About

Sara Beery is a Homer A. Burnell Career Development Professor and an Assistant Professor at MIT EECS, specializing in artificial intelligence and machine learning, AI and society, and graphics and vision. Her research focuses on developing systems that interact with the external world through perception, communication, and action, while also learning, making decisions, and adapting to changing environments. She contributes to the understanding and advancement of AI technologies with an emphasis on societal impact and ethical considerations.

Research topics

• Computer Science
• Ecology
• Artificial Intelligence
• Data Mining
• Geography
• Machine Learning
• Biology
• Cartography
• Engineering
• Data science
• Mathematics
• Environmental planning

Selected publications

• A Blueprint for Integrated Climate Intelligence

  2026-05-13

  articleOpen access

  Climate information is advancing faster than the decision systems designed to use it. Emergency response operates on timescales of hours, whereas societal adaptation unfolds over decades. Yet climate science, impact assessment and policy remain poorly integrated, limiting coherent action across timescales. We argue that artificial intelligence should be developed not only as a domain-specific tool, but as an integration layer linking fragmented physical, social and institutional systems. This Perspective outlines a blueprint for Integrated Climate Intelligence built on three pillars: fairness, which prioritizes vulnerability alongside data availability; speed, which reduces the latency between assessment and action; and robustness, which quantifies uncertainty under extrapolation and high-stakes decision-making. We further propose a Climate AI Trust Index as an evaluation framework for assessing whether climate AI systems are sufficiently rigorous, equitable and decision-relevant for operational use.

  PublisherOA PDFDOI

• Deep Multi-modal Species Occupancy Modeling

  2026-03-10

  articleOpen accessSenior authorCorresponding

  Occupancy models are tools for modeling the relationship between habitat and species occurrence while accounting for the fact that species may still be present even if not detected. The types of environmental variables typically used for characterizing habitats in such ecological models, such as precipitation or tree cover, are frequently of low spatial resolution, with a single value for a spatial pixel size of, e.g., 1 km2. This spatial scale fails to capture the nuances of micro-habitat conditions that can strongly influence species presence, and additionally, as many of these are derived from satellite data, there are aspects of the environment they cannot capture, such as the structure of vegetation below the forest canopy. To address these gaps, we propose to combine high-resolution satellite and ground-level imagery to produce multi-modal environmental features that better capture micro-habitat conditions, and incorporate these multi-modal features into hierarchical Bayesian species occupancy models. We leverage pre-trained deep learning models to flexibly capture relevant information directly from raw imagery, in contrast to traditional approaches which rely on derived and/or hand-crafted sets of ecosystem covariates. We implement deep multi-modal species occupancy modeling using a new open-source Python package for ecological modeling, designed for bridging machine learning and statistical ecology. We test our method under a strict evaluation protocol on various mammal species across thousands of camera traps in Snapshot USA surveys, and find that multi-modal features substantially enhance predictive power compared to traditional environmental variables alone. To aid in interpreting these models, we propose a technique based on vision-language models that automatically extracts habitat elements that are particularly influential on model estimates. Our results not only highlight the predictive value and complementarity of satellite and in-situ imagery, but also make the case for more closely integrating deep learning models and traditional statistical ecological models while maintaining their interpretability.

  PublisherDOI

• Essential tools but overlooked bias: Artificial intelligence and citizen science classification affect camera trap data

  Methods in Ecology and Evolution · 2025-09-25 · 2 citations

  articleOpen access

  Abstract Camera trapping generates vast image datasets requiring classification before downstream ecological inference, yet the influence of classification errors on subsequent analyses is often overlooked. Classification performance can vary widely depending on the classification method (e.g. citizen science vs. artificial intelligence [AI]), species, illumination conditions (diurnal vs. nocturnal) and other contextual factors. We compared a citizen science classification method to two AI classifiers (EfficientNet and DeepFaune) using an expert‐labelled hold‐out of 51,588 images across seven classes (‘empty’, ‘human’, ‘cervid’, ‘wild boar’, ‘red fox’, ‘leporid’ and ‘European badger’) captured day and night. For each class and method, we quantified precision (accuracy of positive predictions) and recall (ability to detect all positive instances), then fitted single‐season occupancy models to the classified data and compared estimates against expert‐derived benchmarks. Finally, we conducted a large‐scale simulation to investigate how true occupancy, detection probability and classification performance (recall and precision) collectively influence the accuracy (root mean square error [RMSE]) of occupancy estimates. Citizen scientists exhibited consistently high precision but more variable recall. The AI classifiers outperformed the citizen science method in recall for several species, including wild boar, leporid and European badger. Both approaches performed worse on nocturnal images and showed reduced precision for night‐time ‘empty’ images. Bias in occupancy estimates differed across species, methods and space—the AI‐based estimates were generally more biased, with both the magnitude and direction of bias varying spatially, especially for rarer species such as leporids. In our simulation study, precision emerged as the strongest predictor of occupancy model accuracy, with lower precision substantially increasing RMSE. Lower occupancy rates increased RMSE, and precision regulated the impact of detection probability: at low precision, higher detection probability worsened errors; at high precision, RMSE remained low—or even decreased—as detection probability rose. Although AI classifiers offer unmatched processing speed, our findings show that citizen science can reduce classification errors. Moreover, low precision and poor recall, especially for rare or nocturnal species, can substantially bias occupancy models. Based on our results, we recommend improving precision and accounting for classification quality and uncertainty to ensure robust inference from camera trap data.

  PublisherOA PDFDOI

• Adapting the Re‐ID Challenge for Static Sensors

  IET Computer Vision · 2025-01-01

  articleOpen accessSenior authorCorresponding

  ABSTRACT The Grévy's zebra, an endangered species native to Kenya and southern Ethiopia, has been the target of sustained conservation efforts in recent years. Accurately monitoring Grévy's zebra populations is essential for ecologists to evaluate ongoing conservation initiatives. Recently, in both 2016 and 2018, a full census of the Grévy's zebra population was enabled by the Great Grévy's Rally (GGR), a citizen science event that combines teams of volunteers to capture data with computer vision algorithms that help experts estimate the number of individuals in the population. A complementary, scalable, cost‐effective and long‐term Grévy's population monitoring approach involves deploying a network of camera traps, which we have done at the Mpala Research Centre in Laikipia County, Kenya. In both scenarios, a substantial majority of the images of zebras are not usable for individual identification due to ‘in‐the‐wild’ imaging conditions—occlusions from vegetation or other animals, oblique views, low image quality and animals that appear in the far background and are thus too small to identify. Camera trap images, without an intelligent human photographer to select the framing and focus on the animals of interest, are of even poorer quality, with high rates of occlusion and high spatiotemporal similarity within image bursts. We employ an image filtering pipeline incorporating animal detection, species identification, viewpoint estimation, quality evaluation and temporal subsampling to compensate for these factors and obtain individual crops from camera trap and GGR images of suitable quality for re‐ID. We then employ the local clusterings and their alternatives (LCA) algorithm, a hybrid computer vision and graph clustering method for animal re‐ID, on the resulting high‐quality crops. Our method processed images taken during GGR‐16 and GGR‐18 in Meru County, Kenya, into 4142 highly comparable annotations, requiring only 120 contrastive same‐vs‐different‐individual decisions from a human reviewer to produce a population estimate of 349 individuals (within 4.6 of the ground truth count in Meru County). Our method also efficiently processed 8.9M unlabelled camera trap images from 70 camera traps at Mpala over 2 years into 685 encounters of 173 unique individuals, requiring only 331 contrastive decisions from a human reviewer.

  PublisherOA PDFDOI

• DataS^3: Dataset Subset Selection for Specialization

  ArXiv.org · 2025-04-22

  preprintOpen accessSenior author

  In many real-world machine learning (ML) applications (e.g. detecting broken bones in x-ray images, detecting species in camera traps), in practice models need to perform well on specific deployments (e.g. a specific hospital, a specific national park) rather than the domain broadly. However, deployments often have imbalanced, unique data distributions. Discrepancy between the training distribution and the deployment distribution can lead to suboptimal performance, highlighting the need to select deployment-specialized subsets from the available training data. We formalize dataset subset selection for specialization (DS3): given a training set drawn from a general distribution and a (potentially unlabeled) query set drawn from the desired deployment-specific distribution, the goal is to select a subset of the training data that optimizes deployment performance. We introduce DataS^3; the first dataset and benchmark designed specifically for the DS3 problem. DataS^3 encompasses diverse real-world application domains, each with a set of distinct deployments to specialize in. We conduct a comprehensive study evaluating algorithms from various families--including coresets, data filtering, and data curation--on DataS^3, and find that general-distribution methods consistently fail on deployment-specific tasks. Additionally, we demonstrate the existence of manually curated (deployment-specific) expert subsets that outperform training on all available data with accuracy gains up to 51.3 percent. Our benchmark highlights the critical role of tailored dataset curation in enhancing performance and training efficiency on deployment-specific distributions, which we posit will only become more important as global, public datasets become available across domains and ML models are deployed in the real world.

  PublisherOA PDFDOI

• Author response for "Essential tools but overlooked bias: Artificial intelligence and citizen science classification affect camera trap data"

  2025-06-13

  peer-review
  PublisherDOI

• Seeing Above and Below the Canopy: Modeling and Interpreting Species Occupancy with Multimodal Habitat Representations

  bioRxiv (Cold Spring Harbor Laboratory) · 2025-09-11

  preprintOpen accessSenior author

  Abstract Effective conservation and restoration of species is an increasingly urgent priority. To design management strategies that improve species success, we need a solid understanding of the habitat characteristics that support it. Occupancy models are statistical tools that ecologists use to model these relationships from data. Yet, current models represent habitats with coarse-scale environmental variables that fail to capture important microhabitat features. We show that these limitations can be addressed by incorporating AI-derived, multimodal habitat representations from overhead satellite imagery and ground-level camera-trap imagery. Across geography and species, these representations yield more accurate out-of-sample predictions than models based on conventional covariates alone, and combining satellite and ground-level views provides complementary gains. To translate improved prediction into actionable ecological insight, we further introduce a method that makes black-box AI-derived habitat representations interpretable by summarizing key factors contributing to occupancy probability into text-based descriptions. We then generate a per-site score for each description, which can replace black-box features to transparently link discovered habitat elements to species occurrence while maintaining predictive performance. Our approach provides a path toward microhabitat-aware and interpretable species-habitat models that support restoration planning and management decisions. We implement our method in an open-source Python package bridging AI and statistical ecology.

  PublisherOA PDFDOI

• Pairwise Matching of Intermediate Representations for Fine-grained Explainability

  ArXiv.org · 2025-03-28

  preprintOpen accessSenior author

  The differences between images belonging to fine-grained categories are often subtle and highly localized, and existing explainability techniques for deep learning models are often too diffuse to provide useful and interpretable explanations. We propose a new explainability method (PAIR-X) that leverages both intermediate model activations and backpropagated relevance scores to generate fine-grained, highly-localized pairwise visual explanations. We use animal and building re-identification (re-ID) as a primary case study of our method, and we demonstrate qualitatively improved results over a diverse set of explainability baselines on 35 public re-ID datasets. In interviews, animal re-ID experts found PAIR-X to be a meaningful improvement over existing baselines for deep model explainability, and suggested that its visualizations would be directly applicable to their work. We also propose a novel quantitative evaluation metric for our method, and demonstrate that PAIR-X visualizations appear more plausible for correct image matches than incorrect ones even when the model similarity score for the pairs is the same. By improving interpretability, PAIR-X enables humans to better distinguish correct and incorrect matches. Our code is available at: https://github.com/pairx-explains/pairx

  PublisherOA PDFDOI

• Conservation changed but not divided

  Trends in Ecology & Evolution · 2025-04-24 · 2 citations

  letter
  PublisherDOI

• New frontiers in artificial intelligence for biodiversity research and conservation with multimodal language models

  Methods in Ecology and Evolution · 2025-08-27 · 4 citations

  articleOpen access

  Abstract The integration of artificial intelligence (AI) into biodiversity research and conservation is growing rapidly, demonstrating great potential in reducing the intensive human labour required for data preprocessing, thereby, facilitating larger data collections that offer ecological insights at unprecedented scales. However, most of these AI applications for biodiversity are still in the early stages of development, hindered by challenges inherent in real‐world datasets and the limited accessibility of these technologies to practitioners without extensive programming knowledge. The recent advent of multimodal language models, which can process and generate multiple data modalities, has significantly expanded the realm of possible AI applications in biodiversity research. These models have demonstrated the ability to classify species and recognize more complex concepts, such as animal postures and orientations, without prior exposure during training. Multimodal language models can also provide explanations for their predictions and interact with humans in natural language, thereby making them more transparent, intuitive and accessible to non‐specialists. Despite these advancements, the use of multimodal language models for biodiversity still needs to overcome unique barriers to application, including high computational and financial demands, reliance on prompt engineering for consistent model performance on large datasets and insufficient open‐source sharing of state‐of‐the‐art methods. This paper explores the transformative potential of multimodal language models for biodiversity research and discusses several possible applications in biodiversity research. We also discuss challenges to implement these models in real‐world conservation scenarios and propose directions for future research to overcome these hurdles. Our goal is to encourage robust discussions and research into the integration of multimodal language models to advance AI for biodiversity research and conservation.

  PublisherOA PDFDOI

Frequent coauthors

• Pietro Perona

  17 shared

• Dan Morris

  Google (United States)

  16 shared

• Grant Van Horn

  15 shared

• Neel Joshi

  Northeastern University

  11 shared

• Elijah Cole

  10 shared

• Guanhang Wu

  9 shared

• Jeff Clune

  9 shared

• Jonathan Huang

  Northwestern University

  9 shared

Labs

• MIT EECS - Sara Beery LabPI

Education

• Ph.D., Electrical Engineering and Computer Science

  Massachusetts Institute of Technology

  2009

• M.S., Electrical Engineering and Computer Science

  Massachusetts Institute of Technology

  2005

• B.S., Electrical Engineering and Computer Science

  University of California, Berkeley

  2001