How can agents navigate physical and abstract environments? Many fundamental problems in various disciplines can be formulated as a navigating a complex environment. Neuroscientists and cognitive scientists, for example, hypothesize that the mind connects concepts by their relationships on an abstract map. Reasoning occurs by navigating this abstract, mental map. Many concepts can be formulated as abstract navigation. Programming languages—from computer science—and games—from economics—are equivalent to navigation in an abstract space—such as a graph.
Neural networks—in the context of deep learning—can solve complex tasks such as classifying real-world images and championing world-renowned Go players. Given the success of neural networks, I research how neural networks can navigate in complex, abstract settings. To study this question, I use differential geometry, statistics, and neuroscience.
Ph.D. in Computation and Neural Systems, Division of Physics, Mathematics, and Astronomy
B.S. in Biomedical Engineering, School of Engineering and Applied Science
Conference Proceeding: Dawna Bagherian, James Gornet, Jeremy Bernstein, Yu-Li Ni, Yisong Yue, Markus Meister. “Fine-Grained System Identification of Nonlinear Neural Circuits,” in Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining, 2021.
Conference Proceeding: Yujia Huang, James Gornet, Sihui Dai, Zhiding Yu, Doris Y. Tsao, Anima Anandkumar. “Neural Networks with Recurrent Generative Feedback,” in Advances in Neural Information Processing Systems 32, 2020.
Conference Proceeding: James Gornet, Kannan Umadevi Venkataraju, Arun Narasimhan, Nicholas Turner, H. Sebastian Seung, Pavel Osten, Uygar Sümbül. “Reconstructing neuronal anatomy from whole-brain images.” 2019 IEEE 16th International Symposium on Biomedical Imaging.
Poster Presentation: James Gornet, Kannan Umadevi Venkataraju, Uygar Sümbül, Pavel Osten. “Generating brain atlases across diverse brain sample types.” Society for Neuroscience, San Diego, CA. November 2018.
Poster Presentation: Kannan Umadevi Venkataraju, James Gornet, Janelle Collins, Zainab Khaku, Kadeem Joseph, Nicholas Cain, Pavel Osten. “Whole mouse brain light sheet atlas with iDISCO+ in CCF space.” Society for Neuroscience, Washington, DC, November 2017.
International Symposium on Biomedical Imaging
ISBI 2019 announces the availability of ten travel awards made available through a grant from the US National Institutes of Health, National Institute of Biomedical Imaging and Bioengineering.
The International Genetically Engineered Machine Foundation
The International Genetically Engineered Machine competition is the premiere student competition in Synthetic Biology. Since 2004, participants of the competition have experienced education, teamwork, sharing, and more in a unique competition setting. The 2016 Columbia University iGEM team's project is the design of a genetically engineered mosquito repellent. The repellent is a di-rhamnolipid and is excreted from the P. putida bacteria strain.
Society for Science & the Public
Since 1942, first in partnership with Westinghouse and since 1998 with Intel, SSP has provided a national stage for the country's best and brightest young scientists to present original research to nationally recognized professional scientists. James Gornet was awarded semifinalist for his work in molecular computing and his article “A biological architecture for emulating a central processing unit using serine integrases and transcription factors.”
Columbia University School of Engineering and Applied Science
The C.P. Davis Scholars Program was established in recognition of alumnus C. P. Davis, School of Mines Class of 1922, for his many years of loyalty, leadership and active support of The Fu Foundation School of Engineering and Applied Sciences and Columbia Engineering Alumni Association (CEAA).
Whole-brain imaging offers as a vast amount of information but also increases the amount of required data to analyze. To tackle this problem, I first designed and implemented a convolutional neural network for semantic segmentation of neurons from whole-brain image volumes. To compare data across a wide variety of samples and imaging modalities, I optimized an image registration algorithm to align sample image volumes to the Allen Mouse Brain Atlas. I am pleased that my work led to presentations at the Society for Neuroscience conferences in Washington, D.C. and San Diego, CA.
Reconstructing multiple molecularly defined neurons can reveal organizational principles of the nervous system. Oblique light-sheet microscopy offers a fast, versatile method for acquiring mouse whole-brain images. To reconstruct the neurons in these images, I implemented a convolutional neural network. To prevent topological errors, I developed a connectivity-based regularization method as well as simulated image artifacts. I integrated these techniques into a framework for efficient, scalable processing. To show the merit of this approach, I reconstructed the neurons in a section of an Emx1-Cre mouse cortex section. I am pleased to have submitted a paper on these methods to the 2019 IEEE International Symposium on Biomedical Imaging.
Molecular computing is a method of computing that uses DNA, molecular biology, and biochemistry. In high school, I worked on an independent project to develop a more tractable molecular computer. Using the DNA transistor developed in Drew Endy’s lab, I modeled the molecular computer after a traditional central processing unit. I designed a plasmid such that GFP transcription corresponded to binary numbers. As a result of this research, the Intel Science Talent Search awarded me semifinalist.
NeuroTorch is a framework for reconstructing neuronal morphology from optical microscopy images. It interfaces PyTorch with different automated neuron tracing algorithms for fast, accurate, scalable neuronal reconstructions. It uses deep learning to generate an initial segmentation of neurons in optical microscopy images. This segmentation is then traced using various automated neuron tracing algorithms to convert the segmentation into an SWC file—the most common neuronal morphology file format. NeuroTorch is designed with scalability in mind and can handle teravoxel-sized images.