Researchers at the Museum of Natural History and Birkbeck College used pioneering machine learning techniques to train a computer model to identify the mysterious tooth. This pushes the origins of some of the group’s members back almost 30 million years.
Simon Wills, a Ph.D. student at the Museum of Natural History who led the study, said: “Previous studies suggested that maniraptorans existed during the Middle Jurassic period, but the actual fossil record is unknown. The evidence has been patchy and controversial, and along with fossils found elsewhere, the study suggests that the group had already achieved global distribution by this time.
“The teeth we analyzed currently contain the only troodontid and therizinosaur fossils recorded in the UK and are the oldest evidence of these dinosaurs anywhere in the world.”
Therizinosaurus is a large herbivorous dinosaur from the late Cretaceous period, characterized by long claw-like pincers. The unique look of this extinct animal is now included in the latest Jurassic World movie.
Previous studies have attempted to classify isolated teeth based on various statistical techniques, but have not always been particularly successful. The researchers behind the current study are working to improve this, demonstrating that machine learning models can achieve high accuracy in identifying isolated teeth from known taxa.
“The use of machine learning in vertebrate paleontology, although its use is increasing, is still in its infancy,” adds Simon.
“The main drawback is that it requires a comprehensive training dataset for the model to learn.”
“For our research, we are fortunate that a relatively large dataset of dinosaur tooth measurements is already available that can be used to train the model.”
To convert the information contained in the fossils into data that can be used by machine learning models, researchers first had to create 3D models of each tooth from CT scans. This was because the teeth were too small to be practical to measure by hand.
We trained three different models using thousands of tooth measurements of known dinosaur species. Each model analyzes the data differently and combines their results to provide the most probable ID for each tooth.
Compared to other statistical methods, machine learning models yielded more accurate results, increasing researchers’ confidence in being able to classify unidentified teeth.
As technological innovation continues and various initiatives such as digitization projects make more information available to create training datasets, machine learning becomes more relevant to investigate different paleontological problems. It may become generally applicable.
The results of this study were published in the journal Papers in Paleontology.
Note to editors
Natural History Museum media contact:
Phone: +44 (0)20 7942 5654 / 07799690151
Email: press@nhm.ac.uk
You can download the image here
natural history museum It is a world-leading scientific research center and the UK’s most visited indoor attraction. It has a vision of a future in which both people and the planet thrive, and is uniquely positioned to be a powerful champion for balancing the needs of humanity with those of the natural world.
It is one of the world’s most important scientific collections, with researchers worldwide having direct access to over 80 million specimens, and to date, through over 50 billion digital data downloads. His 350 scientists at the museum are finding solutions to global emergencies, from biodiversity loss to sustainable extraction of natural resources.
The museum uses its global reach and influence to fulfill its mission of creating defenders of the planet. In short, to inform, inspire and empower everyone to make a difference in nature. Welcoming millions of visitors each year, his website was hit 17 million times last year, and his touring exhibitions have been visited by about 20 million people over the past decade.
