I worked with Nir Shavit on a computational neuroscience project - we're extracting a full connectome (3D map of a brain, with all neurons, synapses etc.) of a mammalian brain. Majority of my work is at the intersection of "big data", machine learning (computer vision/image segmentation) and concurrent algorithms.

Feel free to email me if you would like to talk about any of my interests or projects. For now, check out my public GitHub profile: victorj44

You can read my blog with more updates here: www.jakubiuk.net/blog/

Stuff:

• OnSpecta - Deep Learning acceleration software for Caffe and Tensorflow. Co-founder & Chief-Scientist.
  
  
• A Multicore Path to Connectomics-on-Demand - some of the research results published in PPoPP '17.
  
  
• Masters thesis - my Masters research thesis.
  
  
• Hocrux - handwritten optical character recognition. A hackweek project, uses NVIDIA Digits and Caffe to train GoogLeNet classifier. Achieved reasonable accuracy for hand-written english. Joint work with Jon from Dropbox.
  
  
• Computational Connectomics Group - the group I was involved with.
  
  
• engine.datanitro.com - NitroEngine - an alternative computational engine to MS Excel that speeds up statistical analysis (ie. Monte Carlo models) significantly.
  
  
• voyager.datanitro.com - DataNitro Voyager - the simplest REST API to push data to Excel users and their spreadsheets, without having to touch the Windows/Office stack (currently defunct).
  
  
• www.datanitro.com - replaced VBA in Excel with Python. YCombinator backed startup (S12) - founder.
  
  
• Implementation and Performance Analysis of Hash Functions and Collision Resolutions - spring '12 - an implementation project & a paper - performance analysis of different collision resolutions (Linear Probing, Quadratic Probing, Cuckoo hashing and Hopscotch hashing) w.r.t cache performance across multiple cores. The code is available on my github.
  
  
• www.tourmadillo.com - winter & spring '12 - a web & mobile app for travel (currently defunct).
  
  
• Parellization of Dijkstra's Algorithm - Dec '11 - a paper on the Relax Heaps data structure; describes parallelization of the shortest paths algorithm to (efficiently) scale across multiple CPU cores.
  
  
• Parallel memory allocator - Fall '11 - a Hoard-like parallel memory allocator for one of my clas ses. See the source here, it's pretty well commented. Provides concurrent (thread-safe) malloc and free and scales reasonably well on multiple cores.
  
  
• Introduction to C++ - Jan '11 - an intro C++ course I taught during the IAP. You can find the study materials on MIT OCW.
  
  
• Skimmedmilk - Dec '10 - a dinner signup hack (website) for one of my dorms. Written in good old php, the code is quite nasty, but hey, it worked. Uses MIT certificates for auth.