I think what you're looking for - with one exception - is Hiveminder.
I've been using it for a little over a year - the free version is good and the Pro version is (if I remember correctly) $30 / year. It supports:
Task dependencies - so if you can't do task B until A is done, it will (by default) hide task B.
Task priorities
Due dates
You can add comments to each task, describing where you are at as you go
The pro version supports up to 500 Mb of attachments
Other things I really like:
It is also collaborative, so others can assign you tasks (even over e-mail in the pro version)
It supports a "hide until" date - so you can ignore some tasks for a few days while you concentrate on others
Recurring tasks
The downside is that it is an online service, not a desktop app or locally hostable web application. Whether that is a deal-killer will depend on your needs and level of paranoia, but the pro version does allow everything to be done over ssl.
(Just a happy customer, btw - not affiliated in any way with the company)
We won't know how it affects humans until we send a girl up there to get knocked up and see what happens.
Absolutely . . . but then we won't sure until we get a statistically significant sample of those. I'm trying hard to imagine how to put that in a grant.
This is called adaptive cruise control, and it is available on many high-end cars. Note the use of lidar or radar instead of sonar, however.
I once worked on a project to use sonar to guide a robot through hallways. We had very good sensors, but they still gave very inaccurate readings. IMHO sonar is definitely not something I would consider trusting in a transportation context without some serious R&D to improve it. My professor for the project had actually worked with Nissan on an adaptive cruise control system.
Remember that there is a difference between using mitochondrial DNA (the studies you cited) and autosomal DNA (this study). With mitochondrial DNA, the only information that you get is along the maternal line, so you're missing a lot of the data. Looking back 20 generations, for example, you're only looking at one ancestor out of about a million. It would be possible for two groups to come over but only one be reflected in the maternal line.
I am a graduate student in computer science, emphasizing the use of machine learning.
The sound bite conclusion of this blog post is that algorithms are a waste of time and that you are better off adding more training data.
The reality is that a lot of really smart people have been trying to come up with better algorithms for classification, clustering, and (yes) ranking for a very long time. Unless you are already familiar with the field, you really are unlikely to invent something new that will work better than what is already out there.
But that does not mean that the algorithm does not matter - for the problems I work on, using logistic regression or support vector machines outperforms naive bayes by 10% - 30%, which is huge. So if you want good performance, you try a few different algorithms to see what works.
Adding more training data does not always help either, if the distributions of the data are significantly different. You are much better off using the data to design better features which represent/summarize the data.
In other words, the algorithm is not unimportant, it just isn't the place your creative work is going to have the highest ROI.
Who would you really rather have judging the novelty of software patent applications: someone with a Bachelor's degree and no work experience, or someone with 40+ years of patent experience? You don't get that kind of experience by exercising bad judgment and rubber-stamping everything that comes across your desk. I'll take the experienced examiner any day!
Isn't that what Arxiv.org has been doing for ages already? But this service is specifically intended for fields which are not covered by Arxiv.org. Quoting http://precedings.nature.com/about
We do not accept submissions from fields in the physical sciences that are are already well served by preprint servers such as arXiv.org.
This is not simply a proposal, though you have to go to the actual journal article to determine that. The press release is so hyped up though that it's hard to see that basically all they're doing is applying two well-known bioinformatics techniques to the problem of finding previously unknown/unstudied genes related to learning and memory.
The first technique is simply to see what interacts with known genes (CREB and zif268); since proteins function by interacting with each other, you'd expect that most - not all - of the proteins that CREB and zif268 interact with will be related to memory and learning. See http://en.wikipedia.org/wiki/Protein_interaction
The second technique is just an application of the fact that similar proteins from different organisms (i.e. homologs) usually have the same function. See http://en.wikipedia.org/wiki/Homology_modeling
These computational techniques can be very useful in hypothesizing which genes may be involved, so that you can then go to the lab and either confirm or reject your hypothesis. The authors did not do so, but they did do a search of the experimental literature, which gives a partial confirmation. But the fact remains that this work is simply the application of known computational techniques. In all, I'd say it's a nice bit of work and worth my time (as a PhD student with an emphasis in bioinformatics)... but not worth a press release or getting excited about.
The article essentially says as much:
Initial research suggests the software-based system can make it 40 times more likely for caseworkers to accurately predict future lethality than they can using current practices. [emphasis added] Caseworkers have to decide which of the convicted felons assigned to them are the ones that need the most attention. I would argue that - whether you're worried more about reoffense (false negatives) or discriminating against innocents (false positives) - it is clearly better to have repeatable, verifiable procedures in place to decide who gets the attention and who doesn't than to always leave it to a hunch. If the system is verifably and significantly more accurate than current practice, all the better.
Slashdot Mirror
I think what you're looking for - with one exception - is Hiveminder.
I've been using it for a little over a year - the free version is good and the Pro version is (if I remember correctly) $30 / year. It supports:
Task dependencies - so if you can't do task B until A is done, it will (by default) hide task B.
Task priorities
Due dates
You can add comments to each task, describing where you are at as you go
The pro version supports up to 500 Mb of attachments
Other things I really like:
It is also collaborative, so others can assign you tasks (even over e-mail in the pro version)
It supports a "hide until" date - so you can ignore some tasks for a few days while you concentrate on others
Recurring tasks
The downside is that it is an online service, not a desktop app or locally hostable web application. Whether that is a deal-killer will depend on your needs and level of paranoia, but the pro version does allow everything to be done over ssl.
(Just a happy customer, btw - not affiliated in any way with the company)
We won't know how it affects humans until we send a girl up there to get knocked up and see what happens.
Absolutely . . . but then we won't sure until we get a statistically significant sample of those. I'm trying hard to imagine how to put that in a grant.
But do real men use sat-nav?
In the same way that "real men" only program in assembly!
You can read the abstract of the article in the Annals of Neurology at http://www3.interscience.wiley.com/journal/122266379/abstract
This is called adaptive cruise control, and it is available on many high-end cars. Note the use of lidar or radar instead of sonar, however.
I once worked on a project to use sonar to guide a robot through hallways. We had very good sensors, but they still gave very inaccurate readings. IMHO sonar is definitely not something I would consider trusting in a transportation context without some serious R&D to improve it. My professor for the project had actually worked with Nissan on an adaptive cruise control system.
Remember that there is a difference between using mitochondrial DNA (the studies you cited) and autosomal DNA (this study). With mitochondrial DNA, the only information that you get is along the maternal line, so you're missing a lot of the data. Looking back 20 generations, for example, you're only looking at one ancestor out of about a million. It would be possible for two groups to come over but only one be reflected in the maternal line.
I am a graduate student in computer science, emphasizing the use of machine learning.
The sound bite conclusion of this blog post is that algorithms are a waste of time and that you are better off adding more training data.
The reality is that a lot of really smart people have been trying to come up with better algorithms for classification, clustering, and (yes) ranking for a very long time. Unless you are already familiar with the field, you really are unlikely to invent something new that will work better than what is already out there.
But that does not mean that the algorithm does not matter - for the problems I work on, using logistic regression or support vector machines outperforms naive bayes by 10% - 30%, which is huge. So if you want good performance, you try a few different algorithms to see what works.
Adding more training data does not always help either, if the distributions of the data are significantly different. You are much better off using the data to design better features which represent/summarize the data.
In other words, the algorithm is not unimportant, it just isn't the place your creative work is going to have the highest ROI.
Who would you really rather have judging the novelty of software patent applications: someone with a Bachelor's degree and no work experience, or someone with 40+ years of patent experience? You don't get that kind of experience by exercising bad judgment and rubber-stamping everything that comes across your desk. I'll take the experienced examiner any day!
Original Carnegie Mellon press release: http://www.cmu.edu/news/archive/2007/April/april17 _genes.shtml
... but not worth a press release or getting excited about.
The actual journal article: http://www.biomedcentral.com/1471-2202/8/20
This is not simply a proposal, though you have to go to the actual journal article to determine that. The press release is so hyped up though that it's hard to see that basically all they're doing is applying two well-known bioinformatics techniques to the problem of finding previously unknown/unstudied genes related to learning and memory.
The first technique is simply to see what interacts with known genes (CREB and zif268); since proteins function by interacting with each other, you'd expect that most - not all - of the proteins that CREB and zif268 interact with will be related to memory and learning. See http://en.wikipedia.org/wiki/Protein_interaction
The second technique is just an application of the fact that similar proteins from different organisms (i.e. homologs) usually have the same function. See http://en.wikipedia.org/wiki/Homology_modeling
These computational techniques can be very useful in hypothesizing which genes may be involved, so that you can then go to the lab and either confirm or reject your hypothesis. The authors did not do so, but they did do a search of the experimental literature, which gives a partial confirmation. But the fact remains that this work is simply the application of known computational techniques. In all, I'd say it's a nice bit of work and worth my time (as a PhD student with an emphasis in bioinformatics)