In the Twelfth edition of EMC2 workshop, we plan to discuss the enormous impact of agentic AI on diverse industries driving productivity but also creating unprecedented demands for computing capacity and energy. Assessing the sustainability of large-scale AI computing systems involves discussions spanning multiple interrelated areas. First, we continue to serve as the leading forums for discussing the energy-efficiency aspect of GenAI workloads which directly impact the overall viability and economic value of AI technology. Second, we examine the evolving scaling laws of AI with the prevalence of agentic, and reasoning-based models in
conjunction with novel techniques such as sparse architectures, speculative compute, and multi-architecture disaggregation. Finally, we discuss sustainable and high-performance computing paradigms towards efficient datacenters and hybrid computing models that can cater to the exponential growth in model sizes, application areas, and user base. This would allow us to explore ideas to build the hardware, software, systems, and scaling infrastructure, as well as model architectures that make AI technology even more prevalent and accessible.

Focus Areas

• Energy Efficient LLMs/GenAI: In this era of AI, where a trillion is the norm (whether in model parameters, dataset size, or revenue expectations), there is a lot of misdirection and media attention – especially around generative AI and Large Language Models. While the focus often centers on building bigger systems, an equally important question is whether we can afford to run them at scale, both in terms of total cost of ownership (TCO) and return on investment (ROI). We
  address the fundamentals of language understanding and discuss future directions and ideas for building efficient models and systems to tackle more complex language problems and challenges.
• Algorithm-System Co-design for Reasoning: The capabilities of new AI systems have spread beyond plain language—be it multi-modal interaction or step-by-step reasoning for complex problem solving. One can expand this to also include multi-agent systems and retrieval-augmented generation (RAG) pipelines. These directions open up new challenges in model compression, efficiency, and optimization, all while striving to preserve reasoning capabilities and system robustness. We invite discussion about forward-looking projections: what is it going to take to scale the current systems to meet the compute demands of the next few generations of models (Kimi K4, GPT-6, and beyond) and if current scaling laws provide a path to AGI.
• Determinism and Consistency: Persistent challenge remains in LLMs: these models often lack consistency and determinism, especially in tasks requiring spatial or temporal coherence. For example, 3D diffusion models may generate scenes that change unpredictably when the camera is rotated, resulting in non-identical outputs for what should be equivalent view- points. We will discuss methods that improve the reliability, repeatability, and physical plausibility of generative models. The topics will include architectural innovations, training strategies, evaluation metrics, and applications where consistency is critical, such as graphics, simulation, and virtual reality. The goal is to foster collaboration and spark new ideas for making generative AI more robust and trustworthy in real world scenarios.
• Sustainable High Performance Computing: The widespread adoption of Large Language Models (LLMs) in the industry has led to an exponential increase in computing demands and energy consumption, posing sustainability challenges. Efforts have been made in academia and industry to enhance efficiency via novel architectures and optimization techniques, but their success and viability remain open to debate. We discuss what sustainability means for the industry and how to develop new workflows for AI that prioritize environmental impact alongside performance and revenue.

Workshop Format 

The goal of this Workshop is to provide a forum for researchers and industry experts who are exploring novel ideas, tools, and techniques to improve the energy efficiency of Generative AI and Machine Learning as it is practiced today, and would evolve in the next few years. We have tailored our program to serve the participants best in a full-day, in-person setting. Our forum facilitates active exchange of ideas through:

• Keynote presentations by visionaries of the field
• Invited talks about high-impact and ongoing research
• Oral presentations by authors of accepted papers
• Interactive poster session during breaks
• Thought-provoking panel discussion with active audience participation
• Presentation by the winners of our efficiency leader-board

Our tentative lineup of invited speakers includes researchers from Microsoft, Cerebras, UT Austin, MiniMax, Modular, University of British Columbia, Google, d-Matrix, Intel, and more. Further details about the workshop and the kind of discussions we have facilitated can be found at: https://www.emc2-ai.org/
The efficiency leader-board is an open competition where participants devise creative solutions to a specified problem related to efficient AI computing, in the months leading up the Workshop. This year’s challenge calls for the creation of high accuracy draft models to be used as speculators for a chosen target model.

Call for Papers

With the advent of ChatGPT and other language models, Generative AI and LLMs have captured the imagination of whole world!

A new wave of intelligent computing, driven by recent advances in machine learning and cognitive algorithms coupled with process technology and new design methodologies, has the potential to usher unprecedented disruption in the way modern computing systems are designed and deployed. These new and innovative approaches often provide an attractive and efficient alternative not only in terms of performance but also power, energy, and area. This disruption is easily visible across the whole spectrum of computing systems – ranging from low end mobile devices to large scale data centers. Applications that benefit from efficient machine learning include computer vision and image processing, language understanding, speech and gesture recognition, malware detection, autonomous driving, and many more. Naturally, these applications have diverse requirements for performance, energy, reliability, accuracy, and security that demand a holistic approach to designing the hardware, software, and intelligence algorithms to achieve the best outcome.
We invite full-length papers describing original, cutting-edge, and even work-in-progress research projects about efficient machine learning. Suggested topics for papers include, but are not limited to:

• Neural network architectures for resource constrained applications.
• Efficient hardware designs to implement neural networks including sparsity, locality, and systolic designs.
• Power and performance efficient memory architectures suited for neural networks
• Network reduction techniques – approximation, quantization, reduced precision, pruning, distillation, and reconfiguration.
• Exploring interplay of precision, performance, power, and energy through benchmarks, workloads, and characterization.
• Performance potential, limit studies, bottleneck analysis, profiling, and synthesis of workloads.
• Simulation and emulation techniques, frameworks, tools, and platforms for machine learning.
• Optimizations to improve performance of training techniques including on-device and large-scale learning.
• Load balancing and efficient task distribution, communication and computation overlapping for optimal performance.
• Verification, validation, determinism, robustness, bias, safety, and privacy challenges in AI systems.
• Efficient deployment strategies for edge and distributed environments.
• Model compression and optimization techniques that preserve reasoning and problem-solving capabilities
• Architectures and frameworks for multi-agent systems and retrieval-augmented generation (RAG) pipelines.
• Systems-level approaches for scaling future foundation models (e.g. GLM, GPT, Kimi, and beyond).

Paper Submission and Reviews

The EMC2 Workshop uses the EasyChair conference management system to allow submission of papers and to conduct reviews. The Workshop follows a single-blind review process with adequate steps taken to eliminate conflicts of interest. Papers should be no longer than six pages (not including references) using the IEEE double column format. Please see the submission site for details: https://www.emc2-ai.org/submission

Workshop Organizers 

• Satyam Srivastava, d-Matrix Corporation, y2satyam@gmail.com
• Sushant Kondguli, Meta
• Ananya Pareek, Mediatek
• Prashant Nair, University of British Columbia and d-Matrix
• Ravi Bhuchara, Intel
• Daria Soboleva, Cerebras Systems

Workshop Website:

https://www.emc2-ai.org/