| WIKI | FAQ | Tech FAQ | |
 http://csua.com/feed/ | |
| ||||||
| 4/3 |
| 2002/1/8-9 [Academia/GradSchool, Computer/Theory] UID:23497 Activity:very high |
1/8 I am interested in handwriting recognition. In particular,
Chinese handwriting recognition. What are the relevant CS/EE fields?
What is a good grad school to go to for such things?
- person who finally found out what he is interested in, but
don't have much idea on how to pursue his interest.
\_ AI.
Berkeley, MIT, Stanford, CMU, UCLA, UIUC, UW, etc.
\_ Machine Learning and AI. -- CMU, Cornell, Berkeley, MIT, ...
Here's a link to code which recognizes faces (which is similar
to recognizing handwriting):
http://www-2.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/faces.html
\_ Thanks, what about some tier two schools? Are they worth
going? Or should I just find a job at the companies that
are implementing these things.
\_ Gradschool or not depends on where your interest lies.
If you want to work on developing algorithms on handwriting
recognition, then you should go to grad school (in machine
learning). If you're more interested in the engineering
and implementation aspect of the product, then work for
a company. Somewhere in between --> work as a developer
at a research lab. If you decide to go to grad school, then
first find a research AREA you're interested in. Going in
with a specific application in mind is not really good,
because 1) your interests will change, and 2) you may not
find a professor who's interested in EXACTLY the same
application. I would suggest you start by reading some
papers on handwriting recognition, and look at the homepages
of people who do ML, and pick your favorite people. Then
write those people a nice email and demonstrate you know
what you're talking about. If you still have a couple of
years to kill at Berkeley, then start by doing some research
in the area. I don't know why I'm even writing all this
down. You're probably not even going to read this far.
-- alice
\_ Thanks! That helps! You are a very nice person.
I think I am interested in the engineering and
implementation aspect. Does that rule out grad
school or can it still be useful?
\_ In that case, check out http://www.synaptics.com Apparently
they make (made) chinese handwriting recognition
products, according to John Platt's homepage.
http://research.microsoft.com/~jplatt
John, perhaps known in this crowd as the inventor of
ClearType, is smart and a nice guy. He may even answer
general inquiries from random people about about
Synaptics and handwriting recognition. And please,
don't call me nice. It gives people the wrong
impression. -- alice
implementation aspect. Does that rule out grad school
or can it still be useful?
\_ If you have further questions, be VERY nice and send
her an email. She might respond. Sending personal email
reduces the chance that you're just trolling or
interested in the hypothetical
\_ why not a Master's where you concentrate on learning
AI, and machine learning ideas, and maybe work on
a project on machine learning, then go to industry
afterwards? Another option is to take classes at
Stanford's SITN (if they have those courses there)
while working.
\_ Study Dvorak instead.
\_ Japanese Handwriting recognition demo:
http://ai.bpa.arizona.edu/go/mlir/japanesehandwriting.html |
| 4/3 |
|
| www-2.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/faces.html A neural network learning algorithm called Backpropagation is among the most effective approaches to machine learning when the data includes complex sensory input such as images. Visitors from outside CMU are invited to use this material free of charge for any educational purpose, provided attribution is given in any lectures or publications that make use of this material. |
| research.microsoft.com/~jplatt -> research.microsoft.com/%7Ejplatt/ I have also helped to create improved machine learning algorithms, such as * 10 Fast training algorithms, 11 probabilistic outputs, 12 multi-class training, and 13 on-line training for 14 Support Vector Machines. Recent publications Sorted by topic, then by date High-speed machine learning algorithms 18 Learning to Learn with the Informative Vector Machine 19 abstract By N. Platt, International Conference on Machine Learning, to appear, (2004). Platt, Advances in Neural Information Processing Systems 16, (2004). Burges, Microsoft Research Technical Report MSR-TR-2003-38, (2003). Platt, Advances in Kernel Methods - Support Vector Learning, B. Learning to learn (learning priors of Gaussian processes) 30 Learning to Learn with the Informative Vector Machine 31 abstract By N. Platt, International Conference on Machine Learning, to appear, (2004). Zheng, Advances in Neural Information Processing Systems 14, pp. Zheng, Advances in Neural Information Processing Systems 14, pp. Deng, Advances in Neural Information Processing Systems 13, pp. Platt, Microsoft Research Technical Report MSR-TR-2003-82, (2003). Field, Fourth IEEE Pacific Rim Conference on Multimedia (2003), to appear. Longer version as 55 Microsoft Research Technical Report MSR-TR-2002-17, (2002). Schlkopf, Microsoft Research Technical Report MSR-TR-2001-99, (2001). IEEE Workshop on Content-Based Access of Image and Video Libraries 2000, pp. Williamson, 68 Seventh Pacific-Asia Conference on Knowledge Discovery and Data Mining, pp. A longer version appeared as 71 Microsoft Research Technical Report MSR-TR-99-87, (1999). A conference version was 72 Support Vector Method for Novelty Detection by B. Shawe-Taylor, Advances in Neural Information Processing Systems 12, pp. Platt, Advances in Kernel Methods - Support Vector Learning, B. Sahami, 7th International Conference on Information and Knowledge Management, pp. There are 84 abstracts and downloadable Postscript of some older papers. These papers are about neural networks, visual object recognition, and computer graphics. Biography Before coming to Microsoft, I was Director of 85 Research at a small company called 86 Synaptics. They make touchpads and Chinese handwriting recognition products. At Synaptics, I worked on 87 neural network architectures & 88 hardware, 89 neural networks to recognize objects, and 90 handwriting recognition. Other Information I have worked with a number of 94 other people inside and outside Microsoft Research 95 Asteroids I've discovered My Erds number is 3, through 96 John-Shawe Taylor. |
| ai.bpa.arizona.edu/go/mlir/japanesehandwriting.html Users can enter Japanese queries by writing the characters in the HWR Input Test window which will place the recognized characters in the text field of the search form. The Hiragana and Katakana recognition was developed at the UAMIS AI Lab by Marshall Ramsey and the Kanji recognition was developed by Todd David Ruddick as part of his JavaDict project that won first place at the ACM's Quest for Java '97 Contest. The HWR system is not endorsed or developed by any of the search engines. The five best matching characters are displayed in a pull-down list at the bottom of the window. Choose the character you wrote (the top choice is selected by default) and it will be placed in the form after you write another character or press the Force button. If the character you wrote is not listed, select none from the list and write the character again. Press the Clear button if you make a mistake while writing a character. After entering all the characters for your query, press the button to the right of the text field in the search form. A new window will be displayed that shows the search results from the search engine. A Japanese font is also necessary to view Japanese in both Netscape and Java. After installing the font, follow these steps to enable Netscape to view Japanese Web pages: 1. Select Appearance-Fonts from the Category area in the dialog box. Select the font you just installed for both variable and fixed widths. This program has been tested on Windows 95/NT and SGI IRIX. |
| www.synaptics.com The resulting SpeakerPad module offers quality sound, a reliable interface solution, and simplifies product development and manufacturing for OEMs. |