Many thanks to NVIDIA for supplying GPU hardware for our formidable new GPU cluster, which features five GTX280s, five Tesla C1060s and two Tesla S1070 quad GPU 1U rack modules.
We’ll be making these systems available for the students of MIT’s “CUDA @ MIT” independent activities period, in order to help spread the joy of GPU to our friends down the river.
Our latest results on the “Labeled Faces in the Wild” face recognition were just posted to the official LFW website (http://vis-www.cs.umass.edu/lfw/results.html), where we currently have the best reported performance with our “V1-like” model. However, at the same time, we argue that even more realistic test sets are needed – to find out more, see our latest CVPR paper.
In our new paper in PLoS Computational Biology, we describe our efforts to use GPU-accelerated high-throughput screening to find powerful new, biologically-inspired computer vision models.
Also check out the ResearchCast describing the work:
In our new paper in PNAS, we present evidence that rats can perform high-level transformation-tolerant visual object recognition, an ability not traditionally ascribed to rodents. We believe that the availability of rats as a model system for high-level has great potential to accelerate our understanding of high-level vision, by providing a simpler, more accessible example system to study.