Edge computing

Trend 12: Edge AI enhances efficiency and power across industries

Edge AI transforms AI deployment by prioritizing on-device processing, reducing reliance on central servers for instant decisions. This optimization significantly improves latency, response times, security, and bandwidth efficiency — crucial in applications like autonomous vehicles. Edge computing also minimizes data transmission, enhancing privacy in computer vision.

A delicate balance between model complexity and computational efficiency is crucial, especially for compact, portable edge devices like mobile phones, IoT, and drones. Private video analysis through real-time edge AI processing generates less but high-quality data. Edge AI in remote healthcare and manufacturing eliminates the need for human operators, enables 'always-on computer vision,' and empowers firms to avoid sending video streams to the cloud, amplifying its potency.

Edge AI will process inputs from multiple modalities (vision, audio, text) and emerge as a transformative force in our interconnected world. This imminent progression promises heightened potency on the edge, paving the way for more efficient and powerful edge AI models across diverse industries.

A remote power generation entity partnered with Infosys to build a drone system featuring nimble, low-latency computer vision models. Tailored for vast power plants in hard-to-reach terrains, these models seamlessly integrate with drones' edge computing devices. Equipped with cameras and deployed computer vision models, drones deliver real-time data, empowering firefighting teams and authorities with swift, informed decision-making in remote and challenging locations.

Edge computing

Trend 13: Generative AI shapes the future of edge

The next development in edge computing is generative AI use cases. Earlier LLMs like GPT-3.5 or GPT-4 were too big for edge deployment. To optimize performance, smaller models tailored for specific tasks emerged, making generative AI feasible for edge applications such as sentiment analysis, Q&A, and language translation.

With the rise of more powerful LLMs, there is a growing need for full-scale generative AI at the edge. Quadric, a California-based edge AI chip provider, is one such example. It recently unveiled compatibility between its neural processing unit IP core (Chimera) and Meta’s Llama 2 model. Quadric is a key player in the chip provider landscape, particularly due to its support for LLMs. Anticipated proliferation of LLM implementations is driving advancements in mission-critical industries. For instance, autonomous decision-making and suggestions in warehouse environments in healthcare (privacy-preserving medical monitoring) and manufacturing.

In collaboration with Infosys, a manufacturing company developed a generative AI solution on edge for automated review of digitized engineering documents. The solution reads title block information, which helps associate metadata with digitized copies and fastens search and retrieval of relevant drawings from scanned data.