Computational Connectomics Group

Understanding the structure and function of the nervous system is an exceptionally complex task: the system consists of thousands of cells connected to thousands of other cells in microscopic networks that extend over large volumes and exhibit a seemingly endless variety of behaviors. We believe that mapping such networks at the level of synaptic connections, and understanding the relation of their connectivity and geometry to function, will play a key role in unraveling the mystery of thought.

Our group’s goal is to create, based on such microscopic connectivity and functional data, new mathematical models explaining how neural tissue computes. Our modeling spans the connectomics gamut from the behavior of individual neurons in exiguous circuits to collections of neurons in increasingly complex networks. We collaborate with neurobiologists to design experiments based on our theoretical models, and work extensively to analyze the resulting data in order to confirm or disprove our theoretical predictions.

News

June 2019: Cross-Classification Clustering: An Efficient Multi-Object Tracking Technique for 3-D Instance Segmentation in Connectomics
by Yaron Meirovitch, Lu Mi, Hayk Saribekyan, Alexander Matveev, David Rolnick, Nir Shavit; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2019, pp. 8425-8435.
March 27, 2018: Blog on “Deep Learning to Study the Brain to Improve Deep Learning” is Live.
January 2017: Shavit Lab’s PPoPP 2017 paper,
A Multicore Path to Connectomics-on-Demand is selected for
Best Paper Nominee.

June 3, 2016, MIT Commencement: Congratulations to new graduates,
Gregory Odor and Hayk Saribekyan!
(Pictured: Hayk Saribekyan and Professor Nir Shavit.)

February 2016: The Shavit Lab has been awarded research funding under the IARPA Machine Intelligence from Cortical Networks (MICrONS) project.

October 2015: Brain-like chip.

Projects

High Throughput Connectomics
The current design trend in large scale machine learning is to use distributed clusters of CPUs and GPUs with MapReduce-style programming. Some have been led to believe that this type of horizontal scaling can reduce or even eliminate the need for traditional algorithm development, careful parallelization, and performance engineering. This paper is a case study showing the contrary: that the benefits of algorithms, parallelization, and performance engineering, can sometimes be so vast that it is possible to solve “clusterscale” problems on a single commodity multicore machine.

Connectomics is an emerging area of neurobiology that uses cutting edge machine learning and image processing to extract brain connectivity graphs from electron microscopy images. It has long been assumed that the processing of connectomics data will require mass storage, farms of CPU/GPUs, and will take months (if not years) of processing time. We present a high-throughput connectomics-ondemand system that runs on a multicore machine with less than 100 cores and extracts connectomes at the terabyte per hour pace of modern electron microscopes.

Mi, Lu, Wang, Hao, Tian, Yonglong, and Shavit, Nir. Training-Free Uncertainty Estimation for Neural Networks. arXiv: 1910.04858, 2019.
Rosenfeld, Jonathan S., Rosenfeld, Amir, Belinkov, Yonatan, and Shavit, Nir. A Constructive Prediction of the Generalization Error Across Scales. To appear in Proceedings of the International Conference on Learning Representations (ICLR 2020), Addis Ababa, Ethiopia, April 2020. Also, arXiv:1909.12673, September 2019.
Meirovitch, Yaron, Mi, Lu, Saribekyan, Hayk, Matveev, Alexander, Rolnick, Rolnick, and Shavit, Nir. Cross-Classification Clustering: An Efficient Multi-Object Tracking Technique for 3-D Instance Segmentation in Connectomics. The IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 8425-8435, 2019.
Witvliet, Daniel, Mulcahy, Ben, Mitchell, James K., Meirovitch, Yaron, Berger, Daniel R., Holmyard, Douglas, Schalek, Richard L., Cook, Steven J., Xian Koh, Wan, Neubauer, Marianna, Rehaluk, Christine, Wang, ZiTong, Kersen, David, Chisholm, Andrew D., Shavit, Nir, Lichtman, Jeffrey W., Samuel, Aravinthan, and Zhen, Mei. Invariant, stochastic, and developmentally regulated synapses constitute the C. elegans connectome from isogenic individuals. Poster Presentation at Cosyne 2019.
Meirovitch, Yaron, Mi, Lu, Saribekyan, Hayk, Matveev, Alexander, Rolnick, David, Wierzynski, Casimir, and Shavit, Nir. Cross-Classification Clustering: An Efficient Multi-Object Tracking Technique for 3-D Instance Segmentation in Connectomics. CoRR abs/1812.01157, 2018.
Santurkar, Shibani, Budden, David M., and Shavit, Nir. Generative Compression. PCS 2018. Also, CoRR.abs/1703.01467, 2017.
Budden, David, Matveev, Alexander, Santurkar, Shibani, Chaudhuri, Shraman Ray, and Shavit, Nir. Deep Tensor Convolution on Multicores. ICML 2017. Also, CoRR abs/1611.06565, 2016.
Matveev, A., Meirovitch, Y., Saribekyan, H., Jakubiuk, W., Kaler, T., Odor, G., Budden, D., Zlateski, A., and Shavit, N. A Multicore Path to Connectomics-on-Demand. PPoPP 2017 (Best Paper Nominee).
Rolnick, David, Meirovitch, Yaron, Parag, Toufiq, Pfister, Hanspeter, Jain, Vien, Lichtman, Jeff W., Boyden, Edward S., and Shavit, Nir. Morphological error detection in 3d segmentations. CoRR.abs/1705.10882, 2017.
Rolnick, David, Veit, Andreas, Belongie, Serge J., and Shavit, Nir. Deep Learning is Robust to Massive Label Noise. CoRR abs/1705.10694, 2017.
Santurkar, Shibani, Budden, David, Matveev, Alexander, Berlin, Heather, Saribekyan, Hayk, Meirovitch, Yaron, and Shavit, Nir. Toward Streaming Synapse Detection with Compositional ConvNets. CoRR abs/1702.07386, 2017.
Meirovitch, Y., Matveev, A., Saribekyan, H., Budden, D., Rolnick, D., Odor, G., Knowles-Barley, S., Thouis, R., Pfister, H., Lichtman, J., Shavit, N.A Multi-Pass Approach to Large-Scale Connectomics. CoRR abs/1612.02120, 2016.
Shavit, Nir. A Multicore Path to Connectomics-on-Demand. SPAA 2016.
Lichtman, J., Pfister, H., and Shavit, N. The big data challenges of connectomics. Nature Neuroscience, 17, pp. 1448-1454, November 2014.
Allen-Zhu, Zeyuan, Gelashvili, Rati, Micali, Silvio, and Shavit, Nir. Sparse sign-consistent Johnson-Lindenstrauss matrices: Compression with neuroscience-based constraints. Proceedings of the National Academy of Sciences USA; 111(47), pp. 16872-16876, October 2014.

Mapping the Structure of Thought

News

Projects

Principal Investigator

Research Scientist

Postdoc

Graduate Students

Alumni

Publications

Resources