Supply Chains

Innovation Grants

Workshop Papers

Venue Title
NeurIPS 2025 Differentially Private Federated Learning for High-Accuracy Carbon Footprint Prediction that Protects Sensitive Industrial Data (Papers Track)
Abstract and authors: (click to expand)

Abstract: Life Cycle Impact Assessment (LCIA) often lacks accurate data owing to reluctance in industry to share proprietary production information. Here, we present a privacy-preserving framework that improves carbon footprint prediction using federated learning and differential privacy. Our method maintains data confidentiality while enhancing prediction accuracy and consistency. Experiments on public data show strong performance R2 = 0.96 at epsilon=15, comparable to standard and aggregated data models. This approach enables more reliable Scope 3 emissions assessments, supporting accurate and collaborative LCIA amid growing regulatory demands.

Authors: Vijay Narasimhan (EMD Electronics); Hanna Jarlaczyńska (Unit8); Tingting Ou (Columbia University)
NeurIPS 2025 Adaptive Learning in Spatial Agent-Based Models for Climate Risk Assessment: A Geospatial Framework with Evolutionary Economic Agents (Papers Track)
Abstract and authors: (click to expand)

Abstract: Climate risk assessment requires modelling complex interactions between spatially heterogeneous hazards and adaptive economic systems. We present a novel geospatial agent-based model that integrates climate hazard data with evolutionary learning for economic agents. Our framework combines Mesa-based spatial modelling with CLIMADA climate impact assessment, introducing adaptive learning behaviours that allow firms to evolve strategies for budget allocation, pricing, wages, and risk adaptation through fitness-based selection and mutation. We demonstrate the framework using riverine flood projections under RCP8.5 until 2100, showing that evolutionary adaptation enables firms to converge with baseline (no hazard) production levels after decades of disruption due to climate stress. Our results reveal systemic risks where even agents that are not directly exposed to floods face impacts through supply chain disruptions, with the end-of-century average price of goods 5.6% higher under RCP8.5 compared to the baseline in our illustrative economic network. This open-source framework provides financial institutions and companies with tools to quantify both direct and cascading climate risks while evaluating cost-effective adaptation strategies.

Authors: Yara Mohajerani (Pluripotent Technologies Inc)
NeurIPS 2025 Data-Driven Approach for Ship Emissions Prediction: A Case Study on the Saint Lawrence River (Papers Track)
Abstract and authors: (click to expand)

Abstract: The maritime sector plays an important role in global greenhouse gas (GHG) emissions, which presents a major challenge for climate mitigation efforts. As regulatory frameworks and environmental goals become more stringent, accurately predicting emissions in this sector is crucial for informed decision-making and effective policy implementation. This article presents a data-driven approach to predicting maritime emissions using advanced machine learning techniques. Our proposed work predicts emissions from maritime activities, integrating various data sources to improve accuracy and reliability. The aim is to provide actionable insights to monitor ship emissions and assess their environmental impact.

Authors: Abdelhak EL AISSI (UQAR); Ismail Bourzak (Xpert Solutions Technologiques (XST)); Loubna Benabbou (Université du Québec à Rimouski (UQAR)); Abdelaziz BERRADO (Mohammadia School of Engineers (EMI))
NeurIPS 2025 From Sparse to Representative: Machine Learning to Densify IAM Scenario Ensembles for Policy Insight (Proposals Track)
Abstract and authors: (click to expand)

Abstract: This research addresses the challenge of extracting policy-relevant insights from Integrated Assessment Model (IAM) scenario ensembles, which are often sparse, non-representative, and inaccessible to non-experts. We propose a machine learning framework preserving high-dimensional dependencies between variables, enabling generation of plausible in-gap scenarios when one or more outputs are constrained. The intended output is a simplified exploration space for policymakers concerned with crucial climate policy exploration.

Authors: Georgia Ray (Imperial College London)
NeurIPS 2025 Agricultural Monitoring with Fields of The World (FTW) (Tutorials Track)
Abstract and authors: (click to expand)

Abstract: This tutorial demonstrates how to generate field boundaries globally using the Fields of The World dataset, pretrained models, and command line interface (CLI). We then show how to use those boundaries in agricultural monitoring tasks under climate change, including crop type classification and forest loss monitoring. By the end, users will be able to perform the following tasks to support climate change-related decision-making: (1) Extract agricultural field boundaries for any location, (2) Build machine learning models for crop type classification, and (3) Analyze forest loss within agricultural landscapes. By equipping users with the ability to generate field boundaries and link them to climate-relevant monitoring tasks, this tutorial lowers the barrier for researchers, practitioners, and policymakers to access and deploy advanced geospatial AI.

Authors: Hannah Kerner (Arizona State University); Caleb Robinson (Microsoft); Isaac Corley (Wherobots); Matthias Mohr (Taylor Geospatial Engine); Gedeon Muhawenayo (Arizona State University); Ivan Zvonkov (University of Maryland); Tristan Grupp (World Resources Institute); Nathan Jacobs (Washington University St. Louis)
ICLR 2025 Evaluating the Environmental Impact of Language Models with Life Cycle Assessment (Proposals Track)
Abstract and authors: (click to expand)

Abstract: As the scale of machine learning models and the prevalence of AI workloads has grown, so have the computational, financial, and energy requirements of development and deployment. In response, recent research in efficient machine learning and Green AI has proposed interventions aimed at reducing the environmental resource consumption of machine learning, such as model compression, efficient training methods, and data distillation. Additionally, various tools and frameworks have facilitated reporting and measurement of metrics related to efficiency and environmental impact. However, holistic, bottom-up assessment of the end-to-end environmental impacts of ML remains elusive. Inspired by work from the environmental impact community, we propose that holistic lifecycle assessment (LCA) for analyzing language models. We identify use stages for studying LLM development and deployment, propose methods for measuring power utilization, and analysis for comparing the relative environmental costs of individual stages.

Authors: Jared Fernandez (Carnegie Mellon University); Clara Na (Carnegie Mellon University); Yonatan Bisk (Carnegie Mellon University); Emma Strubell (Carnegie Mellon University)
NeurIPS 2024 Large language model co-pilot for transparent and trusted life cycle assessment comparisons (Proposals Track)
Abstract and authors: (click to expand)

Abstract: Intercomparing life cycle assessments (LCA), a common type of sustainability and climate model, is difficult due to basic differences in fundamental assumptions, especially in the goal and scope definition stage. This complicates decision-making and the selection of climate-smart policies, as it becomes difficult to compare optimal products and processes between different studies. To aid policymakers and LCA practitioners alike, we plan to leverage large language models (LLM) to build a database containing documented assumptions for LCAs across the agricultural sector, with a case study on livestock management. The articles for this database are identified in a systematic literature search, then processed to extract relevant assumptions about the goal and scope definition of the LCA and inserted into a vector database. We then leverage this database to develop an AI co-pilot by augmenting LLMs with retrieval augmented generation to be used by stakeholders and LCA practitioners alike. This co-pilot will accrue two major benefits: 1) enhance the decision-making process through facilitating comparisons among LCAs to enable policymakers to adopt data-driven climate policies and 2) encourage the use of common assumptions by LCA practitioners. Ultimately, we hope to create a foundational model for LCA tasks that can plug-in with existing open source LCA software and tools.

Authors: Nathan Preuss (Cornell University); Fengqi You (Cornell University)
ICLR 2024 Empowering Sustainable Finance: Leveraging Large Language Models for Climate-Aware Investments (Papers Track)
Abstract and authors: (click to expand)

Abstract: With the escalating urgency of climate change, it is becoming more imperative for businesses and organizations to align their objectives with sustainability goals. Financial institutions also face a critical mandate to fulfill the Sustainable Development Goals (SDGs), particularly goal 13, which targets the fight against climate change and its consequences. Mitigating the impacts of climate change requires a focus on reducing supply chain emissions, which constitute over 90% of total emission inventories. In the financial industry, supply chain emissions linked to lending and investments emerge as the primary source of emissions, posing challenges in tracking financed emissions due to the intricate process of collecting data from numerous suppliers across the supply chain. To address these challenges, we propose an emission estimation framework utilizing a Large Language Model (LLM) to drastically accelerate the assessment of the emissions associated with lending and investment activities. This framework utilizes financial activities as a proxy for measuring financed emissions. Utilizing the LLM, we classify financial activities into seven asset classes following the Partnership for Carbon Accounting Financials (PCAF) standard. Additionally, we map investments to industry categories and employ spend-based emission factors (kg-CO2/$-spend) to calculate emissions associated with financial investments. In our study, we compare the performance of our proposed method with state-of-the-art text classification models like TF-IDF, word2Vec, and Zero-shot learning. The results demonstrate that the LLM-based approach not only surpasses traditional text mining techniques and performs on par with a subject matter expert (SME) but most importantly accelerates the assessment process.

Authors: Ayush Jain (IBM Research); Manikandan Padmanaban (IBM Research India); Jagabondhu Hazra (IBM Research India); Shantanu Godbole (IBM India); Hendrik Hamann (IBM Research)
ICLR 2023 CaML: Carbon Footprinting of Products with Zero-Shot Semantic Text Similarity (Papers Track)
Abstract and authors: (click to expand)

Abstract: Estimating the embodied carbon in products is a key step towards understanding their impact, and undertaking mitigation actions. Precise carbon attribution is challenging at scale, requiring both domain expertise and granular supply chain data. As a first-order approximation, standard reports use Economic Input-Output based Life Cycle Assessment (EIO-LCA) which estimates carbon emissions per dollar at an industry sector level using transactions between different parts of the economy. For EIO-LCA, an expert needs to map each product to one of upwards of 1000 potential industry sectors. We present CaML, an algorithm to automate EIO-LCA using semantic text similarity matching by leveraging the text descriptions of the product and the industry sector. CaML outperforms the previous manually intensive method, yielding a MAPE of 22% with no domain labels.

Authors: Bharathan Balaji (Amazon); Venkata Sai Gargeya Vunnava (amazon); Geoffrey Guest (Amazon); Jared Kramer (Amazon)
ICLR 2023 Mapping global innovation networks around clean energy technologies (Proposals Track)
Abstract and authors: (click to expand)

Abstract: Reaching net zero emissions requires rapid innovation and scale-up of clean tech. In this context, clean tech innovation networks (CTINs) can play a crucial role by pooling necessary resources and competences and enabling knowledge transfers between different actors. However, existing evidence on CTINs is limited due to a lack of comprehensive data. Here, we develop a machine learning framework to identify CTINs from announcements on social media to map the global CTIN landscape. Specifically, we classify the social media announcements regarding the type of technology (e.g., hydrogen, solar), interaction type (e.g., equity investment, R\&D collaboration), and status (e.g., commencement, update). We then extract referenced organizations via entity recognition. Thereby, we generate a large-scale dataset of CTINs across different technologies, countries, and over time. This allows us to compare characteristics of CTINs, such as the geographic proximity of actors, and to investigate the association between network evolution and technology innovation and diffusion. As a direct implication, our work helps policy makers to promote CTINs by identifying current barriers and needs.

Authors: Malte Toetzke (ETH Zurich); Francesco Re (ETH Zurich); Benedict Probst (ETH Zurich); Stefan Feuerriegel (LMU Munich); Laura Diaz Anadon (University of Cambridge); Volker Hoffmann (ETH Zurich)
ICLR 2023 Emission-Constrained Optimization of Gas Systems with Input-Convex Neural Networks (Papers Track)
Abstract and authors: (click to expand)

Abstract: Planning optimization of gas networks under emission constraints prioritizes gas supply with the smallest emission footprint. As this problem includes complex gas flow physical laws, standard optimization solvers cannot guarantee convergence to a feasible solution, especially under strict emission constraints. To address this issue, we develop an input-convex neural network (ICNN) aided optimization routine which incorporates a set of trained ICNNs approximating the gas flow equations with high precision. Numerical tests on the Belgium gas network demonstrate that the ICNN-aided optimization dominates non-convex and relaxation-based solvers, with larger optimality gains pertaining to stricter emission targets.

Authors: Vladimir Dvorkin (Massachusetts Institute of Technology); Samuel C Chevalier (Technical University of Denmark); Spyros Chatzivasileiadis (Technical University of Denmark)