Glossary of Computer Science and Ecology Terms¶

For implementation examples and further resources, please see the ABC Center Quick-start to Python for Ecologists.

Active learning¶

Sets of methodologies that aim to optimize data collection strategies, or to select the most informative samples among a large quantity of redundant ones by using the uncertainty of the AI model as a guiding principle. Active learning models are iterative by nature and aim to increase model confidence by adding examples to the training data in successive improvement cycles.

Artificial Intelligence (AI)¶

Any computer algorithm that makes predictions, recommendations, or decisions on the basis of a defined set of objectives. Most popular AI currently falls under the subcategory of machine learning, though symbolic AI represents another impactful subcategory.

Benchmarks¶

Standardized, structured challenges posed to the AI community, where methods can be directly compared and improvement can be measured. These often take the form of a fixed dataset, split into training, validation, and testing subsets, alongside a carefully designed metric or set of metrics used to evaluate success. Benchmarks must be carefully designed to ensure they are as representative as possible of the real-world challenge they pose. An example of such a benchmark is WILDS: https://wilds.stanford.edu/.

Category discovery¶

Related to open world/open set classification. A challenge in machine learning where the goal is to identify and group previously unknown or unlabelled categories within a dataset, allowing the model to autonomously discover and define new classes based on patterns or similarities in the data alongside (in some cases) existing domain knowledge. This approach contrasts with open world recognition, where the goal is just to flag data as previously unknown, not also cluster it.

Computer vision¶

A field of artificial intelligence that enables machines to interpret, analyze, and understand visual information from the world. It involves developing algorithms and models to process images and videos, enabling tasks such as image classification, object detection, semantic or instance segmentation, 3D reconstruction, depth estimation, visual question answering, image retrieval, and scene understanding.

Edge computing¶

Computing framework where the processing is performed at the level of the device acquiring the data itself. Instead of downloading the data and processing it on a remote device, an edge computing system will process the data at the sensor level and only broadcast the results. This makes perfect sense in situations where e.g., the sensor device needs the results to control its behavior, when a sensor acquires a lot of unusable data or when the bandwidth for transmitting data is low.

Foundation models¶

Machine learning models that are trained on a wide variety of data with the goal of being useful across a variety of different problems.

Generative AI¶

AI algorithms designed to generate content, often images or text, as opposed to discriminative AI, used to provide information about existing data.

Imageomics¶

An emerging field in which machine learning tools built around biological knowledge are used to analyze image data to characterize patterns and gain insights into traits and relationships at individual, population, and species scales.

Machine learning¶

A subcategory of AI in which algorithms ingest data and "learn" to pick out patterns in the data that are relevant for solving the problem at hand. Machine learning includes tools such as random forests and Bayesian methods, as well as the area of deep learning which has grown extensively in recent years.

Multimodal data¶

Data observing the same entity with a variety of sensors. One modality could be an on-animal sensor, a drone image, or a microphone.

Niche¶

The role or position of an organism within an ecosystem, including its interactions with other species and its environment. It encompasses the range of conditions and resources necessary for the organism's survival, growth, and reproduction, as well as its impact on the ecosystem. A niche includes factors such as the organism's habitat, diet, behavior, and its relationships with other organisms (e.g., predators, prey, competitors).

Trait¶

Phenotypic attributes that impact an organism's fitness and/or influence their ecosystem functions. For plants, leaf, height, and seed characteristics capture strategies for resource allocation, competition, and environmental change. Similarly, traits such as body mass for animals could be surrogates for physiology, behavior, and interactions. Traits also provide insights into the consequences of biodiversity loss for ecosystem functioning and human well-being.

Open world/Open set classification¶

A machine learning approach where the model is designed to classify known classes while also recognizing whether new data belongs to unknown or unseen classes, which were not present during training. This approach contrasts with closed set classification, where the model assumes that all test data belongs to a predefined set of known classes.

Species Distribution Models¶

A family of models (a.k.a. environmental niche models) that predict species occurrence or abundance as a function of abiotic or biotic environmental variables. These models are key tools for ecological inference of species responses to the environment and for mapping present and projected future distribution of species with climate or land-use change.

Borrowed with Permission from: Laura J. Pollock, Sara Beery, Kaitlyn Gaynor, Marta A. Jarzyna, Oisin Mac Aodha, Bernd Meyer, David Rolnick, Graham W. Taylor, Devis Tuia, Tanya Berger-Wolf and Justin Kitzes. Harnessing AI to fill global biodiversity shortfalls. Nat. Rev. Biodivers. 1, 166–182 (2025). https://doi.org/10.1038/s44358-025-00022-3