The Application Prospects and Future Trends of Edge AI
According to a report by Electronic Enthusiasts (by Li Wanwan), due to high computational requirements, early AI started from cloud intelligence, requiring data to be uploaded to the cloud for processing. However, in the subsequent development process, issues related to user experience and data privacy emerged. Edge AI can significantly reduce latency issues and has more relaxed requirements for network environments, greatly enhancing user experience.From a technological development perspective, the iteration of chip and software technology has to some extent driven the development of edge AI. In the past, the computational power of chips could not meet the demands of edge AI applications, and software configurations typically utilized expert systems or basic machine learning systems to implement AI functions. However, with the development of deep learning software and the popularity of high-performance, low-power edge processors, edge AI has also seen better development in terms of technical support.
Wide Application Fields of Edge AI
Edge AI operates on edge devices, specifically in many scenarios that require large amounts of data and high computational real-time performance, without needing to upload data to the cloud for computation, such as intelligent driving, smart factories, and traffic management integrated with security. Edge AI applications can also be understood as within a certain range, such as a car, a train, a factory, or a store. Within this range, some real-time AI decision-making and processing needs are met.Compared to cloud-based AI, edge AI offers higher data security, lower power consumption, shorter latency, higher reliability, and lower bandwidth requirements, while maximizing data utilization and further reducing data processing costs.The application fields of edge AI are very broad, including smart homes, intelligent transportation, restaurant delivery robots, new retail applications, AR/VR/metaverse, robot programming, smart industry/logistics/finance, etc. Especially in edge AI vision, commonly used technologies include image classification, object detection, semantic segmentation, and instance segmentation.Currently, although cloud is still the main segment of AI chip applications, AI chips are developing from the cloud to the edge. Data shows that by 2025, the revenue of the edge AI chip market is expected to reach $12.2 billion, while the revenue of the cloud AI chip market will reach $11.9 billion, indicating that the edge AI chip market will surpass the cloud.IBM previously stated in a study that 94% of surveyed executives indicated that their enterprises would deploy edge computing within the next five years. From smart hospitals and smart cities to unmanned stores and autonomous vehicles, society now needs edge AI more than ever. Logistics issues, labor shortages, inflation, and the uncertainties caused by the pandemic are troubling businesses. Edge AI can serve as a bridge between humans and machines, enabling improvements in prediction, worker allocation, product design, and logistics.
Future Trends of Edge AI Development
Previously, institutions predicted that in the coming years, the growth of edge AI would mainly stem from the Internet of Things (IoT) and the demand for faster computing speeds due to 5G network coverage, as well as other factors. So what are the future trends of edge AI?The integration of edge and industrial IoT solutions is a trend, with smart factories being a field driven by edge AI applications. Gartner previously pointed out in a report that by 2027, machine learning in the form of deep learning will be integrated into over 65% of edge use cases, a significant increase from less than 10% in 2021.NVIDIA previously mentioned that factories could add AI applications to cameras and other sensors for detection and predictive maintenance. But detection is just the first step; once a problem is identified, action must be taken. AI applications can detect anomalies or defects and then alert humans for intervention. However, for safety applications and other use cases requiring immediate action, simply connecting AI inference applications to IoT platforms managing assembly lines, robotic arms, or pick-and-place machines can achieve real-time responses.The integration between these applications relies on custom development work. Therefore, it is expected that more cooperative relationships will be established between AI and traditional IoT management platforms to simplify the adoption of edge AI in industrial environments.Additionally, the number of enterprises adopting AI-on-5G applications will continue to increase. AI-on-5G combined computing infrastructure provides a secure, high-performance connectivity structure that can integrate sensors, computing platforms, and AI applications from the field, local, or cloud. Its main advantages include ultra-low latency in non-wired environments, guaranteed service quality, and higher security.NVIDIA stated that AI-on-5G will unlock new edge AI use cases, such as Industrial 4.0: factory automation, factory robots, monitoring and inspection; automotive systems: toll roads and vehicle telemetry applications; smart spaces: retail, smart cities, and supply chain applications. One of the world’s first full-stack AI-on-5G platforms, Mavenir Edge AI, was launched in November 2021. More full-stack solutions are expected to emerge in the future, providing enterprise 5G environment performance, management, and scalability.Although edge AI has developed rapidly in recent years, the diversification of its application scenarios has brought many challenges to its implementation, such as the need for customized algorithms for different application scenarios, varying requirements for chip computing power and power consumption, and the disconnection between computing power, algorithms, and applications, leading to a lack of overall solutions for edge AI.In the future, innovative solutions to these implementation challenges will gradually emerge. Lin Ming, director of products and markets at NXP Semiconductors, previously mentioned that due to the diversity of edge AI application scenarios, it is difficult to use a universal processor to handle all AI application scenarios, so heterogeneous computing architecture will be an important trend in the future development of edge AI, using the most suitable processing units to handle corresponding AI tasks.Additionally, AI application scenarios typically involve a separate security chip for key management and secure boot management. To reduce power consumption and costs, future edge AI will trend towards integrated security functionalities, such as NXP’s widespread integration of the EdgeLock (secure authentication chip) module in edge processors.More importantly, the implementation of edge AI products involves the integration of different industrial sectors. In the future, the ecosystem of edge AI will undoubtedly require collaboration among chip suppliers, algorithm suppliers, equipment manufacturers, system integrators, and even cloud service providers to provide professional services.
Disclaimer: This article is original from Electronic Enthusiasts, please indicate the above source when reprinting. For group communication, please add WeChat elecfans999, for submission, interview requests, please email [email protected].More Hot Articles to Read
Dong Mingzhu: Gree Supplies Chassis Equipment for Tesla, Do White Goods Companies Reach a Consensus on “Making Cars”?
Russia Produces First Purely Domestic Communication Satellite! How Strong is Russia’s Satellite Capability?
EU Unifies Charging Interface! Apple: Confirmed to Switch to USB-C!
Marvell Significantly Cuts Back Chinese R&D Team, but Another Market is Increasing Investment in China
Beware! Texas Instruments: Chip Demand Weakens, Spreading from Consumer Electronics to Industrial Sector!
