Many recent research efforts aim at developing models able to present results as accurate as Foundation Models and Large Language Models and, at the same time, that can be trained, stored and run with limited computational resources. SusAI workshop aims at gathering researchers, present new trends and highlight new ideas about sustainable AI. In particular, it will present approaches, both at the training and at the inference levels, for obtaining lighter models, requiring less energy for training and/or able to be run on edge devices.

Significant topics are related, but not limited to, different techniques and methods aiming at developing models going towards sustainable AI, which can be applied across a wide range of use cases, with particular regard to Natural Language Processing tasks:

Data scaling and data sampling
Hyperparameters optimization
Memory-efficient model architectures
Model pruning
Knowledge distillation
Quantization
Matrix decomposition
Neuro-morphic paradigm
Quantum computing