Podcast
EDGE AI POD
Hosted by EDGE AI FOUNDATION · EN
Discover the cutting-edge world of energy-efficient machine learning, edge AI, hardware accelerators, software algorithms, and real-world use cases with this podcast feed from all things in the world's largest EDGE AI community.
These are shows like EDGE AI Talks, EDGE AI Blueprints as well as EDGE AI FOUNDATION event talks on a range of research, product and business topics.
Join us to stay informed and inspired!
19episodes
Episodes
Newest firstAll episodes
Cows Chewed Our Sensors And Still Taught Us About Edge AI
1w ago00:22:29Tap to summarizeA failed 5G rollout in a legendary forest forced us to rethink everything we knew about AI infrastructure. Instead of pushing data to distant servers, we turned wearables, sensors, and tiny controllers into a cooperative network that can sense, decide, and act without the cloud. The result is a hands-on tour of decentralized AI: how to split models across devices, why feature fusion matters more than raw horsepower, and what it takes to make ad hoc networks reliable in the wild.We walk through practical patterns for collaboration at the edge, from complementary sensing in search-and-rescue to pooled compute in crowded venues. You’ll hear how we orchestrate parallel processing on microcontrollers, assign inference to one core and radio handling to another, and compress features to keep bandwidth low. We also dig into continual learning and federated averaging, outlining strategies to adapt models locally while protecting privacy and avoiding catastrophic forgetting. Along the way, we share early results from agriculture and public safety pilots, plus the gritty realities of hardware constraints, scarce datasets, and the challenge of testing at scale.If you’re curious about TinyML, edge AI, and how generative models might run collaboratively across many small devices, this conversation lays out a practical path forward. You’ll come away with a clearer picture of when decentralization beats centralized cloud systems, which protocols survive in noisy environments, and why the future of AI may look less like a monolith and more like a swarm. Subscribe, share this episode with a builder who loves constraints, and leave a review to tell us where you’d deploy a swarm of tiny models next.Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →How AI Compensates for PID Controller Limitations in Electric Vehicles with STMicroelectronics
2w ago00:16:59Tap to summarizeHow can artificial intelligence transform electric vehicle performance? Discover the groundbreaking application of neural networks to motor control challenges that even Formula 1 legend Michael Schumacher helped identify.The automotive industry's electrification demands increasingly sophisticated silicon solutions, particularly for traction inverters controlling electric motors. Traditional control systems face a fundamental challenge: they must operate at extraordinary speeds (currently 20kHz, trending toward 100kHz) while managing rapid transitions between states. When drivers make sudden accelerator changes, conventional PID controllers produce energy-wasting overshoots that drain precious battery power.Our research presents a novel approach using neural networks to compensate for these limitations. By generating time-varying correction factors, our AI solution reduces maximum overshoots by up to 70% in demanding scenarios. This innovation represents a critical advancement for electric vehicle efficiency, potentially extending range and improving performance.What makes this application particularly fascinating is the extreme time constraints. While most AI applications process data at relatively leisurely rates (think 30 frames per second for vision systems), motor controllers must complete their calculations within microseconds. Our current implementation achieves 70-microsecond inference times on automotive-grade microcontrollers, with further optimizations planned through hardware acceleration.The collaboration between academic researchers and industry partners (MathWorks and STMicroelectronics) demonstrates the power of combining simulation expertise with real-world deployment capabilities. Using Simulink as the development platform and ST's developer cloud for automatic deployment to physical microcontrollers, we've created a streamlined methodology for applying AI to automotive control systems.Want to dive deeper into the technical details? Check out our published research paper on arXiv and discover how neural networks are transforming the heart of electric vehicle propulsion systems. Share your thoughts on how AI might further revolutionize automotive technology!Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →How to simplify and securely maintain up-to-date AI Models in the Edge
3w ago00:20:47Tap to summarizeEver shipped a smart device and worried what happens after it leaves the lab? We dig into the hard parts of edge security—where models live on-device, firmware updates are routine, and attackers treat your fleet as a supply chain—then break them down into moves any team can adopt. From secure boot that blocks untrusted code at power-on to verified boot with discrete secure elements, we show how to anchor trust in hardware so software can prove itself before it runs.We walk through the real risks teams face—model theft, OTA hijacking, plaintext credentials in flash, and silent downgrades—and map them to practices that actually scale across mixed hardware. You’ll hear why encrypting data at rest frustrates drive cloning, how end-to-end encrypted and signed updates prevent tampering, and why automatic rollback turns “bricks” into recoverable hiccups. Updating AI models becomes a strength when you ship small, signed artifacts instead of full images, with logs that satisfy CRA and NIS2 audits while giving operators the visibility they need.We also tackle the build-versus-buy dilemma with clear-eyed math. Building a secure update stack across Qualcomm, NXP, PSoC, and diverse compute modules takes specialists and months; a platform approach spreads cost, speeds delivery, and still lets you own your keys so you can switch later without stranding devices. That key ownership underpins true end-to-end trust: you sign, devices verify, and the infrastructure moves at your pace. If you care about safeguarding IP, maintaining uptime, and earning customer trust, this is your blueprint.If this deep dive helps, follow the show, share it with your hardware and firmware teams, and leave a quick review—what part of your edge stack needs the strongest lock?Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →AI-Driven Brain-Computer Interface (BCI) Unlocking the Minds Potential
4w ago00:15:23Tap to summarizeImagine steering a game or selecting a letter with nothing but a blink or a glance. We set out to make that feel normal, not magical, by building a non-invasive brain–computer interface that runs entirely on a low-power microcontroller and fits into everyday wearables like glasses. No surgery, no cloud dependency—just smart sensing, tight signal processing, and a tiny neural net that turns eye movements into reliable commands.We start with the “why”: millions live with motor impairments yet can still move their eyes, leaving a powerful window for communication and control. From there, we map the BCI landscape—high-precision invasive implants like Neuralink, BrainGate, and Synchron on one side; accessible non-invasive tools like Emotiv, Muse, and OpenBCI on the other—and unpack the trade-offs across accuracy, latency, cost, and ethics. Our approach uses electrostatic charge sensing to read subtle changes around the eyes, with electrodes positioned for comfort and signal quality. A lean pipeline cleans the data with high-pass, notch, and low-pass filters; a Z-score event detector wakes the model only when something meaningful happens.The model is a compact 1D CNN that classifies four classes—discard involuntary blinks, trigger with a voluntary blink, and detect left or right glances—achieving about 90% accuracy on a small multi-participant dataset. Running on an STM32H7, it uses roughly 18 KB flash and 6 KB RAM, with sub-millisecond inference; the overall response is driven by the short data window at 240 Hz, delivering real-time control for basic tasks. We demo blink-to-jump and look-to-steer gameplay to prove responsiveness and highlight how the same system could power communication aids and smart-home control. Looking ahead, we focus on integrating the electrodes into comfortable glasses, adding quick calibration for personal variability, and expanding the command set without sacrificing simplicity.If this mix of accessibility, edge AI, and practical human–machine interaction resonates with you, follow the show, share it with a friend, and leave a review so we can reach more builders and caregivers working on assistive tech. What would you control first with a glance?Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →An Embedded Transformer- base face recognition system in the STM32N6
Apr 3000:11:52Tap to summarizeWhat if transformer-level face recognition could run on a microcontroller without giving up speed or accuracy? We set out to make that real on the STM32N6 by pairing its neural processing unit with a hybrid model that blends convolutional efficiency and attention-like global context. Along the way, we rewired core assumptions about attention, reworked unsupported operators, and delivered a full on-device pipeline that actually feels instant.We start with the hardware edge: ARM Cortex M55, 4 MB of continuous RAM, and an NPU pushing up to 600 GOPS at remarkable power efficiency. That lets us chain models—RetinaFace-style detection with landmarks, alignment for a stable canonical view, MobileNetV2 anti-spoofing to block print and replay attacks, and a final recognizer that outputs a 512‑dimensional embedding. The recognizer is built on EdgeFace, itself based on EdgeNext, chosen for its sweet spot between parameter count and accuracy. It behaves like a transformer where it matters—capturing long-range relationships—yet fits into the tight compute envelope of a microcontroller.The turning point is attention without the dot product. Because the ST toolchain doesn’t support batch matmul, we replaced it with a convolutional self-attention mechanism. Depthwise and pointwise convolutions encode relationships across pixels and channels, a sigmoid stands in for softmax, and element-wise products reconstruct attention’s weighting behavior. This maps cleanly to the NPU, avoids quadratic costs, and preserves the ability to stabilize identities across pose, lighting, and occlusion.Benchmarks show roughly 40 ms per frame end to end—about 25 FPS—plus substantial speedups over STM32H7 and higher accuracy than MobileFaceNet across validation sets. That opens doors for privacy-first access control, frictionless enrollment on-device, and personalized experiences where latency matters and data should never leave the edge. If you’re exploring embedded AI, this walkthrough shows how to align model design with silicon capabilities and deliver results that feel both fast and trustworthy.Enjoy the deep dive? Subscribe, share this episode with a fellow edge AI builder, and leave a quick review to help others find the show.Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →Verification, Validation & Certification of AI in Safety-Critical Applications
Apr 2300:18:39Tap to summarizeA cyclist disappears to the model, not to your eyes—and that mismatch is the heart of safety-critical AI. We open with the “vanishing cyclist” to show how tiny, imperceptible perturbations can flip life-or-death decisions, then walk through a practical path to trust that spans data, verification, and deployment. Along the way, we share real stories from BMW, Airbus, and Madrid Metro to ground the engineering in results, not hype.We break down how to build a resilient pipeline: domain-specific data labeling, realistic synthetic generation for rare and risky scenarios, and tight interoperability across MATLAB, Python, PyTorch, TensorFlow, and ONNX. We dig into explainability beyond classification with D-RISE for object detectors and semantic segmentation, helping you see what the network actually uses to decide. Then we raise the bar with formal verification for robustness—mathematical guarantees within defined perturbation sets—so you aren’t mistaking the absence of found attacks for true safety.Finally, we get practical about the edge. Model compression and projection recover accuracy with fewer parameters, enabling fast, power-efficient deployment to CPUs, GPUs, and FPGAs, backed by code generation for the entire application. We also cover runtime safeguards like out-of-distribution detection to catch smog-on-the-runway moments and escalate safely. Throughout, we connect the work to evolving standards, the EU AI Act, and updated workflows that adapt the V-model for learning systems, so your process and artifacts are ready for audits and certification.If you care about trustworthy AI for cars, planes, rail, and medical devices—and want tools and habits that survive contact with reality—this one’s for you. Listen, subscribe, and leave a review with your biggest trust gap or the safeguard you’d ship first.Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →Aptos: Creating ML models that fit your edge device like a glove
Apr 1600:20:30Tap to summarizeShipping edge AI shouldn’t feel like a marathon through model zoos, missing ops, and latency ceilings. We lay out a practical path to get from your data and constraints to a hardware-ready model—measured on real boards—without the endless back-and-forth between data science and firmware teams. If you’ve wrestled with quantization loss, unsupported kernels, or picking the “right” NPU, this walkthrough will feel like oxygen.We start by naming the pain: quick demos that collapse under real device limits, foundation models that fail after export, and feedback loops that burn months. From there, we unpack Aptos, our automation engine that turns edge AI into a data in, model out process. The system explores parameterized architecture recipes and neural architecture search, trains promising candidates, and deploys them to a hardware farm packed with evaluation kits. Every candidate returns hard numbers—latency, per-layer timing, memory, on-device accuracy, and power—so tradeoffs are grounded in measurements, not wishful thinking.What makes it fast is the learning layer. As Aptos accumulates results, meta models predict runtime, memory fit, and stable hyperparameter ranges before committing compute. That means less time wasted on dead ends and more time converging on models that satisfy your KPIs, whether you care about sub-5 ms inference on an i.MX 8 Plus, battery life in the field, or non-square inputs that match your camera feed. We also fold in research-backed techniques—pruning, quantization, distillation—so you benefit from the latest without chasing papers.If your team is eyeing a chip migration or evaluating new NPUs, a dropdown swap in Aptos triggers a fresh search tuned to the new hardware, minimizing lock-in and keeping options open. The result is timeline compression: where projects used to take 12–18 months with large teams, we aim to surface strong, deployable candidates in one to two weeks. Subscribe for more deep dives into edge AI deployment, share this episode with your team, and leave a review telling us which device you want to target next.Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →Neural-ART: ST’s New NPU Architecture at the Edge
Apr 900:14:47Tap to summarizeWhat if the fastest path to efficient edge AI isn’t a bigger CPU, but a smarter stream of data? We pull back the curtain on NeuralArt—the flexible, stream‑based accelerator inside the STM32N6—and show how a decade of prototypes led us to rethink how tensors move, how layers are scheduled, and how much work a compiler can save when memory is the real bottleneck. Instead of shuttling activations back and forth, our architecture routes data through specialized units in tightly orchestrated “epochs,” keeping compute hot and bandwidth cool.From there, we tackle the hard limits of standard‑cell designs on practical MCU nodes. Power efficiency stuck around 1–5 TOPS/W and density near 0.1–2 TOPS/mm² pushed us to explore in‑memory computing. We break down digital versus analog IMC—determinism and integration on one side, approximate but highly efficient compute on the other—and share prototype results that hit roughly 40 TOPS/W and about 10 TOPS/mm² at 1 GHz. Along the way, we dig into why half of system power can vanish into data movement and how weight‑stationary strategies change the game.We also get candid about trade‑offs. Embedded phase change memory (PCM) brings remarkable density and multi‑level storage, but demands strict weight‑stationary mapping and drift compensation. No single technology wins every metric, so we lay out a heterogeneous 2D mesh that blends digital IMC, analog IMC, and classical stream units. Our compiler assigns each subgraph to the node that fits its accuracy, throughput, and energy needs, and our NeoSoC research effort moves this vision toward silicon with an upcoming 80‑nm tapeout.If you care about edge inference, memory bandwidth, quantization, and real‑world efficiency beyond spec‑sheet peaks, this conversation is for you. Subscribe, share with a teammate who’s wrestling with on‑device AI, and leave a review with the biggest bottleneck you want us to tackle next.Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →A Unified Neuromorphic Platform for Sparse, Low Power Computation
Apr 200:20:10Tap to summarizeSensors are flooding the edge with data while CPUs juggle denoising, formatting, and inference. We built ADA to flip that script: a Turing-complete neuromorphic processor that computes with time-encoded spikes, slashing power, latency, and memory movement by keeping work inside an event-driven pipeline.We start by unpacking why conventional embedded architectures stall under modern workloads, from pre-processing bottlenecks to compromised security on battery-powered devices. Then we break down neuromorphic fundamentals—how spikes encode information and why sparsity matters—and compare general-purpose frameworks, highlighting the trade-offs that often inflate activity or force manual design. From there, we explain why we chose interval coding and how we solved its biggest flaw. By predicting future spike times, ADA avoids per-tick updates, reducing complexity from linear to logarithmic with precision and mapping neatly to simple add, multiply, and shift hardware.You’ll hear how the architecture comes together: a tiny neuron core that fits in modest FPGAs, standard interfaces like UART and AER for DVS cameras, and our Axon SDK that compiles Python, NumPy, or C algorithms into deployable binaries—no neuron micromanagement required. We demo a three-tap FIR filter built from modular primitives and show ADA acting as a programmable pre-processing element for event vision. On the DVS128 gesture dataset, ADA’s spatial-temporal denoising cut downstream compute by over 50%, keeping the pipeline sparse and fast.Security gets equal attention. We extended the primitive set with modulus arithmetic to support polynomial math central to post-quantum cryptography such as Kyber. The result: 5x better power efficiency and a 2.5x improvement in energy-latency product over MCU baselines, with clear paths to reduce latency further. It points to neuromorphic cryptography that protects implants and IoT sensors without sacrificing battery life.Ready to try it? The Axon SDK is publicly available. Give ADA a spin, share your toughest edge workload, and subscribe for more deep dives into neuromorphic computing. If this sparked ideas, leave a review and pass it to a friend building at the edge.Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →From Fragments to Foundation: The Sound of Progress in Edge Audio AI
Mar 2600:29:18Tap to summarizeWhat if your printer didn’t just spit out pages, but actually understood them? We walk through a hands-on look at multimodal AI on the edge—how visual-language models read layouts, extract tables, translate content, and reformat documents right where data lives, without shipping sensitive files to the cloud. It’s a practical tour from passive peripherals to active intelligence, with real workflows and measurable speedups.We share the architecture behind on-device document intelligence: pre-processing that stabilizes inputs, VLMs that localize and reason over text and images, and post-processing that converts outputs into CSVs, charts, and accessibility-friendly layouts. You’ll hear how Qwen 2.5-VL handles complex visual inputs while maintaining strong language performance, and how a Flux-based diffusion setup enables creative generation and targeted edits—from updating dates in greeting cards to changing borders and colors by prompt. Along the way, we unpack quantization with GGUF to run 7B-class models in tight memory, diffusion sampler and scheduler tuning for latency, and NVIDIA-optimized libraries to squeeze more from modest GPUs.Beyond demos, we dig into business and engineering realities: fine-tuning with enterprise data to reduce hallucinations, building guardrails and fallback paths for reliability, and segmenting large documents to manage VRAM. We also discuss why a companion device—AI PC or smartphone—can orchestrate heavy lifting until printer SOCs catch up, keeping data private and workflows responsive. If you care about document AI, privacy by design, or accessibility features like dynamic type and contrast, this conversation makes the path concrete and actionable.Enjoy the deep dive? Subscribe, share with a colleague who lives in PDFs, and leave a review with the one edge use case you want us to test next.Send us Fan MailSupport the showLearn more about the EDGE AI FOUNDATION - edgeaifoundation.org
Transcribe →