Overview
Do we always need bigger models? As the AI community pushes toward ever-larger architectures, a fundamental question remains underexplored: How much model capacity does a specific task actually require?
This workshop explores the fundamental question of how much capacity (in terms of parameters, architecture, and data) an AI model requires to effectively solve a specific task. This question is central to the ongoing community debate between large multi-task foundation models and smaller task-specific models. We will bring together researchers and practitioners from machine learning and a broad spectrum of applications, including but not limited to geosciences, healthcare, robotics, and biology to discuss theoretical foundations, empirical findings, and real-world applications. We welcome contributions to the full spectrum from TinyML on microcontrollers to Large Language Models, with a focus on principled approaches to right-sizing AI systems. Ideally, workshop contributions should investigate for a specific application scenario or task the tradeoffs in various performance metrics, including but not limited to predictive performance, inference speed, energy efficiency or robustness.
Keywords: model capacity, model selection, multi-objective optimization, tinyML, green AI, domain-specific AI, small language models
Invited Speakers
Confirmed and invited speakers.
Topics of Interest
We invite submissions on (but not limited to) the following topics.
Theory and Methodology
- Theoretical and empirical bounds on model capacity (scaling laws, VC theory, PAC-Bayes)
- Model selection and neural architecture search for right-sized models
- Multi-objective optimization balancing accuracy, efficiency, robustness, and fairness
- Benchmarks and evaluation protocols for capacity-performance trade-offs
Model Compression and Efficiency
- Knowledge distillation, pruning, and quantization
- TinyML and efficient models for edge devices and resource-constrained environments
- Green AI: energy consumption, carbon footprint, and sustainability of model sizing decisions
Domain Applications
- Geosciences: climate modeling, remote sensing, sensor networks
- Healthcare: medical imaging, wearable diagnostics, clinical decision support
- Robotics: manipulation and grasping, human-robot interaction
- Biology: protein folding, genomic analysis, drug discovery
- Other domains with specific capacity constraints or requirements
Empirical Studies
- Comparative studies of small specialized models vs. large general-purpose models
- Task-specific capacity analysis across model families
- Case studies on deploying right-sized models in production
Important Dates
Mark your calendar.
All deadlines are 23:59 AoE (Anywhere on Earth).
Submission Guidelines
Submission guidelines will be announced soon.
Workshop Schedule
Preliminary schedule — details to be confirmed.