Machine unlearning, driven by privacy regulations and the "right to be forgotten", is increasingly needed at the edge, yet server-centric or retraining-heavy methods are impractical under tight computation and energy budgets. We present FiCABU (Fisher-based Context-Adaptive Balanced Unlearning), a software-hardware co-design that brings unlearning to edge AI processors. FiCABU combines (i) Context-Adaptive Unlearning, which begins edits from back-end layers and halts once the target forgetting is reached, with (ii) Balanced Dampening, which scales dampening strength by depth to preserve retain accuracy. These methods are realized in a full RTL design of a RISC-V edge AI processor that integrates two lightweight IPs for Fisher estimation and dampening into a GEMM-centric streaming pipeline, validated on an FPGA prototype and synthesized in 45 nm for power analysis. Across CIFAR-20 and PinsFaceRecognition with ResNet-18 and ViT, FiCABU achieves random-guess forget accuracy while matching the retraining-free Selective Synaptic Dampening (SSD) baseline on retain accuracy, reducing computation by up to 87.52 percent (ResNet-18) and 71.03 percent (ViT). On the INT8 hardware prototype, FiCABU further improves retain preservation and reduces energy to 6.48 percent (CIFAR-20) and 0.13 percent (PinsFaceRecognition) of the SSD baseline. In sum, FiCABU demonstrates that back-end-first, depth-aware unlearning can be made both practical and efficient for resource-constrained edge AI devices.

Together, Aetina and Hailo are launching multi-inference AI solutions that use 4x Hailo-8TM AI accelerators in the AI-MXM-H84A MXM, Aetina's AI inference platform (AIP-SQ67), and object recognition AI models. The AIP-SQ67 platform, powered by Aetina's AI-MXM-H84A MXM module, offers enough processing power to enable real-time video analytics processing and numerous, low-latency AI inference tasks at the edge, with up to 104 Tera-Operations Per Second (TOPS) of AI performance from Hailo's AI processors. The AI technologies are appropriate for a variety of applications in cities and transportation networks since they are capable of identifying diverse objects, such as people and vehicles, and evaluating large video datasets from multiple cameras simultaneously. Aetina and Hailo will present the solutions at ISC West 2023. The MegaEdge family member AIP-SQ67 from Aetina boasts an Intel 12th Gen CoreTM processor and expansion slots for up to two M.2 AI accelerators and one MXM.

As 2023 progresses, demand for AI-powered devices continues growing, driving new opportunities and challenges for businesses and developers. Technology advancements will make it possible to run more AI models on edge devices, delivering real-time results without cloud reliance. Edge AI technology has proven its value and we can expect to see further widespread adoption in 2023 and beyond. Companies will continue to invest in edge AI to improve their operations, enhance products (i.e., safer, additional features) and gain competitive advantages. AI's adoption will also be driven by innovative applications such as ChatGPT, generative AI models (e.g., avatars) and other state-of-the art AI models that will be used for applications in medtech, industrial safety and security.

The arrival of artificial intelligence (AI) in embedded computing has led to a proliferation of potential solutions that aim to deliver the high performance required to perform neural-network inferencing on streaming video at high rates. Though many benchmarks such as the ImageNet challenge work at comparatively low resolutions and can therefore be handled by many embedded-AI solutions, real-world applications in retail, medicine, security, and industrial control call for the ability to handle video frames and images at resolutions up to 4kp60 and beyond. Scalability is vital and not always an option with system-on-chip (SoC) platforms that provide a fixed combination of host processor and neural accelerator. Though they often provide a means of evaluating the performance of different forms of neural network during prototyping, such all-in-one implementations lack the granularity and scalability that real-world systems often need. In this case, industrial-grade AI applications benefit from a more balanced architecture where a combination of heterogeneous processors (e.g., CPUs, GPUs) and accelerators cooperate in an integrated pipeline to not just perform inferencing on raw video frames but take advantage of pre- and post-processing to improve overall results or handle format conversion to be able to deal with multiple cameras and sensor types.

Since the last decade, it is becoming increasingly clear that Artificial Intelligence is going to be the central technology around which all other advanced technologies will revolve and co-exist. Artificial Intelligence, in simplest terms, refers to'smart' machines or'intelligent' systems which perform tasks that are usually associated with human intelligence and reasoning. Artificial Intelligence (AI), in order to maximize its utility, has been paired with robotics, machine learning, Internet of Things (IoT), and many such technologies. Edge AI computing is one such example of integration of two technologies. As said earlier, edge AI is the combination of two technologies, viz., Edge computing and Artificial Intelligence.

Edge AI addresses the processing and the implementation of machine learning algorithms locally on the hardware systems. This form of local computing reduces the network delay for data transfer and solves the security challenges as everything happens on the device itself. This diagram that appears above summarizes all the processes of the Edge AI. Edge AI's local processing doesn't mean that the training of the ML models should happen locally. Generally, the training takes place on a platform with a greater computational capacity to process a larger dataset.