publications

2025

1. arXiv

   Towards Operational Automated Greenhouse Gas Plume Detection

   Brian D. Bue, Jake H. Lee, Andrew K. Thorpe, and 7 more authors

   2025

   Abs HTML

   Operational deployment of a fully automated greenhouse gas (GHG) plume detection system remains an elusive goal for imaging spectroscopy missions, despite recent advances in deep learning approaches. With the dramatic increase in data availability, however, automation continues to increase in importance for natural and anthropogenic emissions monitoring. This work reviews and addresses several key obstacles in the field: data and label quality control, prevention of spatiotemporal biases, and correctly aligned modeling objectives. We demonstrate through rigorous experiments using multicampaign data from airborne and spaceborne instruments that convolutional neural networks (CNNs) are able to achieve operational detection performance when these obstacles are alleviated. We demonstrate that a multitask model that learns both instance detection and pixelwise segmentation simultaneously can successfully lead towards an operational pathway. We evaluate the model’s plume detectability across emission source types and regions, identifying thresholds for operational deployment. Finally, we provide analysis-ready data, models, and source code for reproducibility, and work to define a set of best practices and validation standards to facilitate future contributions to the field.

2024

1. IEEE Xplore

   Finding Common Ground: A Two-Opinion Approach to Reducing Polarization in Networks

   Sulekha Kishore^* and Anagha Satish^*

   In 2024 IEEE MIT Undergraduate Research Technology Conference (URTC), 2024

   Abs DOI

   The most obvious strategy for reducing polarization in social networks is to connect people with opposing views. However, this is neither realistic nor likely to result in productive connections or conversations. To reduce polarization in networks, we propose a two-dimensional model with two opinion sets, each representing a different issue. We introduce a new \textitcommon ground strategy to more realistically reduce polarization on a primary issue, by creating connections between individuals with similarities on a separate, secondary issue. We propose a new polarization metric and measure our common ground strategy’s impact on polarization on both real-world networks. We find a significant reduction in polarization through utilization of this strategy.

2. NeurIPS OptML

   Consensus Based Optimization Accelerates Gradient Descent

   Anagha Satish, Ricardo Baptista, and Franca Hoffmann

   2024

   Abs HTML

   We propose a novel algorithm for integrating gradient information into Consensus Based Optimization (CBO), a recently proposed multi-particle gradient-free optimization method. During each iteration, a subset of particles are updated using local gradient information, while others are updated using a traditional CBO step. We propose a method for subset selection and investigate its empirical performance. The algorithm combines gradient and gradient-free optimization to encourage exploring the state space while maintaining fast convergence. We investigate the tradeoff between accuracy and computational cost when adjusting the number of gradient evaluations. When applied to classification tasks in machine learning, the proposed algorithm attains a similar accuracy to ensemble gradient methods based on Gradient Descent or Adam at a reduced computational cost.