I attended an interesting talk on by Daniel Lopresti on
a new approach to machine perception at the
BYU Computer Science colloquium on Thursday. Machine perception refers to the ability of computers to mimic human behavior for tasks such as computer vision, document analysis, image processing, speech recognition, and natural language understanding.
Dr. Lopresti is advocating many approaches that we have discussed as part of the free software movement:
- open, shared resources: the research community shares data, algorithms, citations, and other work
- crowd intelligence: people can rate the quality of the resources, so that the community develops an interpretation of which are the best
- transparency: algorithms and results are publicly available so they can be modified and improved by other researchers
As a side benefit, results are verifiable and repeatable. Beginning researchers can build off of existing work more easily, instead of starting from scratch.
Essentially, this idea does away with the status quo of research in many fields, where each researcher works independently, rarely shares algorithms, doesn't always share data, and runs tests that are limited and not easily reproducible.
 |
| by NASA's Marshall Space Flight Center |
|
Scientific research seems like the perfect match for openness and transparency. Science is often done for purely altruistic reasons -- to simply advance the truth and knowledge. The complicating factors are that (1) corporations want to patent their research to monopolize it for themselves, and (2) academics want to keep their data and algorithms private for as long as possible, in order to publish more papers. Open source science is a big dream, but we haven't yet figured out how to balance these concerns with the benefits that an open source approach would provide.
2 comments:
I find it interesting that much of the scientific community tends to keep its research under wraps. It seems to go against the very basis and purpose of the Scientific Method, as taught in grade school, that there is a standardized procedure so that others can look at what the scientist has done and verify it through reproduction.
Perhaps if we found a way to feed information to a computer about decisions humans have made in the past regarding documents, speech recognition, images, etc. The computer could take in this information and then use it when necessary to choose the most likely human response for certain situations.
Now to get the computer to recognize human decisions and to recognize exceptions to the rules is the tricky part.
Post a Comment