[2] During her time in the lab, she published two first author scientific papers, one in the Journal of Neuroscience which offered a novel protocol for generating human limb-innervating neural subtypes in vitro for use in neurological disease research,[3] and the other in Nature Biotechnology on benchmarking iPS stem cell lines ability to make motor neurons.
[5] In her postdoctoral work, Mathis focused on pioneering deep learning tools for neural and behavioral analysis which served as a critical step towards her independent career.
[8] In 2017, Mathis started her lab at the Rowland Institute at Harvard University with a goal of reverse engineering neural circuits that drive adaptive motor behavior.
This tool relies on transfer learning to optimize an ImageNet-pretrained foundation model and the feature detectors from DeeperCut,[10] to fine-tune on a desired new dataset after sufficient training.
[12] Mathis has shown the versatility of this tool on many diverse datasets highlighting the robust design and potential for wide use in fields even beyond neuroscience.