Sven Dorkenwald will present his General Exam "A spatial graph database for proofreading and analyzing dynamic segmentations in connectomics" on May 16, 2019 at 10am in in PNI 130. The members of his committee are as follows: Sebastian Seung (adviser), Michael Freedman, and Kai Li. Everyone is invited to attend his talk, and those faculty wishing to remain for the oral exam following are welcome to do so. His abstract and reading list follow below. Title: A spatial graph database for proofreading and analyzing dynamic segmentations in connectomics Connectomics is concerned with high resolution imaging of brain tissue and analysis of the neurons, their connectivity and ultrastructure within these datasets. Ongoing improvements in volume electron microscopy and automatic segmentation have created large, fully reconstructed datasets with cubic-millimeter sized datasets (PBs of data) in reach. However, current automated segmentations still contain errors prohibiting immediate analysis and requiring proofreading, both automated and manual. Proofreading such segmentations can be viewed as a graph problem with nodes representing supervoxels (SVs), accumulations of voxels, and neurons being connected components in this graph. Proofreading edits add and remove edges and trigger recomputations of connected components (CCs). Interactive proofreading requires fast CC computations and two-way lookups (CC id <-> SV id), as well as versioning, multiuser support, and spatially restricted queries. We present a database management system, the PyChunkedGraph, which enables interactive proofreading of graphs from large datasets (more than tens of billions of nodes and hundreds of billions of edges). We achieve this by decoupling the local nature of proofreading changes from the spatially distributed nature of neurons by implementing an octree on top of the SV graph storing the CC information. Hence, changes to the graph only need to propagate through the height of the tree lending it great scalability. We implemented the PyChunkedGraph with Google’s BigTable making use of its timestamp feature for versioning and read-modify-write guarantees for locking. We present performance on large automatically segmented datasets. Reading list Papers: DVID: Distributed Versioned Image-Oriented Dataservice Katz WT, Plaza SM; Frontiers in Neural Circuits, 2019 NeuTu: Software for Collaborative, Large-Scale, Segmentation-Based Connectome Reconstruction Zhao T, Olbris DJ, Yu Y, Plaza SM; Frontiers in Neural Circuits, 2018 NeuroBlocks – Visual Tracking of Segmentation and Proofreading for Large Connectomics Projects Al-Awami AK et al.; IEEE Transactions on visualization and computer graphics, 2016 Design and Evaluation of Interactive Proofreading Tools for Connectomics Haehn D et al.; IEEE Transactions on visualization and computer graphics, 2014 Dense connectomic reconstruction in layer 4 of the somatosensory cortex Motta A, Berning M, Boergens KM, Staffler B, et al.; bioarxiv, 2018 Bigtable: A Distributed Storage System for Structured Data Chang F, et al., 2006 Large-scale Incremental Processing Using Distributed Transactions and Notifications Peng D and Dabek F; OSDI'10 Proceedings of the 9th USENIX conference on Operating systems design and implementation, 2010 Pregel: A System for Large-Scale Graph Processing Malewicz et al., SIGMOD '10 Proceedings of the 2010 ACM SIGMOD International Conference on Management of data, 2010 Functional selectivity and specific connectivity of inhibitory neurons in primary visual cortex Znamenskiy P, Kim MH, Muir DR et al.; bioarxiv, 2018 Single-neuron perturbations reveal feature-specific competition in V1 Chettih SN and Harvey CD; nature, 2019 Functional specificity of local synaptic connections in neocortical networks Ko H, Hofer SB et al.; nature, 2011 Book: Principles of Neurobiology, Luo et al.