Anand Rajaraman, an associate professor who teaches data mining at Stanford had a nice post on the power of more data today over algorithms.
One of the stumbling blocks, though, is acquiring that data in an intelligent fashion. For instance, in the TRECVID evaluation - where teams develop systems to identify the presence of concepts, such as U.S. flags, in a large of collection videos, there are often very few positive examples to train on. On the other hand, the evaluation in general had discouraged enhancing the training set by designating those teams' systems that do so 'C' type - a rarely used designation thus limiting what can be learned by comparison with other systems.
That is, how do you properly evaluate systems when you open the dataset and simultaneously allow varying algorithms?
Monday, March 31, 2008
Tuesday, March 25, 2008
Science!
I recently was invited to help judge a middle school science fair. There was a wide range of experiments from hilarious: "Do people lie more often to kids or adults?" based upon responses after subjects ate salt-laden cookies, to engineering-like: "Can water be made into hydrogen and oxygen?"
Perhaps the best one I saw measured how well different thicknesses of wire would transduce vibration to electricity.
It's great to see young minds taking part in what gets demonized in American culture as a 'for geeks' process - the scientific method. Some projects were better than others, as some kids have more of an inclination or knack to tackle science, but doubtlessly the exposure will lead to less fear of 'what's hard' as they grow into adults.
One observation though: the format of the science fair submission were all uniform. Cardboard stands, demos, and a report - no videos, no websites, no audio explanations, no flat layouts, etc... A certain creativity in display design was missing. It's not surprising to see such 'to-the-letter' assignment following and typical pedantry in an environment where the state controls the educational content.
Even undergraduates in college suffer from this mentality - I did task A, to get points B, to get grade C - rather than look beyond the process in order to grasp true understanding of the assignment's lesson.
---
UPDATE: Make your own aerogel - a ten year old's awesome science project.
Perhaps the best one I saw measured how well different thicknesses of wire would transduce vibration to electricity.
It's great to see young minds taking part in what gets demonized in American culture as a 'for geeks' process - the scientific method. Some projects were better than others, as some kids have more of an inclination or knack to tackle science, but doubtlessly the exposure will lead to less fear of 'what's hard' as they grow into adults.
One observation though: the format of the science fair submission were all uniform. Cardboard stands, demos, and a report - no videos, no websites, no audio explanations, no flat layouts, etc... A certain creativity in display design was missing. It's not surprising to see such 'to-the-letter' assignment following and typical pedantry in an environment where the state controls the educational content.
Even undergraduates in college suffer from this mentality - I did task A, to get points B, to get grade C - rather than look beyond the process in order to grasp true understanding of the assignment's lesson.
---
UPDATE: Make your own aerogel - a ten year old's awesome science project.
Thursday, March 20, 2008
Dawes fakes Brodeur out of his jock strap
Forward to 2:52 to see Nigel Dawes hypnotize the living legend:
Thursday, March 13, 2008
Subscribe to:
Posts (Atom)
