Unlock the secrets of ultra-efficient AI with Julian Musman from ETH Zurich, as we explore the cutting-edge world of smart glasses. Discover how Julian and his team have mastered the art of deploying powerful AI algorithms on the tiniest of devices, tackling the monumental challenge of performing object detection on microcontrollers. Tainissimo YOLO, a breakthrough network, is not just a marvel of engineering but a testament to the balance between computational power, network size, and battery life, revolutionizing the way smart glasses operate without relying on the cloud.

Join us as we unpack the hardware innovations behind these intelligent glasses, featuring components like the NRF53 microcontroller and the GAP9 neural engine, meticulously designed for energy efficiency and high-speed performance. With an astonishing end-to-end latency of 18 frames per second, these glasses promise all-day functionality on a single charge. Julian sheds light on the intricate harmony of PCB design and neural network architecture and explores the potential of advanced convolutional changes, all while emphasizing the strategic choice to remain loyal to traditional networks for unmatched performance with hardware accelerators. Get ready for an insightful journey into the future of AI-powered wearable technology!

Send us a text

Support the show

Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

Curious about how you can run a colossal 405 billion parameter model on a device with a mere 2 billion footprint? Join us with Mahesh Yadav from Google, as he shares his journey from developing small devices to working with massive language models. Mahesh reveals the groundbreaking possibilities of operating large models on minimal hardware, making internet-free, edge AI a reality even on devices as small as a smartwatch. This eye-opening discussion is packed with insights into the future of AI and edge computing that you don't want to miss.

Explore the strategic shifts by tech giants in the language model arena with Mahesh and our hosts. We dissect Microsoft's investment in OpenAI’s Phi model and Google's development of Gamma, exploring how increasing the parameters in large language models leads to emergent behaviors like logical reasoning and translation. Delving into the technical and financial implications of these advancements, we also address privacy concerns and the critical need for cost-effective model optimization in enterprise environments handling sensitive data.

Advancements in edge AI training take center stage as Mahesh unpacks the latest techniques for model size reduction. Learn about synthetic data generation and the use of quantization, pruning, and distillation to shrink models without losing accuracy. Mahesh also highlights practical applications of small language models in enterprise settings, from contract management to sentiment analysis, and discusses the challenges of deploying these models on edge devices. Tune in to discover cutting-edge strategies for model compression and adaptation, and how startups are leveraging base models with specialized adapters to revolutionize the AI landscape.

Send us a text

Support the show

Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

Imagine a world where satellites can make real-time decisions without waiting for ground-based commands. Learn how edge AI is revolutionizing space technology with David Maloney from Ubotica, who brings unparalleled expertise from his time at Movidius and Intel. Uncover the transformative power of performing AI inference directly on satellites, and why traditional ground-based systems are no longer sufficient. We explore advanced imaging techniques like hyperspectral imaging, which can now detect forest fires or monitor environmental changes with unprecedented speed and accuracy.

Delve into the fascinating intersection of image compression and generative AI, and see how these technologies can overcome the challenges of power consumption and energy efficiency in space. Hear about real-world implementations such as the FISAT-1 satellite's cloud detection and removal capabilities and the impressive radiation tolerance of hardware tested at CERN's Large Hadron Collider. By examining the trade-offs and legacies of space technology, we reveal how cutting-edge AI applications are optimizing data transmission and processing.

Finally, we discuss the advanced applications of hyperspectral sensors and image pattern recognition in environmental monitoring and security. Learn how these technologies can assess vegetation health, evaluate soil nutrients, and even detect illegal activities. Discover the challenges of deploying sensors in the harsh conditions of space and the significance of early warning systems for space weather events. With insights into the European Space Agency's Vigil mission and real-time edge processing in space, we also cover the exciting potential of Tiny Train technology for on-board training and updates, ensuring that our satellites are smarter and more responsive than ever before.

Send us a text

Support the show

Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

Ready to unlock the secrets behind real-time sensing algorithms for embedded devices? In this episode, you’ll gain valuable insights into transforming raw sensor data into actionable intelligence. We begin by breaking down the complexities of signal preprocessing, event triggering mechanisms, and feature extraction. Learn how event detection can drastically reduce false positives and computational overhead, setting the stage for robust and efficient real-time systems.

Next, we delve into the world of AutoML tailored for embedded devices. Discover the nuances of hyperparameter tuning, cross-fold validation, and model ranking based on key performance metrics like F1 score and accuracy. We also introduce you to Piccolo AI, an open-source game-changer for model building and sensor data management. Plus, we answer listener questions about practical implementation details, making this segment a must-listen for any embedded systems enthusiast.

Finally, get hands-on with Piccolo AI as we guide you through setting up your environment using Docker and exploring its powerful web interface. From adding new feature extractors to using synthetic data for model validation, we cover everything you need to contribute to or leverage this open-source project. We also emphasize community engagement, highlighting how collaborative efforts can drive innovation and improve the versatility of machine learning in embedded systems. Join us and become part of a thriving community pushing the boundaries of what's possible.

Send us a text

Support the show

Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

Can machine learning models be both powerful and tiny? Join us in this episode of TinyML Talks, where we uncover groundbreaking techniques for making machine learning more efficient through high-level synthesis. We sit down with Russell Clayne, Technical Director at Siemens EDA, who guides us through the intricate process of pruning convolutional and deep neural networks. Discover how post-training quantization and quantization-aware training can trim down models without sacrificing performance, making them perfect for custom hardware accelerators like FPGAs and ASICs.

From there, we dive into a practical case study involving an MNIST-based network. Russell demonstrates how sensitivity analysis, network pruning, and quantization can significantly reduce neural network size while maintaining accuracy. Learn why fixed-point arithmetic is superior to floating-point in custom hardware, and how leading research from MIT and industry advancements are revolutionizing automated network optimization and model compression. You'll gain insights into how these techniques are not just theoretical but are being applied in real-world scenarios to save area and energy consumption.

Finally, explore the collaborative efforts between Siemens, Columbia University, and Global Foundries in a wake word analysis project. Russell explains how transitioning to hardware accelerators via high-level synthesis (HLS) tools can yield substantial performance improvements and energy savings. Understand the practicalities of using algorithmic C data types and Python-to-RTL tools to optimize ML workflows. Whether it's quantization-aware training, data movement optimization, or the fine details of using HLS libraries, this episode is packed with actionable insights for streamlining your machine learning models.

Send us a text

Support the show

Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

Unlock the secrets to high-efficiency embedded computer vision with industry experts Sandeep Mistry from Arm and Alex Chang from Himax Technology. We promise insights into the TinyML community and the collaborative efforts that are setting new benchmarks in the field. Alex Chang shares an in-depth look at Himax Technology’s cutting-edge AI processors and always-on sensors designed for low-power, battery-operated devices. Learn how Himax’s dual-mode CMOS sensors are revolutionizing applications like doorbells and smart devices, and why efficient AI processing at the edge is increasingly crucial.

Curious about the architecture of state-of-the-art AI processors? Discover the intricate details of a sophisticated Wi-Fi processor designed specifically for AI applications. We explore groundbreaking features such as DC-DC converters, dynamic voltage and frequency scaling (DVFS), and advanced security measures like Physical Unclonable Functions (PUF) and qubit-toe accelerators. You'll also understand how specialized hardware accelerators for motion detection, JPEG decoding, and voice detection enhance comprehensive audio and video AI processing. The development ecosystem, complete with essential software tools and downloadable resources on GitHub, is laid out to ease your implementation process.

Ready to see AI in action? Follow our guided process of porting a pose estimation model to the iMac's YSI2 processor, aimed at tasks like fall detection and human-machine interfaces. We detail how to use a modified version of the UltraLytics GLLv8 postmodel for better accuracy and how to optimize the model for the ARM Ethos U55 MPU using the Ethos U Vella compiler. Witness a live demo showcasing the model’s real-time performance, achieving over 10 frames per second. We wrap up with a discussion on revolutionary partnerships driving Edge AI forward, spotlighting contributions from companies like Syntient, Atreus, ARM, and many others. Don't miss this chance to stay at the forefront of embedded computer vision and Edge AI technology!

Send us a text

Support the show

Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org

Unlock the transformative potential of AI in healthcare with Jim Wang and Steve Yu from the University of Sydney and listen to their revelatory insights on TinyML Network models in medical devices. Our latest episode is a treasure trove of innovation, where machine learning meets wearable technology to redefine patient care. From the pressing need for low latency in critical health monitoring to the intricacies of biosensor evolution and data privacy, this conversation maps out the future of medical diagnostics and the promise of personalized treatments.

Imagine a world where your medical device knows you better than you know yourself. That's the future Jim and Steve unveil as we scrutinize the optimization of machine learning algorithms for edge devices, balancing the scales between performance and power constraints. We explore the regulatory hurdles and breakthroughs in model tailoring, ensuring that your health is monitored with meticulous precision. Our discussion also peels back the layers on the importance of public datasets and hospital partnerships in enhancing the accuracy and responsiveness of life-saving technology.

Stepping into the realm of neuromorphic computing, we navigate the promising landscape of on-device training and model compactness. With an eye on patient privacy, we delve into the adaptability of models like the S4D and NCP, overcoming the challenges of limited data and simplifying complex architectures. Jim and Steve's research opens up a dialogue on the future of medical devices, where health monitoring is not just personalized but also proactive, ensuring that the care you receive is as unique as your heartbeat. Join us for an enlightening journey into the heart of healthcare innovation where edge computing is not just smart—it's genius.

Send us a text

Support the show

Learn more about the EDGE AI FOUNDATION - edgeaifoundation.org