Closed. This question is off-topic. It is not currently accepting answers.
Want to improve this question? Update the question so it's on-topic for Stack Overflow.
Closed 11 years ago.
Improve this question
In developing a face recognition we first need to detect faces.Recent way is to train a system on known databases i.e artificial intelligence and neural networks.I would like to know how this training is done?
When training your system, you will need to train a classifier to distinguish between faces and non-faces relying on a set of features.
These features can be defined differently, but for a simple approach, the whole array of pixels can be used as features. One pixel = one feature.
Then you would need to apply a training algorithm. Common ones are logistic regression and support vector machines.
The training consists in correctly weighing the individual features to obtain the correct classification.
This can be done by minimizing a cost function.
You will need to separate your examples into at least two sets - the training set and the test set, to validate the results you get from training on a test set.
Related
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 8 months ago.
Improve this question
Many chat channels (e.g. Omegle, mIRC) produce the the following conversations when engaging a person one doesn't know.
M
asl?
F 25
wonna sexchat?
This behavior is extremely prevalent. Detection of nudity in videos or images is an open research problem. What I am listing here should be simpler to accomplish. Any off-the-shelf solutions? Any language and any open and free library is fine(including papers on how to do it)!
There is a library on Python that name is profanity. Link: profanity-check
With machine learning algorithm I'm super sure that we can handle it easily. For that you should collect whole words as that. After collecting, you can add a label line, like 0 if acceptable, 1 if not. Then, with some successful machine learning algorithms, you can associate the data with the euclidian, manhattan type distance criteria and create a decision mechanism about whether the content is bad or not. It is difficult to predict how much the success rate will be, but I think that with such an approach, a success rate close to 90% can be achieved. I would like to share with you an academician whose work I trust on this subject. Yılmaz Kaya
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 7 years ago.
Improve this question
I have the task to write some crypto stuff in C and make it lightweight. The idea behind making it lightweight is, that it could run on a smartcard which doesn't offer much computational power and memory. It won't come to actually running it on a smartcard and it won't be for any practical use.
However, I'm curious if I could run the program on a smartcard without major adjustments. I'm aware that I'd probably have to change something in the IO-part but let's keep that aside. And by "smartcard" I mean a regular smartcard which could be afforded by the majority of private individuals and not some fancy stuff.
To restrict the question a little more:
Could I run the program without modification if I only use 8-bit integers in my program and the architecture is >= 8-bit, aswell as stay below the memory limit?
If no, why not?
Due to their limited CPU power, SCs mostly have their own security/encryption hardware and OS. The latter for instance controls access to critical features like the interface and key storage. Also, some of them have countermeasures against typical attack scenarios like differential cryptoanalysis, etc.
There are standards available, but which to pick depends on the actual card used. There are various SCs on the market with different capabilities and demands.
It is unlikely that your program will run without major modifications.
Note that the specs are mostly only available under NDA and possibly with additional guarantees from your side. The actual level depends on the capabilities and the card vendor.
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 9 years ago.
Improve this question
I'm curious if there are any advantages to writing your own sorting algorithm when programming something that needs sorting versus simply using a database to sort the data? The only thing i can think of would be if the data set is small enough where it wouldn't be worth it to set up a database with some queries + the extra time that it would take to feed the data back and forth to the program. Databases do a pretty good job at sorting efficiently so besides what i mentioned above, i have no clue.
One common reason to employ a sorting algorithm outside of the database is to provide sorting facilities controlled by end-users of your system in the presentation layer. In situations like that it would be too costly to go back to the database and fetch the information that you already have simply for the sake of having it sorted differently.
However, it is highly unusual to implement sorting yourself: all popular and most not-so-popular programming environments supply built-in sorting facilities, recognizing their widespread use.
you should implement sorting outside the database when your key field is complex and you need complex comparators to sort them.
Closed. This question needs to be more focused. It is not currently accepting answers.
Want to improve this question? Update the question so it focuses on one problem only by editing this post.
Closed 9 years ago.
Improve this question
Anybody used Neural Network approaches for clustering data? Particularly
ART Neural Network (Adaptive Resonance Theory) or
Kohonen self organizing maps
How are they as compared to k-means or any other distance based clustering Algorithms?
Self-organizing maps (SOMs) have some internal similarities with K-Means, but also important differences. A SOM actually maps your data from the original data space (usually high-dimensional) onto the map space (usually two-dimensional), while trying to preserve the original data densities and neighborhood relationships. It won't give you directly the clustering, but may help you to visually inspect the data and recognize clusters.
I know too little about ART nets.
Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.
We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.
Closed 8 years ago.
Improve this question
I'm working on a project that requires finding the most intersected set among a great number of other sets.
That is, I have a large number (~300k) of sets with hundreds of entries each. Given one of the sets, I need to rank the other sets in order of how intersected they are. Additionally, the set entries contain properties which can be used as a filter, e.g. For set X, order the other sets by how much they intersect with the "green" entries subset.
I have free reign to architect this solution, and I'm looking for technology recommendations. I was initially thinking a relational DB would be the best suited, but I'm not sure how well it will perform doing these real time comparisons. Somebody recommended Lucene, but I'm not sure how well that would fit the bill.
I suppose it's worth mentioning that new sets will be added regularly and that the sets may grow, but never shrink.
I don't know exactly what you are looking for: method, library, tool?
If you want to compute your large datasets really fast with distributed computing, you should check out MapReduce, e.g. using Hadoop on Amazon EC2/S3 services.
Lucene can easily scale to what you need. Solr will probably be easier to set up, and hadoop is most likely overkill for only a few million data points.
Something you need to think about is what definition of "how intersected" you want to use. If all the sets have the same size I suppose it's easy, but Jaccard distance might make more sense in other contexts; Lucene's default scoring is often good too.
My advice would be: try running the default Solr instance on your local workstation (it's a cllick-and-run jar type of deal). You'll know pretty quickly whether Solr/Lucene will work for you or if you'll have to custom code your own thing via Hadoop etc.