Presentations

Costas Pantelides, CTO, Siemens Process Automation Software

• Title: Autonomy in Process Operations and the role of AI

• Abstract: Autonomy in process operations aims to automate important tasks which currently require significant human intervention, skill and judgment. This presentation will present a recent analysis of the key drivers for autonomy in multiple sectors of the process industries, and outline concrete requirements and use cases arising from this analysis. It will also provide a brief description of a comprehensive Autonomous Process Operations System (APOS) designed to address the above requirements, and discuss the increasing role of AI-based models and technologies in this context, particularly when tightly integrated with complementary modeling methodologies and computational techniques.

Leaelaf Hailemariam, Digital Innovation Leader, DuPont Water Solutions

• Title: AI in DuPont – Discovery, Productivity & Customer Support 

• Abstract: Leaelaf Hailemariam, R&D Manager and Digital Innovation Leader at DuPont Water Solutions, will provide an overview of AI adoption at DuPont with specific examples of AI tools used to discover new solutions and applications, increase productivity and support customers. He will describe externally accessible Ai enabled tools such as Minerva and the RO Operations Advisor.

Elizabeth Balapitiya, Director - AI Technologies, RD&E, Ecolab & Albert Goldfain, Director - Data Science, Advanced Analytics, Ecolab

• Title: AI in Action: Accelerating Innovation and Value Creation in Applied Chemistry

• Abstract: Innovation sits at the intersection of scientific discovery, product performance, and real-world operational constraints. At Ecolab, advances in artificial intelligence are increasingly enabling new ways to accelerate innovation, improve decision making, and deliver measurable cost and sustainability benefits across the R&D pipeline. This presentation will share an industry perspective on how AI is being integrated into applied chemistry workflows to enhance formulation development, experimental design, and process optimization. The talk will illustrate how data-driven models and digital tools complement deep domain expertise, enabling teams to explore broader design spaces, reduce experimental cycles, and more efficiently translate laboratory insights into scalable solutions. The discussion will also highlight practical considerations for deploying AI in applied chemical environments, including data quality, model interpretability, and cross functional collaboration between chemists, engineers, and data scientists. We will also provide a high-level overview of our technical AI innovation initiatives and review Ecolab’s responsible AI framework.

Leo Chiang, Senior Director, Lubrizol & Robert Duan, VP of Technology, Lubrizol

• Title: Amplifying AI and Digital Impacts in the Specialty Chemical Industry

• Abstract: In the digital era, artificial intelligence (AI) is transforming how industries operate— powering breakthroughs in image analytics, large language models (LLMs), deep learning, reinforcement learning, hybrid modeling, and real-time decision-making. This talk will reframe the narrative around AI, not as a hype or threat to replace human decisions, but as an opportunity to amplify impact in industry. The talk will show industrial examples on how humans and AI must be working in the loop. One aspect is to understand how to incorporate AI methods to assist humans to accelerate discovery in research and to make well-informed decisions in manufacturing operations. The other aspect is to allow humans to incorporate engineering and science domain knowledge to make AI methods smarter. This talk aims to showcase AI success stories in the specialty chemical industry such as the use of molecular dynamic for physics based molecular modeling; the use of computer vision tools for quantifying the improvements seen related to our beauty products on skin. The discussion will extend to anticipated research paths, the necessity for workforce development, and the imperative collaboration between academia, technology providers, and the industry to shape the AI future together.

Anand Chandrasekaran, Product Manager - Materials Science Informatics, Schrödinger

• Title: Empowering the Digital Chemistry Laboratory: Generative AI, Agentic Workflows, and the Future of Materials Design

• Abstract: The rapid evolution of artificial intelligence is shifting materials science from expensive, trial-and-error experimentation to autonomous, data-driven design. Drawing on recent advancements at Schrödinger, this talk will address the efficient adoption of AI, its impact on daily scientific practice, and strategies for future-proofing R&D organizations. First, we will explore how AI accelerates existing operations and enables new opportunities. By leveraging Machine Learning Force Fields (MLFFs) like MPNICE, we can drastically speed up computationally expensive quantum mechanics and molecular dynamics workflows while maintaining ab initio accuracy. Furthermore, Generative AI capabilities like REINVENT unlock the true de novo inverse design of novel molecules and complex formulations optimized for specific target properties using reinforcement learning. Next, we will examine how AI enhances autonomy in materials design. We will highlight Formulation Machine Learning and Optimization, which maps chemical structure and composition directly to physical properties to autonomously suggest the "next best experiment". We will also introduce agentic AI, such as Schrödinger's digital chemistry assistant, demonstrating how expert digital assistants capable of multi-tasking, pipelining jobs, and executing complex research projects are fundamentally altering how scientists interact with computational platforms. Finally, we will discuss how a digital chemistry strategy built on the synergy of physics-based modeling and machine learning is the most effective way to future-proof AI adoption. By utilizing rigorous physics-based simulations to generate massive, highly accurate training datasets, organizations can overcome data scarcity limitations and confidently extrapolate into the vast space of synthesizable chemistry.

Jayshree Seth, Corporate Scientist & Chief Science Advocate, 3M

• Title: Gen AI for R&D: Go with the (work)flow

• Abstract: In this presentation Jayshree will discuss key learnings from an ongoing initiative for incorporating generative AI powered tools to assist R&D personnel at 3M. Researchers routinely rely on information to generate insights, ideas and inventions for product innovations that deliver value. Where traditional AI/ML can power quantitative analysis, generative AI can accelerate qualitative synthesis of information for strategic interpretation and action. Driving adoption involves mapping workflows to address pain-points, communicating the benefits with real examples, improving modalities through feedback, influencing with testimonials from lead-users and incentivization from leadership. All indications are that AI fluency within familiar workflows intuitively sparks meaningful discussions around improvements as well as AI-first process redesign.

Michail Vlysidis, Senior Engineer, AbbVie

• Title: Real-World AI Integration in Drug Discovery and Development 

• Abstract: Artificial intelligence is transforming biopharma, but its impact varies significantly across the drug development lifecycle. This talk examines how different AI approaches, from machine learning to generative AI agents, are being deployed across discovery, development, and manufacturing stages, with particular focus on biologics pipelines. The talk will map the drug development journey from target identification through manufacturing and will identify where specific AI types deliver the most value. Through concrete examples, it will discuss how AI agents streamline information access, how biologics property prediction models inform pipeline decisions, and how AI-powered analytics detect common themes in manufacturing errors; all while emphasizing that AI's power lies in amplifying, not replacing, human expertise.

Jen McKay, Lead - Health on Cloud, Senior Clinical Specialist, Google Health

• Title: The "To Err is Human" Moment Revisited: AI as the System Design Response to a Culture of Safety

• Abstract: Two decades after the Institute of Medicine's "To Err is Human" report, healthcare still faces a "Retributive Gap", where system failures and design discrepancies contribute to preventable errors. This talk frames the current critical juncture—marked by high clinician burnout and rapid technological integration—as a renewed "To Err is Human" moment. We examine how system failure is rooted in the gap between "Work-as-Imagined" (WAI) and the realities of "Work-as-Done" (WAD), where high cognitive load and interface discrepancies lead to fatal errors, such as look-alike drug packaging. Generative AI is not merely a tool but the necessary "System Design" response to address these failures. We explore four systemic potentials—to Respond, Monitor, Learn, and Anticipate—leveraging advancements from fields like commercial aviation and manufacturing. Finally, we showcase how Google is applying rigorous AI research, including LearnLM for health workforce training and the AMIE diagnostic agent, to safely and consistently augment clinical workflows, championing a Human-Led approach to ensure scaled, ubiquitous AI.

Ju Sun, Associate Professor, Department of Computer Science & Engineering

• Title: Accelerating materials discovery via physics-informed constraints

• Abstract: Machine learning struggles with predicting material stability. We hypothesize this stems from models treating materials independently, whereas material stability is intrinsically a group-wise property. Our work imposes thermodynamic constraints on training, aiming to capture underlying relationships. This will enable machine learning models to serve as effective, low-cost surrogates for DFT, accelerating solid-state inorganic materials discovery. The computational tools we will be developing, especially those for faithfully handling physical constraints, have far-reaching impacts for scientific machine learning with hard constraints. 

Aryan Deshwal, Assistant Professor, Department of Computer Science & Engineering

• Title: Adaptive Experimental Design for Accelerating Materials Discovery  

• Abstract: Many problems in chemical design and materials discovery involve searching over a large number of candidates, where evaluating each candidate is expensive because it requires significant resources. For example, searching the space of materials for a desired property while minimizing the total resource-cost of physical wet-lab experiments or computationally expensive simulations for their evaluation. The key challenge is how to select the sequence of experiments to uncover high-quality solutions for a given resource budget. In this talk, I will introduce novel adaptive experiment design algorithms to tackle this challenge. This includes both designing new probabilistic models over combinatorial objects that work well in small data settings and provide principled uncertainty quantification and decision policies to carefully select each experiment for complex real-world settings where there are usually multiple objectives, multi-fidelity experiments, and black-box constraints. I will also present results on applying these algorithms to solve problems in domains including nanoporous materials discovery and surfactant design. In the end, I will also cover some open challenges and future directions for steering generative AI that I am excited about.

Ben Hackel, Professor, Department of Chemical Engineering and Materials Science

• Title: Interpretable Protein Design

• Abstract: Proteins empower natural and engineered biology. Protein sequence/stability relationships remain incompletely understood. While machine learning models trained on extensive datasets enable ‘black-box’ stability prediction, such methods do not elucidate the factors that inform stability. We applied network theory to construct a model that leverages perceived knowledge of amino acid interactions and physicochemical communities to effectively predict absolute stability across broad protein topologies and sequences. Parametric analysis reveals that network connectivity – presence/absence of interactions between all site pairs – is the most informative model element. The results highlight the utility of network topology analysis for protein stability and illuminate the importance of interaction networks more than structural coordinates. Merging the network topology perspective with other protein modeling provides compelling future opportunities.

Sapna Sarupria, Associate Professor, Department of Chemistry

• Title: AI and Soft Matter

• Abstract: Molecular systems often behave not because of strong interactions, but because weak, solvent-mediated effects select which dynamical pathways are accessible. This makes it non-trivial to explore the available design space for elucidating the structure-to-function connection. This also leads to specific challenges and the need for novel ways to apply AI for designing soft matter. In my talk, I will highlight the innovative use of AI to gain molecular insights into nucleation and its application for polymorph selection. I will also present recent work on formulation design for biologics and the challenges of applying AI to this problem.

Chris Bartel, Assistant Professor, Department of Chemical Engineering & Materials Science

• Title: Exposing gaps in AI-accelerated materials discovery and synthesis 

• Abstract: Modern computational materials discovery approaches leverage universal machine learning (ML) interatomic potentials (“foundation” potentials) along with generative artificial intelligence (AI) to rapidly search materials space. From the standpoint of materials discovery, the first challenge is efficiently identifying new inorganic crystals that are thermodynamically stable. Once these materials are identified on the computer, the next step is to make them in the lab. Unsurprisingly, ML models are now emerging with the goal of predicting which hypothetical materials are likely to be synthesizable. While these developments are exciting, a pervasive challenge for any ML problem is to anticipate the performance of models outside the curated experiments used for initial training and evaluation. This talk will detail my group’s efforts to systematically assess the quality of generative models and synthesis prediction models through the lens of first-principles thermodynamics.

Bernard Agyeman, Postdoctoral Associate, & Prodromos Daoutidis, Professor, Department of Chemical Engineering & Materials Science

• Title: AI-Accelerated Benders Decomposition for Mixed-Integer Optimization 

• Abstract: Mixed-integer programming (MIP) problems arise in many engineering applications, including process design, supply chain optimization, and model predictive control (MPC). Benders decomposition (BD) is a widely used solution strategy that splits the problem into an integer master problem and a continuous subproblem, both of which are coordinated through iteratively generated Benders cuts. A key bottleneck is the repeated solution of the master problem, whose complexity grows as cuts accumulate over iterations. To address this, we develop a learning-based approach that augments BD with a graph neural network (GNN) to predict integer variable assignments at each iteration, which reduces reliance on expensive MIP solver calls. The master problem is represented as a bipartite graph, and the GNN policy is trained via imitation learning, which is then fine-tuned within the BD loop using reinforcement learning. To preserve the finite convergence guarantees of classical BD, a confidence-based assignment mechanism screens predictions for feasibility and optimality. The framework is evaluated on a benchmark mixed-integer nonlinear programming problem and on a large-scale irrigation scheduling problem for a field in Lethbridge, Alberta, where closed-loop scheduling is performed within a mixed-integer MPC framework. In both cases, the proposed framework achieves substantial reductions in solution time while maintaining solution quality, showing that AI can meaningfully accelerate established optimization algorithms without compromising their theoretical guarantees.

Jnana Sai Jagana, Ph.D Candidate, & Qi Zhang, Associate Professor, Department of Chemical Engineering & Materials Science

• Title: Robust demand response aggregation under customer response uncertainty

• Abstract: Demand response (DR), where electricity consumers adjust their consumption profiles in response to price signals from the electricity market, is considered a crucial strategy for maintaining a stable power grid and achieving a sustainable energy system with net-zero carbon emissions. In the context of DR, smaller loads are often combined such that they can collectively participate in the wholesale electricity market. This aggregation of DR resources is typically coordinated by DR aggregators that serve as intermediaries between the electricity consumers and the market or grid operator. In this work, we take the perspective of a DR aggregator that operates in the wholesale market to perform DR activities on behalf of their customers by offering incentives in exchange for customers’ modifying their power load. Here, a major challenge is due to the fact that the customers may not always respond as per the DR aggregator’s request, introducing uncertainty into the DR aggregator’s problem for which we use robust optimization techniques to solve the problem effectively.