Is it possible to finish a 60000×10000 matrix decomposition (NMF, PCA, etc) or completion in 6 seconds on your laptop’s matlab? Can we make it even faster by a simple distributable scheme?
How to summarize a huge-scale dataset (ratings, movie, action, text) in seconds by representative REAL instances (users, key frames, anchor word) rather than artificial basis?
How to learn graphical models like LDA only by distributed max and min operations rather than iterative algorithm?
Learning by selecting real instances from real world, is usually much faster, more robust and more convincing than learning by optimizing and creating artificial numbers.
Summarizing the world, is the way how human learn from it.
Please come to my “Divide-and-Conquer Anchoring” talk at ICDM this year at Dallas. We win the Best student paper award! My talk will be located at Houston Ballroom B Sheraton Dallas, between 16:00-18:00 this Sunday (Dec 8). Here is the paper’s PDF. But I will talk more in Sunday 🙂
Warning of Winter Storm and Freezing Rain in Dallas this weekend!