Archives: Livestream

Taking models from development to production at the edge is hard, sometimes really hard. Phones, SoCs, and devboards each introduce their own constraints that make performance unpredictable.

This session presents a practical, vendor-neutral workflow for on-device verification based on remote device execution, and performance profiling. The approach replaces trial-and-error with repeatable processes, enables fair comparisons across hardware, and delivers on-device data that empowers developers to make informed design decisions.

Local AI model execution is essential for time-critical and data-sensitive edge AI applications. Modern edge-AI chips offer improved performance by integrating heterogeneous compute cores—such as CPUs, GPUs, and NPUs—into a single system. However, current deployment frameworks fall short in effectively leveraging such heterogeneous platforms.

We present a flexible and lightweight SDK designed to address this gap. Our SDK supports ahead-of-time model compilation for a range of formats, including PyTorch (custom or Hugging Face), TensorFlow Lite, and ONNX. It abstracts the complexities of heterogeneous systems, enabling seamless deployment across multiple compute targets without requiring specialized hardware knowledge.

With a Python-friendly API, the SDK empowers developers to switch workloads dynamically between different cores—for example, balancing inference between CPU and GPU on ARM-based SoCs. We demonstrate how this capability impacts model throughput for both image classification tasks and on-device generative AI workloads.

As edge devices push the limits of performance and battery life, EMASS’s “Atoms-to-Apps” vision connects every layer of the stack. ECS-DoT is the embodiment of that philosophy: an ultra-efficient, event-driven SoC that fuses tightly interwoven compute-and memory architectures with a lean AI software toolchain—purpose built for intelligent sensing at the very edge.
This talk explores how ECS-DoT collapses the traditional energy intelligence trade-off,delivering always-on sensing, context-aware decision-making, and instant power-gated wake-up. Whether extending drone flight time or enabling wearables to monitor health for days on a single charge, ECS-DoT proves that true edge AI excellence starts at the atom and ends in the app—with no energy wasted in between.

We unveil the principles behind our model-agnostic AI deployment framework, OwLite, that lets developers reuse a single pipeline for any neural network on diverse edge devices. By abstracting away device-specific quirks and conversion scripts, teams can eliminate redundant engineering effort and focus on optimizing core model performance. We’ll discuss how this unified approach accelerates roll-out, simplifies maintenance, and future-proofs your AI infrastructure against evolving AI landscapes.

As generative AI moves from the cloud to the edge, balancing safety, efficiency, and performance becomes increasingly complex. This livestream will explore critical innovations—such as federated learning, multi-agent systems, and embodied AI—that are shaping the future of edge-native foundation models. Join us to discuss how engineers and researchers must evolve AI infrastructure into a full-fledged systems discipline to enable adaptive, context-aware intelligence across diverse edge environments.

From micromobility to two-wheelers, edge AI is becoming essential to how people and goods move safely and intelligently through the world. But designing smarter, more connected vehicles takes more than hardware—it requires vision systems that are fast, accurate, and built for real-world complexity.

In this edition of EDGE AI BLUEPRINTS, we’ll explore how iRider and NXP Semiconductors are tackling the challenges of connected safety at the edge—from AI-powered digital cockpits to embedded intelligence for urban mobility.

Featuring:
Frédéric Delahais, CEO, iRider
Ali Ors, Global Director of AI/ML Strategy, NXP

Together, they’ll share insights on system design, safety-critical AI performance, and how edge AI is reshaping transportation—followed by live audience Q&A.

In this talk we will discuss Responsible AI Governance and its complexity, particularly within the semiconductor industry, a sector which is characterized by a diverse AI landscape with distinct risks, intense competition, and the challenges of operating globally while adhering to a wide range of regulations and legal obligations.

We will also highlight how implementing best practices in Trustworthy AI not only ensures compliance with emerging AI regulations, such as the EU AI Act, but also provides significant strategic benefits. These include gaining a competitive advantage, creating new business opportunities, and driving innovation. At this critical juncture, where clear requirements and benchmarks are still lacking and harmonized standards are yet to be released, it is important to continue advancing AI literacy, awareness, and competencies to shape the future of responsible AI.

In March and October 2024, the EDGE AI FOUNDATION organized two of the most successful community events in its history. They were the cornerstones marking the transition from tiny machine learning (known as tiny ML) to  Generative Edge AI.

Since then, new studies, evidence, innovations have appeared. The EDGE AI FOUNDATION community has a need to stay up-to-date and educate each other on the latest progress from industry and academia to develop a shared future for generative edge AI.

This two-day “third edition” livestream forum help to gain more understanding about the impact of generative edge AI platforms and to witness the transformative impacts of generative edge AI pushing hardware, software, tooling, applications, services at the next level with the impact of new efforts with agentic AI and physical AI.

Day 1 Sessions Include:

• Fireside Chat: The State of GenAI at the Edge – Samir Jaber & Pete Bernard

• Building Optimized GenAI Use Cases with OpenVINO – Raymond Lo & Ashutosh Kumar, Intel

• Generative AI at NXP – Alberto Alvarez, NXP

• Real-World Deployment & Industry Applications – Anirban Bhattacharjee, Wipro

• When is GenAI Useful at the Edge? – Pete Warden, Useful Sensors

• AI’s Center of Gravity Shifts to the Edge – Vinesh Sukumar, Qualcomm

• Optimizing LLMs for Hardware Accelerators – Chen Lai, Meta

Day 2 Sessions Include:

• Expanding the GenAI Horizon with Qualcomm Dragonwings – Evgeni Gousev, Qualcomm

• Bridging the Digital Divide Through the Edge – Danilo Pau, STMicroelectronics

• AI Agents at the Edge: Architectures & Algorithms – Aruna Kollaru, Dell

• NANDA: The Internet of AI Agents – Pradyumna Chari, MIT

• Continues integration with Executorch – the way LiquidAI utilizes PyTorch + Executorch for getting the optimal frontier – Yuri Khrustalev, Liquid AI

• Feasibility & Tradeoffs in LLM Inference at the Edge – Roberto Morabito, EUROCOM

• Agentic AI + SLMs for Airport Boarding – Andreas Basso, MITO

This two-day livestream event is made available for FREE thanks to the support of EDGE AI FOUNDATION Strategic Partners and our Generative Edge AI Working Group.