Join Transform 2021 this July 12-16. Register for the AI event of the year.

AI-driven activity recognition has been used to suss out emotions from walking style, estimate how many people are in a room, and create videos from start and end frames alone. But that’s not all it’s good for. In a new paper (“A Framework for Identifying Group Behavior of Wild Animals“) published on the preprint server, scientists at the University of Illinois, Politecnico di Milano in Italy, and the University of California, Davis propose a statistical framework for identifying wild animal group behavior. They say that in preliminary experiments, it exhibits “significantly [better]” classification accuracy compared with baseline methods.

“Understanding animal behavior is central to answering the fundamental question of why do animals (including humans) do what they do,” wrote the coauthors. “Recently, biologist[s] started to use wearable technologies, such as GPS, accelerometers, and radio sensors, to track animals and their activities. However, the collected raw data are not human-interpretable and needs to be processed to extract behavioral patterns … Activity recognition models can be used to learn the relations between the raw time series and the behavioral annotations collected through observations or other modalities.”

As the researchers explain, time series classification — i.e., classification of quantity values obtained at successive times, often with equal intervals between them — is most often tackled using temporal sequence analysis or machine learning. The former is based on an explicit description of the raw signals, while the latter automatically infers features from input data.

The researchers’ approach employs a two-step sequence analysis process: First, they select the “best” global temporal resolution (the measurement resolution with respect to time) for a given corpus, and then they encode social relations among animal groups by extracting topological and relational components relevant for classification.

In experiments, the team sourced a publicly available data set of baboon group activities containing 26 animals tracked for 35 days. They defined the baboons’ social network based on proximity, such that baboons passing within two meters of each other were considered to have interacted.

The researchers report that their approach achieved accuracy roughly 10% better than previous methods, and that adding social information resulted in a 7% improvement over the initial results.

“Our evaluation on a real-world data set shows that the proposed framework better identifies the complex behavioral dynamics of groups of wild animals,” wrote the coauthors, who say they plan to include other data sets in future work. “We are currently working on extending the temporal resolution step to a more dynamic approach allowing varying temporal steps, which will allow to better identify the critical components of each different behaviors.”


VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative technology and transact. Our site delivers essential information on data technologies and strategies to guide you as you lead your organizations. We invite you to become a member of our community, to access:
  • up-to-date information on the subjects of interest to you
  • our newsletters
  • gated thought-leader content and discounted access to our prized events, such as Transform 2021: Learn More
  • networking features, and more
Become a member