The 3rd edition of AutoMLPerSys will focus on advancing Automated Machine Learning (AutoML) for Pervasive sensing Systems, with an emphasis on edge devices, wearables, and IoT applications. Building on the established foundation of previous editions, the 2027 workshop continues to address critical challenges and opportunities in the field such as: 

• 1. AutoML for Tiny and Edge Devices:

• Neural Architecture Search (NAS), hyperparameter optimization, and quantization/pruning for microcontrollers, wearables, and IoT.

• Energy-aware training/inference and hardware-aligned model compression.

2. Open Datasets for AutoML Research:

   • Call for curated datasets (e.g., health, activity, environmental sensing) with initial benchmarking results to accelerate reproducibility.

3. Small Language Models (SLMs) for Edge AI:

   • Optimization techniques for SLMs:

     • Efficient tokenization, distillation, and sparsity-aware training.

     • Automated SLM design (e.g., NAS for latency-accuracy trade-offs).

   • Applications in on-device NLP, real-time decision-making, and personalized AI.

4. Explainable AI (XAI) for Tiny Models:

   • Automated interpretability: Tools to generate explanations (e.g., SHAP, LIME) during model design.

   • Self-explaining models: Lightweight architectures with built-in transparency (e.g., attention maps for edge vision).

5. Emerging Challenges:

   • Ethical AI: Privacy-preserving AutoML (federated learning, differential privacy).

   • Dynamic adaptation: AutoML for non-stationary sensor data (concept drift, lifelong learning).

6. Core Challenges:

• Data quality, scalability, adaptability, and personalization .

• New focus: Ethical AI and privacy-preserving AutoML for pervasive systems.

Workshop Goals

• Explore novel datasets to accelerate AutoML research in pervasive sensing.

• Discuss SLMs and XAI as key enablers for edge AI.

• Foster collaboration between academia/industry on scalable, ethical AutoML solutions.