Computer Vision News - September 2016

be the same. Yet you cannot reveal anything about the individuals from the k-means. There shouldn’t be any connection between compressing the data and adding this small noise to preserve their privacy, but there is such a connection. CVN: It sounds like we have a distinctive coreset in our mind. Feldman: Kind of… We always need to decide what is important and what is not. Think at how we manage our time, what we see, what we hear and so on. The idea of compression and compressing the right thing is problem- dependent. I am sure that we do it all the time for all the senses that we have. Some people do it better than others. CVN: What is the biggest breakthrough that you have seen? Feldman: The main breakthrough in this field is the fact that we now have a general framework coreset for any problem . Unfortunately, I think the coresets are still buried in theoretical computer science conferences. Engineers are not using them because there is very little code out there. I hope that in coming years we will have more implementations so that people who cannot read my papers can still run the coresets and evaluate them for their business or for their research. We want to bring the coreset to the people and bring the research into the industry. Computer Vision News Guest 23 Guest “ Coresets are a new paradigm. It is more a state of mind than an exact mathematical definition ” This image and the previous one were taken on June 28 at the Simons Institute for the Theory of Computing: Core-sets for Real-Time Tracking using Caratheodory Theorem, with Applications to Drones. The video of the conference is here .