| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: Help Conquer Cancer
Thread: Accelerating the process |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 10
|
|
| Author |
|
|
BigMike
Cruncher USA Joined: Nov 14, 2005 Post Count: 13 Status: Offline Project Badges: 
|
After watching a depressing number of obviously negative samples take 6 hours of CPU time, I have a suggestion.
----------------------------------------Why not adopt the strategy used by Stardust@Home? Train people what to look for. Then have the program and scientists look at the ones that people tag as "possible". Behind every stupid computer that's participating, there's a smart person. Will people pick out the positives correctly every time? Of course not. Will they be right about which ones aren't worth looking at? Almost always. And it won't take 6 hours. With multiple people looking at each sample, the odds of accuracy greatly improve. The folks at S@H were stunned by the accuracy of the participants in finding minute particles in their samples. Some of them had even been missed by the experts. We may be trying to harness the wrong resource. --Mike
Don't believe everything you think.
|
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Yes, I've looked at SD@H and clasified galaxies and gazers and what not on a project just to see if i could see, but would I by able to know if something would dock to a particular fold, protein, receptor at what energy... many times per job?
----------------------------------------Rosetta has developed a game like project too but not heard how it fares. Meantime, 400,000 results per day with tens of millions ways to assess. That's allot of people instead of spare cycles that are used in set and forget. Anyhow in this field you'll need at least 3 persons to opine any sample, for if there is a dozer missing the rosetta stone, it will be lost for longer.
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
You posted this question previously in another thread with a response from a HCC scientist:
http://www.worldcommunitygrid.org/forums/wcg/viewthread?thread=22462 I like Mike's suggestion of interrupting the image analysis process if initial computations can already classify the image with certainty. It is a strategy we are investigating. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Hi, BigMike.
Could you identify a phase separation, skin or precipitate? The challenge here isn't really spotting the crystals - that much is fairly trivial. The problem is identifying the more subtle results. And yes, these more subtle results do look very similar to "no change". Can the automated process be improved? Almost certainly. Training human observers would be more difficult than for Stardust@Home, and as Sekerob says, the cost of an error would be higher. The whole goal of this project is a) to remove the subjectivity of the observer, and b) allow processing at the rate that samples are produced. |
||
|
|
BigMike
Cruncher USA Joined: Nov 14, 2005 Post Count: 13 Status: Offline Project Badges:
|
You posted this question previously in another thread with a response from a HCC scientist No, I didn't. The question I asked before had to do with the program recognizing the shape of the light source. You're trying pretty hard to be a spoiler. ==Mike
Don't believe everything you think.
|
||
|
|
BigMike
Cruncher USA Joined: Nov 14, 2005 Post Count: 13 Status: Offline Project Badges:
|
Could you identify a phase separation, skin or precipitate? Didn't say I could. But I *can* spot a completely clear cell a lot faster than the program seems to be able to. If I sat down and identified 5 clear cells in a few minutes, we're way ahead of a program that can only do 4 in a day. If the question is whether there's *anything* there or not, it should be pretty simple. And if you require a quorum of 2 or 3 to agree on a cell being *completely* clear, the error probablility should still be extremely low. Maybe even lower than program logic can produce. ==Mike
Don't believe everything you think.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
That's just the problem. You think you know what a completely clear cell looks like. Do you? I don't know. But I understand that some of the indeterminate results may look very like a clear cell.
----------------------------------------edit: I had a look at some guide images, and I stand by my assertion that some situations are almost impossible to distinguish. Experts can do it, but not me or you. Phase separation can easily be confused with clear, and crystals with garbage. The combination states are even harder to classify. [Edit 1 times, last edit by Former Member at Mar 12, 2009 5:59:24 PM] |
||
|
|
BigMike
Cruncher USA Joined: Nov 14, 2005 Post Count: 13 Status: Offline Project Badges:
|
But do we care about the ones that are hard to distinguish from a clear cell? Do we care if it's phase separation, or really clear? Not if what the project cares about are crystals.
----------------------------------------If the question is simply "is there anything there", then any kind of recognizable result, whatever it is, would be answered "yes". Then someone, or the program, needs to look at it. SD@H has shown people are extraordinarily good at answering that particular question. If we answer "no, there's nothing there" and it's really phase separation, what have we lost? There's still no crystal. ==Mike
Don't believe everything you think.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
BigMike, read some of the updates from the researchers - they care very much about partial results.
----------------------------------------Partial results, combined, may lead to finding successful conditions for crystallisation. It's kind of the whole point of the experiment. If all they wanted to do was identify definite crystals, they could probably get a load of students to go through them all, scanning through them dozens at a time. edit: Maybe you think I'm being too harsh on the idea. I don't mean to be - if things were just a little different, human identification would be the ideal solution. It would work for many similar situations, too. [Edit 1 times, last edit by Former Member at Mar 13, 2009 2:34:46 AM] |
||
|
|
BigMike
Cruncher USA Joined: Nov 14, 2005 Post Count: 13 Status: Offline Project Badges:
|
If all they wanted to do was identify definite crystals, they could probably get a load of students to go through them all, scanning through them dozens at a time. Point taken. It's just that, as a computer scientist for the last 30 years, the idea of wasted CPU time makes me go into problem determination mode. You can never get back a wasted CPU cycle. ==Mike
Don't believe everything you think.
|
||
|
|
|