Call for Papers

Since its inception in 2017, the EMDL workshop has tracked how breakthroughs in deep learning transformed the interpretation of sensor data for mobile systems like smartphones and wearable devices. In the early years, the community focused on making standard inference feasible, overcoming the severe demands that deep models exerted on local resources. By 2022, these methods had matured, successfully adapting CNN and RNN architectures to meet the stringent needs of mixed-reality and cyber-physical systems.

The Shift: From Discriminative to Generative

However, the landscape has shifted once again. We are witnessing a transition from Discriminative AI (classifying sensor data) to Generative AI (reasoning, explaining, and acting on context). While Generative AI (GenAI) brings unprecedented capabilities, it also presents a resource wall. Modern edge devices operate under constraints in memory bandwidth and energy availability that standard GenAI architectures—which are memory-bound and autoregressive—fundamentally exceed. Currently, the most advanced models reside almost exclusively on the cloud, challenging the autonomy of mobile platforms.

In this context, the mobile computing community is in a unique position to begin the careful study of two core technical questions re-framed for the GenAI era. First, how should systems be architected to partition these massive workloads? We must move beyond simple offloading to explore dynamic collaboration where mobile devices handle context and lightweight generation while the cloud supports heavy lifting. Second, what is required to integrate GenAI into resource-constrained systems? This necessitates a re-examination of efficiency, spanning from the compression of Transformer architectures and diffusion models to the software/hardware optimization of mobile processors (CPUs, GPUs, NPUs) for memory-intensive generation rather than traditional convolution.

Scope and Goals

EMDL 2026 explores the intersection of Systems and Generative AI. Unlike traditional AIoT approaches that focus on lightweight classification, this workshop addresses the unique systems challenges of GenAI deployment. We focus on the full stack of efficient deployment: from algorithmic compression to hardware-software co-design for resource-efficient reasoning on wearables, robots, and mobile devices. We invite researchers to submit work that answers core technical questions for the GenAI era:

  1. Architecture: How should systems be architected to partition massive workloads between the cloud and the edge?
  2. Efficiency: What is required to integrate memory-bound GenAI into resource-constrained systems?
  3. Edge-Native Design: How do we define edge-native generative models designed explicitly for physical constraints?

Topics of Interest

We solicit submissions including full technical workshop papers, white position papers, and work-in-progress/demos. Topics include, but are not limited to:

Systems & Runtime

Models & Algorithms

Applications & The Physical World