Past Research

In the neuroscience field, novel experimental techniques and increased availability of inexpensive high-speed video cameras are generating data at an unprecedented rate. In particular, advances in microscopy facilitate imaging large brain areas and more complex behaviors with fine time resolution producing data at rates surpassing 1 TB/hour. The resulting data deluge represents a bottleneck in science progress and calls for formal analysis methods that are reliable, fully automated, efficient, and scalable to datasets generated over the course of multiple days. Against this background, Dr. Giovannucci‘s research (@Flatiron Institute, Simons Foundation) has focused on developing algorithms and tools for the analysis of large imaging datasets and animal behavior. This effort has culminated in an open source software suite – CaImAn – that is widely employed by the neuroscience community and includes algorithms to solve several preprocessing problems.

Large scale analysis of calcium imaging data

CaImAn approaches the problem of big data with a map-reduce approach, where movies are processed in small chunks, therefore preventing memory issues and enabling parallelization. By applying advanced machine learning and computer vision methods to pre-processing problems, including motion correction, source separation, and deconvolution, this suite allows individual multi-core machines to process large-scale datasets, and provides user-friendly means of data review and modification.  CaImAn provides an intuitive programming interface and several demonstration examples.

Simplified representation of the analysis pipeline for calcium imaging data

 

Giovannucci, A., Friedrich, … & Pnevmatikakis, E. A. (2018). CaImAn: An open source tool for scalable Calcium Imaging data Analysis. bioRxiv, 339564. Python and Matlab Software. Article.

Online real-time analysis of calcium imaging data

Another approach to the big data problem in calcium imaging is to analyze movies incrementally, i.e. one frame at a time. This has the double advantage of reducing memory requirements to a minimum and to enable real-time closed-loop experiments.  CaImAn is also equipped with an algorithm to process data in real-time and streaming fashion. The underlying algorithm combines online optimization strategies and deep neural networks.

Online analysis of calcium imaging data. From top left to bottom right. Raw data, residual, detected components, denoised reconstruction. White squares are proposed regions, purple squares are regions accepted by the deep network. Data courtesy, Tank Lab.

Giovannucci, A., Friedrich, … & Pnevmatikakis, E. A. (2018). CaImAn: An open source tool for scalable Calcium Imaging data Analysis. bioRxiv, 339564. Python and Matlab Software. Article.

Giovannucci, A.*, Friedrich, J.*,  .., & Pnevmatikakis, E. A. (2017). Onacid: Online analysis of calcium imaging data in real time. In Advances in Neural Information Processing Systems (pp. 2381-2391). Software. Article.

Pnevmatikakis, E. A., & Giovannucci, A. (2017). NoRMCorre: An online algorithm for piecewise rigid motion correction of calcium imaging data. Journal of neuroscience methods, 291, 83-94. Python and Matlab Software. Article.

Automatic gesture tracking in head-fixed mouse preparation

CaImAn also provides some routines for the analysis of animal behavior. The underlying unsupervised algorithms combines optical flow and nonnegative matrix factorization to segment stereotypical movements in head-fixed mouse.

Giovannucci, A., Pnevmatikakis, … & Masip, D. (2018). Automated gesture tracking in head-fixed mice. Journal of neuroscience methods, 300, 184-195. Software. Article.

Online principal component analysis

Big data problems (as in the case of calcium imaging for instance) frequently require processing datasets in a streaming fashion, either because all data are available at once but collectively are larger than available memory or because the data intrinsically arrive one data point at a time and must be processed online. This project introduced a computationally efficient version of similarity matching, a framework for online dimensionality reduction that incrementally estimates the top K-dimensional principal subspace of streamed data while keeping in memory only the last sample and the current iterate. Open source code is available in Python and Matlab.

Giovannucci, A.*, Minden, V.*, Pehlevan, C., & Chklovskii, D. B. (2018). Efficient Principal Subspace Projection of Streaming Data Through Fast Similarity Matching. IEEE Big Data 2018. In press. Article. Software in Matlab and Python

Neuroprosthetics

Within the EU funded project ReNaChip, aimed at replacing compromised brain functions with a neuroprosthetic device, Dr. Giovannucci conceived signal-processing algorithms to decode neural activity in real-time and an artificial cerebellum that restores lost learning functions. To witness the success of his endeavor, the cerebellar model he implemented was subsequently employed in the new generation of neuroprosthetic devices stemming from the original project (Hogri et. al, Scientific Reports, 2015).

S. A. Bamford, R. Hogri, A. Giovannucci, A. H. Taub, I. Herreros, P. F. M. J. Verschure, M. Mintz, and P. Del Giudice, A VLSI field-programmable mixed-signal array to perform neural signal processing and neural modeling in a prosthetic system, IEEE Trans Neural Syst Rehab Eng, vol. 20, no. 4, pp. 455–467, Jul. 2012.

Herreros, A. Giovannucci, A. H. Taub, R. Hogri, A. Magal, S. Bamford, R. Prueckl, and P. F. M. J. Verschure, A Cerebellar Neuroprosthetic System: Computational Architecture and in vivo Test, Front Bioeng Biotechnol, vol. 2, pp. 14–14, Jan. 2014.

Engineering the organization of complex systems

During his PhD, Dr. Giovannucci developed mechanisms and algorithms for the self-organization of complex systems that have found application in industrial procurement scenarios, received a competitive prize from the EU and resulted in his PhD to be shortlisted for the best European thesis in Artificial Intelligence. 

Giovannucci, J. Cerquides, and J. A. Rodriguez-Aguilar, Composing supply chains through multiunit combinatorial reverse auctions with transformability relationships among goods, Systems, Man and Cybernetics, Part A, IEEE Transactions on, vol. 40, no. 4, pp. 767–778, 2010.

Giovannucci, J. A. Rodriguez-Aguilar, A. Reyes, F. X. Noria, and J. Cerquides, Enacting agent-based services for automated procurement, Engineering Applications of Artificial Intelligence, vol. 21, no. 2, 2008.

J. Cerquides*, U. Endriss*, A. Giovannucci*, and J. A. Rodríguez-Aguilar*, Bidding languages and winner determination for mixed multi-unit combinatorial auctions, IJCAI’07: Proceedings of the 20th International Joint Conference on Artificial Intelligence, 2007.

A. Giovannucci, J. A. Rodriguez-Aguilar, J. Cerquides, A. Reyes, and F. X. Noria, iBundler: an agent-based decision support service for combinatorial negotiations, AAAI’04: Proceedings of the 19th National Conference on Artificial intelligence, 2004.

A. Giovannucci, J. A. Rodriguez-Aguilar, A. Reyes, F. X. Noria, and J. Cerquides, Towards Automated Procurement via Agent-Aware Negotiation Support, AAMAS ’04: Proceedings of the 3rd Intl. Conference on Autonomous Agents and Multiagent Systems, 2004, vol. 1.