5th International Workshop on

Advancing AI Through Efficient Computing

Over the past decade, Deep Learning (DL) has revolutionized numerous research fields, transforming AI into a computational science where massive models are trained on large-scale infrastructures. This paradigm shift has accelerated scientific discovery and improved results across various applications. However, harnessing such computational power demands meticulous optimization of neural architectures and training procedures to ensure model effectiveness, scalability, and energy efficiency.

The advent of transformer-based language and vision models has further intensified this challenge. The substantial development, training, and inference costs associated with these architectures have limited their accessibility to a select few major players in the research arena.

CADL Workshop: Bridging AI and High-Performance Computing

Now in its fifth edition, the Computational Aspects of Deep Learning (CADL) workshop aims to address these critical challenges. CADL brings together experts from Deep Learning and High-Performance Computing (HPC) backgrounds to:

• Discuss computational challenges in AI

• Exchange innovative ideas

• Identify solutions for advancing AI in a computationally efficient and energy-conscious manner

• The workshop aligns closely with the ISC High Performance conference, focusing on the intersection of AI and high-performance computing.

• Evolution and Focus

• Since its inception, CADL has continuously adapted to the rapidly evolving AI landscape, with an increasing emphasis on scalability and energy efficiency. This edition builds upon previous successes, incorporating emerging trends and addressing new challenges in the field.

• Key Areas of Interest

• The workshop welcomes submissions across a wide range of applications, including:

• Applied DL in compute-limited environments (e.g., embedded or automotive systems)

• Novel architectures for data-intensive scenarios

• Energy and power reduction in DL

• Efficient training frameworks and algorithms

• Distributed learning approaches (reinforcement, imitation, and training)

• Large-scale pre-training techniques for real-world applications

• HPC and massively parallel architectures for DL

• Model optimization (pruning, compression, quantization)

• Memory and data transmission efficiency

• Differentiable metrics for computational and energy cost estimation

• Hardware accelerators for DL

• Efficient data handling in training

• Integration of vision and language models

Impact and Contributions

The insights and techniques shared at CADL aim to benefit both the scientific community and industry practitioners. By focusing on optimizing DL models and infrastructures, the workshop contributes to the development of more efficient and accessible AI systems across various domains.

Submission and Participation

We invite full paper submissions, which will be presented as talks or posters during the workshop. Outstanding contributions may be selected for oral presentations. Detailed submission guidelines and deadlines are available in the designated section of this website.

Networking and Collaboration

CADL offers an exceptional platform for networking and collaboration among researchers, practitioners, and industry professionals. Attendees will have ample opportunities to engage in discussions, share experiences, and forge new partnerships in the dynamic field of computational deep learning.

Join us at CADL to be part of shaping the future of efficient and accessible AI computing!