver value and so expands its overall market potential.
Matroid Is Democratizing AI for Nontechnical Clients
Matroid is innovating by providing customers with an easy-to-use dashboard for creating, training, and deploying CV models for object, action, or event recognition (i.e., detectors) either on the cloud or at the edge.
The company offers both a device embedded solution where the detectors run on a camera and where the detectors run on a customer workstation in a private data center. These models can also run on the cloud at Matroid.com. Matroid offers Matroid Studio, a developer service which runs on-premises (through the Matroid API) or in the cloud (through the Matroid browser). With Matroid Studio, customers can build, test and deploy CV detectors. These detectors run on Matroid monitoring software (installed on on-premises devices) which supports multiple hardware platforms.
Matroid is focused on bringing more pretrained models to market. This can significantly support the democratization of Edge AI. Business demand is increasing for easy-to-use CV solutions to gain operational insights and improve business operations. Solutions, such as Matroid’s, will make CV more commonplace in the average workplace.
Reality AI Breathes New Life Into Old Processors
Reality AI is innovating with technology that extracts key features from legacy sensor data, automatically builds ML models based on identified features, and then deploys those models on existing (often very resource constrained) edge systems. They target the extreme edge characterized by low cost, low resource processors such as children’s toys, automotive components and building infrastructure and subsystems.
Reality AI offers automated AI feature discovery and optimization from existing sensor data. Their proprietary algorithms search through more than 10,000 different feature combinations to find the set that delivers high model accuracy at lower computational cost. The market impact of innovations like Reality AI is tremendous because they leverage a customer’s existing assets and add significant value on top of substantial prior investments. This greatly expands the market for edge AI and allows for a gradual evolution to full modern edge AI as upgrades extend the serviceable life of existing assets. This is often far more desirable to organizations than a “rip and replace” approach.
This New Research
A big THANK YOU to Eric Goodness and Danielle Casey who conducted the ~9 month case-based research effort on Edge AI. Mr. Goodness is the lead author for the full Edge AI Tech Innovators analysis. His dedication to quality was truly remarkable.
We plan on delivering five “Tech Innovators” articles in 2020 covering a variety of emerging technologies and trends.
The vendors chosen were included based upon Gartner’s opinion of the nature of their innovation against the below qualities.
- Demonstrates uniqueness of technologies and solutions
- Holds potential for significant market impact
- Has feasibility for implementation at scale
- Solves significant, well-known technology or business problems
- Has catalyst potential to stimulate additional innovations
This analysis is specific to the innovations presented. It is not an endorsement of the company or an assessment of the company’s viability as a business, a vendor or a potential partner. Though Gartner is highlighting the innovations of this set of technology providers, we recognize that there are other innovations out there and that this is not an exhaustive list.
As always, your comments/feedback is much appreciated.