About

Carla P. Gomes is the Ronald C. and Antonia V. Nielsen Professor of Computing and Information Science at Cornell University, where she also serves as the director of the Institute for Computational Sustainability and co-director of the Cornell University AI for Science Institute. She earned her Ph.D. in computer science with a focus on artificial intelligence from the University of Edinburgh. Her research centers on Artificial Intelligence, particularly large-scale constraint reasoning, optimization, and machine learning. Recently, Gomes has focused on the emerging field of Computational Sustainability, which aims to develop computational methods to address critical environmental, economic, and societal challenges to promote a sustainable future. This interdisciplinary field integrates areas such as constraint reasoning, optimization, machine learning, and dynamical systems to tackle complex problems involving combinatorial decisions in dynamic and uncertain environments. Examples of her work include planning and optimization for wildlife preservation and biodiversity conservation, poverty mapping, and accelerating the discovery of renewable materials like solar fuels through the combination of deep data-intensive learning with inference, reasoning, and optimization. Gomes has been the lead principal investigator on two NSF Expeditions in Computing awards and has co-authored over 200 publications in prestigious venues including Nature and Science, as well as numerous AI and computer science conferences and journals, earning several best paper awards. She has been recognized as the most influential Cornell professor by a Merrill Presidential Scholar in 2020. Her accolades include the AAAI Feigenbaum Prize in 2021 for her high-impact contributions to artificial intelligence, particularly in constraint reasoning, optimization, the integration of reasoning and learning, and founding the field of Computational Sustainability with impactful applications in ecology, species conservation, environmental sustainability, and energy materials discovery. In 2022, she received the ACM/AAAI Allen Newell Award for her contributions bridging computer science and other disciplines. Gomes is a Schmidt AI2050 Senior Fellow and a Fellow of the Association for the Advancement of Artificial Intelligence (AAAI), the Association for Computing Machinery (ACM), and the American Association for the Advancement of Science (AAAS).

Research topics

• Sociology
• Political Science
• Socioeconomics
• Geography
• Ecology
• Environmental science
• Environmental resource management
• Business
• Environmental planning
• Medicine
• Environmental health

Selected publications

• Position: Multimodal Large Language Models Can Significantly Advance Scientific Reasoning

  arXiv (Cornell University) · 2025-02-05 · 1 citations

  preprintOpen access

  Scientific reasoning, the process through which humans apply logic, evidence, and critical thinking to explore and interpret scientific phenomena, is essential in advancing knowledge reasoning across diverse fields. However, despite significant progress, current scientific reasoning models still struggle with generalization across domains and often fall short of multimodal perception. Multimodal Large Language Models (MLLMs), which integrate text, images, and other modalities, present an exciting opportunity to overcome these limitations and enhance scientific reasoning. Therefore, this position paper argues that MLLMs can significantly advance scientific reasoning across disciplines such as mathematics, physics, chemistry, and biology. First, we propose a four-stage research roadmap of scientific reasoning capabilities, and highlight the current state of MLLM applications in scientific reasoning, noting their ability to integrate and reason over diverse data types. Second, we summarize the key challenges that remain obstacles to achieving MLLM's full potential. To address these challenges, we propose actionable insights and suggestions for the future. Overall, our work offers a novel perspective on MLLM integration with scientific reasoning, providing the LLM community with a valuable vision for achieving Artificial General Intelligence (AGI).

  PublisherOA PDFDOI

• No Trick, No Treat: Pursuits and Challenges Towards Simulation-free Training of Neural Samplers

  ArXiv.org · 2025-02-10 · 2 citations

  preprintOpen access

  We consider the sampling problem, where the aim is to draw samples from a distribution whose density is known only up to a normalization constant. Recent breakthroughs in generative modeling to approximate a high-dimensional data distribution have sparked significant interest in developing neural network-based methods for this challenging problem. However, neural samplers typically incur heavy computational overhead due to simulating trajectories during training. This motivates the pursuit of simulation-free training procedures of neural samplers. In this work, we propose an elegant modification to previous methods, which allows simulation-free training with the help of a time-dependent normalizing flow. However, it ultimately suffers from severe mode collapse. On closer inspection, we find that nearly all successful neural samplers rely on Langevin preconditioning to avoid mode collapsing. We systematically analyze several popular methods with various objective functions and demonstrate that, in the absence of Langevin preconditioning, most of them fail to adequately cover even a simple target. Finally, we draw attention to a strong baseline by combining the state-of-the-art MCMC method, Parallel Tempering (PT), with an additional generative model to shed light on future explorations of neural samplers.

  PublisherOA PDFDOI

• Deep mutual learning: incentives and trust through collaborative integration of artificial intelligence into sustainability science

  RSC Sustainability · 2025-01-01

  articleOpen access

  Sustainability science increasingly requires computationally intensive predictive and decision-making tasks across varied temporal and spatial scales.

  PublisherOA PDFDOI

• Science-policy: UNFCCC policymakers’ perspective of scientific scenarios and their policy relevance

  npj Climate Action · 2025-05-22 · 7 citations

  articleOpen access

  Abstract Scenarios play a pivotal role in linking climate science to policy action, informing the Intergovernmental Panel on Climate Change (IPCC) reports and international negotiations within the United Nations Framework Convention on Climate Change (UNFCCC) and its annual Conferences of the Parties (COPs) 1–3 . However, policymakers’ (PMs) perspectives remain understudied. Here, we surveyed UNFCCC National Focal Points (N = 278/n = 57), assessing the knowledge base of international-national PMs, perceptions of scenarios’ policy relevance, and plausible improvements. Results highlight a significant regional knowledge gap, with lower scenario familiarity for PMs representing low- and middle-income countries. Furthermore, policymakers request more straightforward scenario communication and more detail. To improve scenario relevance (credibility and legitimacy) 4 , we recommend more actively disseminating scenario knowledge (enhancing institutional capacity) in the Global South and providing more policy-relevant detail into global scenarios and national extensions (linking scenarios to on-the-ground policy action). This also means reassessing the IPCC’s cautiousness concerning being policy-neutral.

  PublisherOA PDFDOI

• Critic Loss for Image Classification

  2024-12-18 · 1 citations

  articleSenior author

  Modern neural network classifiers achieve remarkable performance across a variety of tasks; however, they frequently exhibit overconfidence in their predictions due to the cross-entropy loss. Inspired by this problem, we propose the Critic Loss for Image Classification (CrtCl, pronounced Critical). CrtCl formulates image classification training in a generator-critic framework, with a base classifier acting as a generator, and a correctness critic imposing a loss on the classifier. The base classifier, acting as the generator, given images, generates the probability distribution over classes and intermediate embeddings. The critic model, given the image, intermediate embeddings, and output predictions of the base model, predicts the probability that the base model has produced the correct classification, which then can be back propagated as a self supervision signal. Notably, the critic does not use the label as input, meaning that the critic can train the base model on both labeled and unlabeled data in semi-supervised learning settings. CrtCl represents a learned loss method for accuracy, alleviating the negative side effects of using cross-entropy loss. Additionally, CrtCl provides a powerful way to select data to be labeled in an active learning setting, by estimating the classification ability of the base model on unlabeled data. We study the effectiveness of CrtCl in low-labeled data regimes, and in the context of active learning. In classification, we find that CrtCl, compared to recent baselines, increases classifier generalization and calibration with various amounts of labeled data. In active learning, we show our method outperforms baselines in accuracy and calibration. We observe consistent results across three image classification datasets.

  PublisherDOI

• Critic Loss for Image Classification

  arXiv (Cornell University) · 2024-09-23

  preprintOpen accessSenior author

  Modern neural network classifiers achieve remarkable performance across a variety of tasks; however, they frequently exhibit overconfidence in their predictions due to the cross-entropy loss. Inspired by this problem, we propose the \textbf{Cr}i\textbf{t}ic Loss for Image \textbf{Cl}assification (CrtCl, pronounced Critical). CrtCl formulates image classification training in a generator-critic framework, with a base classifier acting as a generator, and a correctness critic imposing a loss on the classifier. The base classifier, acting as the generator, given images, generates the probability distribution over classes and intermediate embeddings. The critic model, given the image, intermediate embeddings, and output predictions of the base model, predicts the probability that the base model has produced the correct classification, which then can be back propagated as a self supervision signal. Notably, the critic does not use the label as input, meaning that the critic can train the base model on both labeled and unlabeled data in semi-supervised learning settings. CrtCl represents a learned loss method for accuracy, alleviating the negative side effects of using cross-entropy loss. Additionally, CrtCl provides a powerful way to select data to be labeled in an active learning setting, by estimating the classification ability of the base model on unlabeled data. We study the effectiveness of CrtCl in low-labeled data regimes, and in the context of active learning. In classification, we find that CrtCl, compared to recent baselines, increases classifier generalization and calibration with various amounts of labeled data. In active learning, we show our method outperforms baselines in accuracy and calibration. We observe consistent results across three image classification datasets.

  PublisherOA PDFDOI

• Feitiço contra o feiticeiro? Maré Verde pela legalização do aborto avança no Brasil

  2024-07-03

  article1st authorCorresponding

  A quatro meses das eleições municipais, novo ataque aos direitos das mulheres está no centro das estratégias eleitorais de setores da direita. Mas a reação dura da sociedade pode estar fazendo o tiro sair pela culatra

  PublisherDOI

• AiSciVision: A Framework for Specializing Large Multimodal Models in Scientific Image Classification

  arXiv (Cornell University) · 2024-10-28

  preprintOpen accessSenior author

  Trust and interpretability are crucial for the use of Artificial Intelligence (AI) in scientific research, but current models often operate as black boxes offering limited transparency and justifications for their outputs. We introduce AiSciVision, a framework that specializes Large Multimodal Models (LMMs) into interactive research partners and classification models for image classification tasks in niche scientific domains. Our framework uses two key components: (1) Visual Retrieval-Augmented Generation (VisRAG) and (2) domain-specific tools utilized in an agentic workflow. To classify a target image, AiSciVision first retrieves the most similar positive and negative labeled images as context for the LMM. Then the LMM agent actively selects and applies tools to manipulate and inspect the target image over multiple rounds, refining its analysis before making a final prediction. These VisRAG and tooling components are designed to mirror the processes of domain experts, as humans often compare new data to similar examples and use specialized tools to manipulate and inspect images before arriving at a conclusion. Each inference produces both a prediction and a natural language transcript detailing the reasoning and tool usage that led to the prediction. We evaluate AiSciVision on three real-world scientific image classification datasets: detecting the presence of aquaculture ponds, diseased eelgrass, and solar panels. Across these datasets, our method outperforms fully supervised models in low and full-labeled data settings. AiSciVision is actively deployed in real-world use, specifically for aquaculture research, through a dedicated web application that displays and allows the expert users to converse with the transcripts. This work represents a crucial step toward AI systems that are both interpretable and effective, advancing their use in scientific research and scientific discovery.

  PublisherOA PDFDOI

• Local Context of Climate Change Adaptation in the South-Western Coastal Regions of Bangladesh

  Climate change management · 2023-01-01 · 6 citations

  book-chapter
  PublisherDOI

• The Changing Climate Is Changing Safe Drinking Water, Impacting Health: A Case in the Southwestern Coastal Region of Bangladesh (SWCRB)

  Climate · 2023 · 52 citations

  • Sociology
  • Environmental health
  • Geography

  This study focuses on investigating the impact of climate change on the availability of safe drinking water and human health in the Southwest Coastal Region of Bangladesh (SWCRB). Additionally, it explores local adaptation approaches aimed at addressing these challenges. The research employed a combination of qualitative and quantitative methods to gather data. Qualitative data were collected through various means such as case studies, workshops, focus group discussions (FGDs), interviews, and key informant interviews (KIIs). The study specifically collected qualitative data from 12 unions in the Shyamnagar Upazila. On the other hand, through the quantitative method, we collected respondents’ answers through a closed-ended questionnaire survey from 320 respondents from nine unions in the first phase of this study. In the next phase, we also collected data from the three most vulnerable unions of Shyamnagar Upazila, namely Poddo Pukur, Gabura, and Burigoalini, where 1579 respondents answered questions regarding safe drinking water and health conditions due to climate change. The findings of the study indicate that local communities in the region acknowledge the significant impact of sea-level rise (SLR) on freshwater sources and overall well-being, primarily due to increased salinity. Over 70% of the respondents identified gastrointestinal issues, hypertension, diarrhea, malnutrition, and skin diseases as major waterborne health risks arising from salinity and lack of access to safe water. Among the vulnerable groups, women and children were found to be particularly susceptible to waterborne diseases related to salinity. While the study highlights the presence of certain adaptation measures against health-related problems, such as community clinics and health centers at the upazila level, as well as seeking healthcare from local and paramedical doctors, it notes that these measures are insufficient. In terms of safe drinking water, communities have adopted various adaptation strategies, including pond excavation to remove saline water (partially making it potable), implementing pond sand filters, rainwater harvesting, and obtaining potable water from alternative sources. However, these efforts alone do not fully address the challenges associated with ensuring safe drinking water.

  PublisherOA PDFDOI

Recent grants

• Collaborative Research: CompSustNet: Expanding the Horizons of Computational Sustainability

  NSF · $8.1M · 2015–2023

• Collaborative Research: Computational Sustainability: Computational Methods for a Sustainable Environment, Economy, and Society

  NSF · $7.9M · 2008–2016

• INSPIRE Track 1: UDiscoverIt: Integrating Expert Knowledge, Constraint-Based Reasoning and Learning to Accelerate Materials Discovery

  NSF · $700k · 2013–2017

• EAGER: Exploratory Research in Automated Computational Analysis of Inorganic Materials Libraries

  NSF · $133k · 2013–2014

Frequent coauthors

• Md. Ashrafuzzaman

  University of Chittagong

  48 shared

• Luísa Schmidt

  34 shared

• João Guerra

  University of Lisbon

  16 shared

• Susana Guerreiro

  10 shared

• Pedro Prista

  Iscte – Instituto Universitário de Lisboa

  9 shared

• Filipe Duarte Santos

  Sustainability Institute

  8 shared

• Iyad Rahwan

  Max Planck Institute for Human Development

  8 shared

• Mirko Viroli

  8 shared

Labs

• Carla Gomes LabPI

  Artificial Intelligence with a focus on large-scale constraint-based reasoning, optimization, and machine learning

Education

• Ph.D., computer science in artificial intelligence

  University of Edinburgh

Awards & honors

• AAAI Feigenbaum Prize (2021)
• 2022 ACM/AAAI Allen Newell Award
• ACM Fellow
• AAAS Fellow
• AAAI Fellow