Max Balandat (Meta Adaptive Experimentation) "Bayesian Optimization for Sustainable Concrete"

Max Balandat, from Meta's adaptive experimentation team, shared insightful developments in the field of sustainable concrete using Bayesian optimization at a recent hackathon. Highlighting the environmental urgency, Balandat pointed out the concrete industry's substantial carbon footprint, responsible for a significant percentage of global CO2 emissions. With an aim to mitigate this, his team at Meta focuses on long-term, generalizable research, collaborating across various organizational domains and striving to incorporate their innovations into open-source tools like BoTorch and Ax.
Balandat's presentation dove into the complex challenge of developing sustainable concrete mixes that balance strength and eco-friendliness. Traditional concrete, while strong, is highly CO2-intensive due to its cement content. By optimizing the mix of cement with supplemental cementitious materials (SCMs), it's possible to reduce CO2 emissions without compromising on strength. However, finding the right balance requires navigating a maze of variables, from material proportions to environmental conditions, making Bayesian optimization an ideal tool for this intricate problem.
The team employed a Gaussian Process (GP) model tailored to capture the nuances of concrete's strength development over time. This approach allowed them to predict concrete strength with remarkable accuracy, paving the way for optimization. Leveraging multi-objective Bayesian optimization, they explored trade-offs between early and long-term strength against CO2 emissions. The results were promising, with AI-driven mixes outperforming human-designed ones in both sustainability and strength, marking a significant step towards greener construction materials.
Balandat also elucidated the distinctions and synergies between BoTorch and Ax, Meta's open-source libraries facilitating Bayesian optimization. BoTorch caters to the research community, offering modularity and cutting-edge methods, while Ax presents a user-friendly interface for a broader audience, embodying best practices and simplifications garnered from extensive experience in optimization projects.
Concluding his talk, Balandat expressed enthusiasm for the hackathon's potential to further advancements in sustainable concrete and beyond. His work demonstrates a powerful intersection of environmental science and cutting-edge technology, showing how thoughtful application of Bayesian optimization can lead to tangible improvements in one of the world's most pivotal industries.

[@christophchristoph9909]

Great to see work on sustainable concrete! So important and yet so underexplored! 👏👏