3 ms·
Yes, they do. Reminds me of the AI algorithm designed to tell US tanks from Soviet ones. They accidentally trained it on one type of film for US tanks, and anot
by tomsaffell 18y ago
Yes, they do. Reminds me of the AI algorithm designed to tell US tanks from Soviet ones. They accidentally trained it on one type of film for US tanks, and another type for Soviet tanks, each with different color tones, and ended up with a system that could spot different color tones...
http://books.google.com/books?id=JNhRSAMFn6YC&pg=PA271&lpg=PA271&dq=computer+vision+tanks+mistake+brightness&source=bl&ots=f8mLCJ1BsT&sig=SV1CSLMXal40j5yr8HDjSSM-M-c&hl=en&ei=lsTGScP7OIaNtgeZ64zJCg&sa=X&oi=book_result&resnum=1&ct=result http://books.google.com/books?id=JNhRSAMFn6YC&pg=PA271&#...
- endtime 18y agoThat's one of the problems with neural nets...there's usually no easy way for a human to understand what the actual decision process is (in sharp contrast to, say, decision trees).
- trapper 18y agoYes there is. There are a few methods of converting a trained nn to a decision tree. It's actually a better way of training decision trees for some problems.