[ Skip to the content ]

Institute of Formal and Applied Linguistics Wiki

• ÚFAL Wiki Home
• ÚFAL Home
• Sitemap
• RSS
• ÚFAL calendar
• Recent changes
• Login
• Hledat:

[ Back to the navigation ]

Differences

This shows you the differences between two versions of the page.

View differences: Side by SideInline Go

Link to this comparison view

Both sides previous revision Previous revision 2012/03/12 22:59 longdt 2012/03/12 22:53 longdt 2012/03/12 22:53 longdt 2012/03/12 22:51 longdt 2012/03/12 22:50 longdt 2012/03/12 22:50 longdt 2012/03/12 22:49 longdt 2012/03/12 22:47 longdt 2012/03/12 22:45 longdt 2012/03/12 22:44 longdt 2012/03/12 22:42 longdt 2012/03/12 22:41 longdt 2012/03/12 22:41 longdt 2012/03/12 22:39 longdt 2012/03/12 20:22 longdt 2012/03/12 20:21 longdt 2012/03/12 20:19 longdt 2012/03/12 20:17 longdt 2012/03/12 20:16 longdt 2012/03/12 20:11 longdt vytvořeno Go Next revision		Previous revision 2012/03/12 22:59 longdt 2012/03/12 22:53 longdt 2012/03/12 22:53 longdt 2012/03/12 22:51 longdt 2012/03/12 22:50 longdt 2012/03/12 22:50 longdt 2012/03/12 22:49 longdt 2012/03/12 22:47 longdt 2012/03/12 22:45 longdt 2012/03/12 22:44 longdt 2012/03/12 22:42 longdt 2012/03/12 22:41 longdt 2012/03/12 22:41 longdt 2012/03/12 22:39 longdt 2012/03/12 20:22 longdt 2012/03/12 20:21 longdt 2012/03/12 20:19 longdt 2012/03/12 20:17 longdt 2012/03/12 20:16 longdt 2012/03/12 20:11 longdt vytvořeno Go
courses:rg:2012:longdtreport [2012/03/12 22:47] longdt		courses:rg:2012:longdtreport [2012/03/12 22:59] (current) longdt
Line 24:		Line 24:
	__Sorted Array__		__Sorted Array__
-	+ Use n array for n-gram model (array i-th is used for i-gram)	+
-	+ Each element in array in pair (w,c)	+	- Use n array for n-gram model (array i-th is used for i-gram)
-	+ w : index of that word in unigram array	+	- Each element in array in pair (w,c)
-	+ c : offset pointer	+	w : index of that word in unigram array
-	+ Sort base on w	+	c : offset pointer
		+	- Sort base on w
	Improvement : Implicitly encode W (all n-gram ending with particular word wi are stored -> wasteful. So, maintain another array save the beginning and the end of the range		Improvement : Implicitly encode W (all n-gram ending with particular word wi are stored -> wasteful. So, maintain another array save the beginning and the end of the range
Line 38:		Line 40:
	__Variable length coding__		__Variable length coding__
-
-	Most of the attendants apparently understood the talk and the paper well, and a
-	lively discussion followed. One of our first topics of debate was the notion of
-	skyline presented in the paper. The skyline was somewhat of a supervised element
-	-- the authors estimated initial parameters for a model from gold data and
-	trained it afterwards. They assumed that a model with parameters estimated from
-	gold data cannot be beaten by an unsupervisedly trained model. Verily, after
-	training the skyline model, its accuracy dropped very significantly. The reasons
-	of this were a point of surprise for us as well as for the paper's authors.
-	Complementary to the skyline, the authors presented a baseline which should	+	Idea :
-	definitely be beaten by their final model. This baseline, they called	+	Context offset tend to be close with each others, => Only save the first offsets address and the difference of others with the it
-	"uninformed", but were vague about which exact probability distribution they	+
-	used in this model. We could only speculate it was a uniform or random	+
-	probability distribution.	+
-	A point about unsupervised language modeling came out: Many linguistic phenomena	+	Question :
-	are annotated in a way that is to some extent arbitrary, and reflects more the	+	how the unigram be sorted ?
-	linguistic theory used than the language itself, and an unsupervised model	+	Martin suggest that it must be sorted base on frequency
-	cannot hope to get them right. The example we discussed was whether the word	+
-	"should" is governing the verb it's bound with, or vice versa. The authors	+
-	noticed that dependency orientation in general was not a particularly strong	+
-	point of their parser, and so they also included an evaluation metric that	+
-	ignored the dependency orientations.	+
-	Perhaps the most crucial observations the authors made was that there is a limit	+	__Block Compression__
-	where feeding more data to the model training hurts its accuracy. They	+	compress the key/value array in the blocks of 128 bytes,
-	progressed from short sentences to longer, and identified the threshold, where	+	the underline reason is only to support binary search
-	it's best to start ignoring any more training data, at sentences of length 15.	+
-	However, we were not 100% clear how they computed this constant.	+
-	If the model was to be fully unsupervised, it remains a question, how to setup	+
-	this threshold, because it cannot be safely assumed that it would be the same	+
-	for all languages and setups.	+
-	The writing style of the paper was also a matter of differing opinions.	+	==== Decoding ====
-	Undeniably, it is written in a vocabulary-intensive fashion, bringing readers	+	**I. Exploiting Repetitive Queries**
-	face to face with words like "unbridled" or "jettison", which I personally had	+
-	never seen before.	+
-	==== Conclusion ====	+	The method use cache to speed up the process
		+	This simple implementation increase performance of 300% over conventional implementation
		+	**II. Exploiting Scrolling Queries**
-	All in all, it was a paper worth reading, well presented, and thoroughly	+	We can quickly form the context encoding of the next query by concatenating new words with saved offset from previous query
-	discussed, bringing useful general ideas as well as interesting details.	+
		+	==== Conclusion ====
		+	In summary, it was a worth reading paper, well presented, and thoroughly
		+	discussed. Paper contain many detail techniques that definitely helpful for actual implementation.

---

[ Back to the navigation ] [ Back to the content ]