MIT doctoral student Justin Kay has introduced a groundbreaking approach to AI model selection known as CODA, which stands for “consensus-driven active model selection.” This creative approach allows conservationists to make an informed choice about the best AI model to apply to analyzing ecological data. It is especially aimed at photos taken by wildlife monitoring cameras.
The development of CODA emerged from collaboration between Kay and his colleagues at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT and the University of Massachusetts Amherst. With this new approach using modern technology, the data-crunching process gets accelerated and streamlined, creating a much more efficient workflow for researchers and conservationists alike.
Today, there are around 1.9 million pre-trained models at the HuggingFace Models hub. So how do you choose the best model from this deep, wide pool? CODA streamlines this process by aggregating predictions from multiple AI models. This method usually yields deeper findings than relying solely on a single model’s output.
In real-world implementations, CODA requires minimal input and can be used to find the leading model for any task. Users may only want to annotate 25 examples. This number, ideally small, is useful to them in picking an optimal model among their candidates of choice. This feature is particularly advantageous in wildlife ecology, where researchers often deal with limited labeled data, sometimes as few as 50 images.
Kay spoke about CODA’s promise at a recent study on tracking and safeguarding fragile ecosystems. The research focused on the application of AI methods to interpret data gathered from camera traps set out in animals’ natural environments. By addressing model selection, CODA increases the validity of ecological analyses, thereby increasing the effectiveness of conservation efforts that rely on such analyses.
At the International Conference on Computer Vision (ICCV 2025) held this October, CODA’s influence was unmistakable. It was even recognized as a Highlight Paper, evidence of the significance of TBI’s work to the field. The research has been made publicly accessible through arXiv with the DOI: 10.48550/arxiv.2507.23771.