Researchers from the Indian Institute of Technology Madras and Harvard University have developed a new machine learning algorithm to save wildlife.
According to him, hunting of animals can be prevented by joint and coordinated use of drones and forest rangers. They named their solution ‘CombSGPO’ (Combined Security Game Policy Optimization).
Because both the drone and the ranger are limited resources, the researchers developed an algorithm that provides highly efficient, scalable strategies. Once a resource boundary is identified, the algorithm works to handle resource allocation and create a patrolling strategy. For this it uses animal data population in the protected area and assumes that poachers are aware of patrols carried out at various sites.
Professor in the Department of Computer Science and Engineering. Balaraman Ravindran and head of the Robert Bosch Center for Data Science and Artificial Intelligence (RBCDSAI), IIT-M, collaborated with TeamCore – a research group led by Professor Milind Tambe at Harvard University. study.
“The only resources we consider are human patrols,” said Mr. Balaraman. [forest personnel] and surveillance drones, which are equipped with object detectors for animals and predators and can make strategic signals and communicate with each other as well as with human patrols.
The algorithm uses a game theory-based model. Game theory is concerned with predicting areas where hunting may occur based on first hunting events and interactions between predators and defenders.
Arvind Venugopal, first author of the study, said: “The game model and the kind of resources we use to simulate this kind of ‘illegal play’ between defenders.[foresters and drones] and attackers [poachers] They are based on the widely studied ‘Stackelberg Security Game Model’ and are linked to drones that have already been deployed.”
The team is looking to learn sample-efficient multi-agent reinforcement learning with minimal data so that it can be applied in domains such as security, search and rescue, and aerial mapping for agriculture.
The data collection for such work in a real-world scenario is expensive, the researchers said.