Welcome everyone to my presentation on, H computing and Tiny ML Intelligence at the ferry. Today we'll embark on a journey exploring the frontier of this, transformation technologies like, that are reshaping our digital landscape. This complementary innovations or fun changing how intelligent systems operate by bringing competition power directly to where data. For enabling real time analytics, even in the most resource constrained environments. What is edge computing? Edge computing is a proximity focused processing approach where computation occurs directly at or near data generation points. This dramatically reduces the need to transfer information to distant, centralized servers and back again. It features a low latency architecture. Enabling near, instantaneous response times measured in milliseconds, which is essential or time since to applications like, autonomous vehicles, industry safety systems, and real time analytics. Additionally, edge computing distributes computing power across the network edge, ensuring operation continuity during connectivity disruptions, and enhancing data privacy by processing sense to information locally. So let's, look at the practical applications of H computing in, autonomous vehicles, self-driving cars process estimates of sensor data in milliseconds. H computing enables, critical split second edition, making locally, ensuring safety and, responsiveness regarding of, regardless of the code, connectivity in, smart factories, industrial. Equipment leverages edge processing for continuous real time monitoring and analysis. This technology identifies subtle anomalies, predicts potential, failures, and trigger, provincial maintenance without relaying on distance. Centralized systems in healthcare monitoring advanced variable devices analyze complex patient vital signs directly on the device. They intelligently process health data locally. Preserving privacy and battery life while only transmitting, critical alerts when potentially life-threatening pattern image, And, introduction to Tiny ml. Tiny ml. A all throw low power, technology that consumes mirror, micro watts, tots of power enabling months or years of operations on small. Batteries or energy harvesting systems. It is ideal for long-term deployments in, remote environments. Tiny ML dramatically compresses, neural networks to function within just a kilobyte of memory through advanced techniques like, quantization, pruning and, knowledge legislations. It runs on microcontrollers with, severely constrained resources, one device interface process data, and, Make intelligent decisions directly at the source without, cloud dependencies, eliminating latency, enhancing privacy, and ensuring functionality even in disconnected environments. the technical foundation of, tiny ml, right? So the technical foundation of ML involves several key components. First, modern optimization techniques like quantization, pruning, and, Knowledge distillations are used to compress neural networks. This methods, reduce numerical ions from 32 bit to eight bit or lower while maintaining critical accuracy thresholds for deployment. Second, specialized frameworks like tensor floor light for microcontrollers. These, frameworks enable seamless, deployment on several cons constrained hardware with a little at. 2 56 KB of flash memory. Third hardware acceleration. In latest generation microcontrollers units integrates dedicated ML acceleration hardware. These purpose build silicone employment dramatically employ interference speed by up to 10 times while simultaneously reducing more conception to the micro world range. And, tiny ML development workflow, the tiny ML development. Workflow consists of four main steps. First, data collection involves gathering representative sensor data that captures all essential edge cases. This meticulously labeled dataset creates the foundation for robust, modern, like model performance. Second model training involves a development in shell neural networks using powerful computing infrastructure. Start with full precision, architecture before beginning the optimization, gen third optimization. Transform model through strategic quantitation pruning and knowledge distillations. This, critical phase reduces modern requirement like, memory requirement, to me kilobytes while maintaining, functional, accuracy. the fourth deployment integrates optimization models into target, microcontrollers, and verify real world performance, carefully balanced interference, speed, power consumption, and accurate. C for product prediction readiness and the synergy. Okay. Edge computing and tiny ml combining edge computing and tiny ml offers several benefits. Enhanced privacy is achieved since to data remains on device. Eliminating cloud secu, security vulnerabilities and ensuring regulatory complaints reduced latency is another advantage with sub millisecond re. Once time enabling real time applications critical for autonomous systems and time sense to monitoring lower bandwidth is achieved as pre-processor insights. Reduced network traffic by up to 90%, optimizing connectivity costs and infrastructure requirements. Finally, energy efficiencies enhanced through specialized hardware, accelerations, and optimized models extending battery life or harvest two months for I OT deployments. Applic. Applications across industries. Each computing and tiny ml have applications across various industries. In healthcare, intelligent monitoring device detects, critical patient anomalies in, real time on device algorithms enable instantaneous, failed detection and lifesaving, recognition without cloud connectivity in agriculture, precision field sensors continuously analyze soil moisture, nutrition levels, and Camera conditions. Automated irrigation systems dramatically responds to environmental changes, optimizing water usage and crop fields In manufacturing, advanced equipment monitors, utilize vibration and acoustic signatures to identify subtle failure patterns. Data-driven pre predictive maintenance algorithms prevent catastrophic breakdowns, reducing downturn by up to 70% in the consumer sector. Sophisticated variables recognize complex activities and health patterns with medical grade accuracy, energy efficient voice interference, understand natural language commands, and maintaining privacy by processing all data locally. So with this, we have implementation challenges as well, implementing each computing and time comes with several challenges. Resource constraints are significant issue with extreme limitations in memory and processing power and energy capacity. Restricting model complexity and functionality model accuracy is another challenge as balancing performance State ops while maintaining acceptable interference accuracy during aggressive optimization process can be difficult. Development complexity requires, specialized expertise in embedded systems Modern. Magician and hardware specific implementation techniques. Security concerns are also prevent as v vulnerable H device devices faces increase risk where ADV attacks and modern theft and privacy breaches requiring robust protection mechanism. So let's compare the technologies. So we have cloud computing, edge computing, and tiny. Cloud computing process data in remote data centers with latency ranging from a hundred milliseconds, two seconds. High power requirements in kilowatts, constant connectivity needed, and a typical memory in gigabytes or more each computing process, data in local gateways or servers with latencies ranging from 10 to hundred milliseconds. Moderate power requirements in wats, intermediate connectivity needed and difficult memory in megabytes. Tiny ML process data. On end devices with latency ranging from one to 10 milliseconds, low power requirements in milliwatts, minimal or no connectivity needed, and typical memory in kilobytes. And let's see the future of the edge, intelligence here. So it includes several exciting trends. Federated learning is a distributed training architecture that enable devices to collectively improve models while presuming data. Privacy and sovereignty. Neuromorphic computing involves biologically inspired process architectures that mimic neurostructures for unprecedented energy efficiency. In ML workloads, energy harvesting involves autonomous ml. Systems that capture ambient energy from surroundings enable perpetual influence without battery replacements. Tiny transformations are radically compressed attention based. That bring sophisticated language understanding to resource constraint. Micro, thank you. Thank you for your attention. I hope you found this, presentation on each computing and tiny email inform me too and insightful. If you have any questions, feel free to reach out to me. Thank you.