Přepis řeči - PARALLEL IMPLEMENTATION OF MULTI-DIMENSIONAL ENSEMBLE EMPIRICAL MODE DECOMPOSITION

0:00:15	after after everyone
0:00:18	and
0:00:19	but they meant hobbies the pair implementation for multi
0:00:22	that mission we them in simple improve
0:00:25	cool mode decomposition
0:00:28	i first that were talking about the big one for the uh
0:00:31	much what the measuring simple in three promoting something we call we
0:00:35	emd
0:00:36	and
0:00:37	and what our what the why we wanna palace that and lay i with the
0:00:41	show you how we per rest of that and then show the result
0:00:45	in cruise
0:00:49	so
0:00:49	and we I D's that fine
0:00:51	frequency and it's for the multi
0:00:54	um at at mission data
0:00:56	uh
0:00:57	and as you can be price that time to the spatial uh to this
0:01:00	spatial domain data
0:01:02	and
0:01:04	it's hits be under uh multi data
0:01:07	different uh and that a he's P the block is uh uh empirical mode decomposition emd
0:01:13	and i in emd D
0:01:15	we
0:01:16	only can work on the one D
0:01:18	data
0:01:18	and then we use then
0:01:20	is then that use a uh is that a
0:01:24	use then uh
0:01:25	is
0:01:26	a a bad know
0:01:28	okay
0:01:28	i didn't know it so we have a in simple
0:01:31	empirical mode decomposition because record E M D but i E N D self
0:01:35	still working on the one D data
0:01:37	then
0:01:38	but
0:01:39	E E E and D provide the possibility to extend to the much but i'm use that because of making
0:01:44	a cake
0:01:45	have a a uh
0:01:46	some probabilities of copper that to uh
0:01:50	and
0:01:51	a a to to date a uh along the sent mission but uh two
0:01:55	post cruel uh neighbouring sliced data
0:02:00	so a i sure that uh
0:02:03	the part a diagram for E and D
0:02:05	so in the you know the you you are around the multi but i have a procedure to extract that
0:02:09	interesting model
0:02:11	uh frequency
0:02:12	i we live
0:02:13	and the each i have for see joe you have a several shift in procedure
0:02:17	in uh
0:02:19	in the several system procedure in that
0:02:21	it just different procedure
0:02:23	we best your try to find a a a a envelope and use use that
0:02:27	that difference of the envelope
0:02:29	and uh all the of of envelope a and all real signal
0:02:33	to to the and have a
0:02:36	and but to finesse in moderation
0:02:39	and
0:02:40	in order to to lot we have to uh we have to to find the the we have to find
0:02:45	the each trampling of the the stick no local each point of the second
0:02:49	and the use a lap uh this lost if from a points to your uh to or the envelope
0:02:55	and you in this
0:02:56	E
0:02:57	uh
0:02:58	more most come we use a spline to to build that
0:03:02	the envelope
0:03:04	and
0:03:04	we also know the the
0:03:06	this my of issues
0:03:08	uh is quite uh is
0:03:09	hi close to do so
0:03:11	so
0:03:12	and based on the fighting
0:03:14	we with discover the strip the post us if civil inverse joe uh
0:03:19	i a nineteen a percent of the computation
0:03:21	in the emd
0:03:22	and
0:03:23	uh as the or the several be a uh several or be of the block of the shift shifting
0:03:28	the
0:03:29	and the sum uh some a shell from this process
0:03:32	not not that i think that percent of the exclusion
0:03:37	so and
0:03:43	so
0:03:44	and
0:03:45	if we movie seem uh we use uh uh
0:03:48	uh if what when you uh if he and the will come on the one C can a one T
0:03:52	signal and if he's in sacks
0:03:55	i he's out voice S C one to C K
0:03:57	and
0:03:58	E and the
0:03:59	can be that show ask the the for uh this speaker so for that each D don't we add the
0:04:04	he's
0:04:05	different with to generate
0:04:07	uh a a a symbol set of a signal
0:04:09	and they for the H T going we passed through order
0:04:12	uh have the reading in the uh independent and the which are right it's
0:04:17	it's own output and the we have a to the output
0:04:20	a a and the
0:04:21	we defined the final output
0:04:23	i
0:04:24	uh as a as a uh as the ever to so
0:04:29	so a
0:04:31	so i so we can see the but you in the E do you have a a a lot of
0:04:35	a uh a a lot of U M the you we need to S Q
0:04:39	in a
0:04:40	in the and me and D
0:04:41	that's and we use a E and the S appeared building block
0:04:44	so if we uh i i i use a two D uh two D image just example
0:04:49	so we first uh a uh from the emd a the one dimension
0:04:55	so which generates a
0:04:57	some partial sell
0:04:58	and the we perform a be uh would have a E and D
0:05:01	and to they are not a technician
0:05:04	so we have a a two D uh we have a two D result
0:05:07	and we basis
0:05:08	would best to uh
0:05:10	oh oh
0:05:12	which uh use the this as a coefficient uh and uh
0:05:16	to generate the final result
0:05:19	and the put in the uh
0:05:20	the close in probably is to because so i'm only show in the in the pit
0:05:25	and in a E i E and E we uh for the one i i i M emd of the
0:05:31	well fight of by five have image
0:05:34	oh we need to exit you have to the
0:05:36	a three uh thirty meaning and
0:05:39	emd
0:05:41	and
0:05:41	and we can see that's a computation in uh come tension tense if and that network
0:05:46	now use series
0:05:47	so
0:05:47	the basic a were late and the need a is that uh
0:05:53	it's a a a real
0:05:54	so our go is try to the paradise that i was a set or a the and M D
0:05:59	the people i told before i talk about uh
0:06:02	how how we pay the i sure that the some X uh and some resolve the and even D
0:06:08	so that's a the
0:06:10	that's a a uh a you mean you of the uh
0:06:12	a two D image of the
0:06:13	however
0:06:15	and
0:06:16	i have put for from the T and the
0:06:19	the or or don't you meet you can be T cost all of the uh
0:06:23	because to the fight fight image
0:06:28	and
0:06:28	that differ uh that if and that different or a more a frequency
0:06:33	so
0:06:34	you can think of the the their first speakers is that high frequency of the
0:06:39	uh a high i have to have a high was because the of the original so should
0:06:44	and that that the last wise of the lowest frequency so that scrolls the the but high pass and low
0:06:49	pass
0:06:50	okay
0:06:53	so i'll go is high to paradise the any and the uh use a general part uh general purpose processor
0:07:01	so and
0:07:02	base a general process forces uh process support the power a ration
0:07:06	like uh uh a was P you
0:07:09	i
0:07:10	or the
0:07:11	yeah
0:07:12	uh and uh also the the many code you Q
0:07:15	a graphic are a a a a
0:07:17	graph uh graphic
0:07:18	processing you need angle go graphic are
0:07:21	so why we P could general part i job as processes
0:07:23	yes system more common
0:07:25	so if it comes to the almost every time has a
0:07:28	if even cell phone has a
0:07:29	but we are not talking on the set of
0:07:31	and
0:07:33	also he's
0:07:33	he uh he's dislike cycles ace
0:07:36	compare the has to use used this is a work
0:07:39	and you have some i for by the more uh flexibility so sometimes has you need to my the uh
0:07:44	you need to to the function of your program
0:07:46	so
0:07:48	a a so
0:07:49	so that's the a some which was so uh
0:07:51	uh general purpose
0:07:53	portions
0:07:54	and
0:07:55	so and i was lose the the that uh several the a pair of a model
0:08:00	so
0:08:01	like a a uh old maybe for if use magic was if you and the could a and be that
0:08:06	could have for the cheap used
0:08:07	i also for uh
0:08:09	that's that's some and then to open C or is target on there
0:08:13	a a a lot of about picture
0:08:15	like us if you to if U S P F P G A event i feature
0:08:19	but we listen that's of future work
0:08:21	so you nice talk we only talk about the open C O and the
0:08:24	could the
0:08:25	is uh for the see if you entropy was
0:08:29	so
0:08:30	basic a all is the appropriate model for the multi close if you
0:08:33	and he
0:08:35	he use uh us that annotation to to label the loop
0:08:39	and
0:08:40	use the have to label the loose when self
0:08:42	and
0:08:43	but well is annotation
0:08:45	the compatible to understand that which should double can be powerless
0:08:48	and
0:08:50	and for the H uh for they should power loop
0:08:52	that you to ration a large group can be to your schedule to this stem rows
0:08:57	see use the rest
0:08:59	and the rest
0:09:00	the rest part the open all that still run time manager
0:09:06	could i an a uh can model for the be to you
0:09:10	and
0:09:11	could apple right the the two layer uh a to layer at a picture
0:09:15	uh which are that at the red a plot to control to people program
0:09:20	and the right
0:09:21	is something that the scalar execution of the a uh
0:09:25	i time did mod and there it is some fact that the scalar as a fusion of the middle of
0:09:31	that
0:09:32	uh a tell uh
0:09:34	which to uh
0:09:36	uh
0:09:37	major two should
0:09:38	and that third rock face of the
0:09:40	something like a vector speech itself
0:09:44	the in in that there that each there were performed the send
0:09:47	send reach an expression your in your codes
0:09:50	and the
0:09:51	in a just the record
0:09:53	uh
0:09:54	it can't if is the map a many need there it
0:09:57	and the
0:10:00	we have a how a a ha on shape on scratch the memory to come you to
0:10:05	to help to come kate
0:10:07	a that the red
0:10:09	and in this that sent there to brock
0:10:11	and that is that there about half true
0:10:14	have to be uh independent
0:10:19	so
0:10:20	yeah i sure the the out six to to the uh
0:10:23	all opaque at imputation
0:10:25	of the uh of a i mean E D
0:10:27	so basically basic are signed the one third of uh wise if you there to handle the one em M
0:10:32	D
0:10:33	so why would do would itself read
0:10:35	a the the
0:10:37	one stiff used there to handle the one
0:10:40	emd
0:10:41	because in the D we have the we share the an input data
0:10:46	so we
0:10:46	in a so this cage doing can put up of fight that they capture in in in the case
0:10:57	so a all sure the X one
0:11:00	if we would do the office uh the outer scheduling joint late words they would not have a ship they
0:11:04	have sure so that was all on performance
0:11:08	also
0:11:09	uh
0:11:11	as as if you uh as
0:11:13	you also provide that the vector or to um bit bidding you next Q shouldn't i say C
0:11:18	so the in this case we only we we around the the i I C in table uh C compiler
0:11:24	to help of us to to find a vector um vector iteration
0:11:30	and the for high dimensional data
0:11:32	and night
0:11:33	the a two do you have a
0:11:35	the you it process S Q a E and the in the
0:11:38	one the and U S as the D and then not that mission but had
0:11:42	for this i mission high dimensional data
0:11:45	because you S is paid to use a different value of that
0:11:49	you are the
0:11:50	the members storage
0:11:51	that direction
0:11:52	so in this case is you was uh
0:11:55	uh
0:11:56	the performance of a cash will be slowed down
0:11:58	so would base it to that in split as
0:12:01	express a transport of its per the pull
0:12:05	uh
0:12:06	of the data
0:12:07	so make a like
0:12:08	um uh clay have
0:12:10	um a like a sheet
0:12:11	so that will provide the better duration of cash
0:12:16	and the in that you for uh in the
0:12:18	could i for station we have a different strategies for for different strategy
0:12:22	we basic a signed the one if you there
0:12:24	uh
0:12:25	to the white and these that emd big how that
0:12:29	it tells the the
0:12:31	the because
0:12:33	G you have a most to right
0:12:35	so he's a he have to be the more find a more fine of grand uh full matt final in
0:12:40	the program
0:12:41	so
0:12:42	this cajun will five provide the mess of data power
0:12:46	but that was a cost another not problem
0:12:49	so we basis cage order
0:12:52	sketch to what they're to the what one white
0:12:55	is self for one
0:12:56	the whole why wise
0:13:00	so
0:13:01	so
0:13:02	so we
0:13:03	so the
0:13:03	but if you
0:13:04	they tie in the uh the sent uh it in the same thread block
0:13:09	may have to uh bridge direction
0:13:12	uh because
0:13:13	uh because we had as noise
0:13:15	so
0:13:17	so would basic a or as this thing uh
0:13:20	schedule wife this send X
0:13:22	to the same that block
0:13:24	uh a to the sent their brought to avoid the a range that richard
0:13:28	that's all they all
0:13:29	come from this send X
0:13:31	so
0:13:31	they have a higher as us an energy
0:13:36	so we basically scale use something that is
0:13:38	is that often this
0:13:40	and
0:13:41	the rest how we just by up there that one cache
0:13:48	and in in the put i'm measure what didn't do that the the ah a
0:13:51	the uh had them uh exposition transport
0:13:55	transpose
0:13:56	but
0:13:57	but we that have a high dimensional man chorus in a that has is proper
0:14:02	so we
0:14:03	best again low that data how to the function of sketch pen memory
0:14:07	so we will low the something like this
0:14:09	to the question had a memory
0:14:11	but
0:14:12	in disk
0:14:13	and in the switch pay very with to this set the this to has S
0:14:18	and i i'm because
0:14:19	the pen we for scratchpad memories really hide and they since is and really low so
0:14:24	we still can benefit
0:14:28	so and an outer is some we didn't do the the uh it's is especially transpose
0:14:32	because
0:14:33	even you to the transpose you still need to store data also this
0:14:37	this this local transport
0:14:40	she's is what ten remove the the read this memory S as
0:14:45	and that's an an you should do you have a limited them a memory
0:14:49	so compare was strip you
0:14:51	uh
0:14:52	how how C if you
0:14:53	yeah
0:14:54	yeah
0:14:56	so in in your man you base you probably uh a basic can has uh i to the sits for
0:15:01	key up lights
0:15:02	memory
0:15:03	but for the one to if you only can have a a for the three yeah up by man
0:15:08	but in this
0:15:09	in this
0:15:10	and me and D
0:15:12	uh we have to sub for the up to the uh a tough the up i is for fight to
0:15:16	have a five tough
0:15:17	image
0:15:18	to get the flow pair
0:15:20	so
0:15:21	the nice
0:15:22	uh in feasible so
0:15:24	so would
0:15:24	we basically
0:15:26	to the pitch exclusion so with tie out that they uh how the data
0:15:30	to several slice and it just slides can fit in
0:15:32	that you memory
0:15:34	and you this then says size
0:15:37	like
0:15:38	lay can as the execute
0:15:40	in the power or in that difference that size two
0:15:43	in the sequential
0:15:46	a list the hours
0:15:48	uh all
0:15:49	oh is yeah uh
0:15:51	is very environment
0:15:52	uh is we run on the
0:15:55	and me idea and of
0:15:56	full uh full stock it's
0:15:58	a call
0:15:59	uh to get a entails
0:16:02	a X um X seven five fifty with this six will be out by memory
0:16:06	but only the first
0:16:08	first sixteen in are use
0:16:11	an a if you run on the ready it uh i the yeah for me test uh C twenty fifty
0:16:16	two you
0:16:17	they only provide a a a a a very gap by memory
0:16:19	and i would you could there for has us C to fifty
0:16:22	because for for me
0:16:24	from yeah at a trace their first at a first
0:16:27	right
0:16:28	really top yeah under that a a of procedure for point is
0:16:32	and i'll whole result is
0:16:35	uh is
0:16:35	is double precision
0:16:38	and we'll on that's three and three uh
0:16:42	test case
0:16:43	and this is their size
0:16:45	and that's zero exclusion we based
0:16:47	we best the only schedule the one call
0:16:50	on this motion
0:16:51	so and that that
0:16:53	then
0:16:54	then you is second
0:16:56	so and the in the could ice occlusion
0:16:59	uh in in the could i from and in implementation
0:17:03	we can see we got a have for the uh a a wrong around uh a of fifty
0:17:07	X P compare of a sequential budget
0:17:11	and in all then
0:17:12	and he implementation
0:17:14	we gather up to that you avon
0:17:16	uh speed up compare of a sequential
0:17:21	so we also yeah evaluate the the scalability of the
0:17:25	uh a T
0:17:27	by
0:17:28	by changing the how many to read that we schedule
0:17:31	so
0:17:32	we base
0:17:33	basing for as a base to provide almost noon near
0:17:36	scalable
0:17:37	no more almost a P scalability for prime T
0:17:40	or if you only schedule no then
0:17:43	a called
0:17:45	and
0:17:45	but after lot
0:17:47	he was like to right
0:17:49	so the members some we believe that's the the
0:17:52	the memory bound
0:17:54	can i cannot for uh is limited
0:17:57	and also lost
0:17:58	probably no my fact will also
0:18:00	cost not
0:18:04	we also plot provide
0:18:06	for the
0:18:07	for the uh the present you just keep a of the exclusion
0:18:11	uh uh the
0:18:13	uh a the but uh a among the block
0:18:16	and
0:18:18	oh my is basically provide an this then a simple find has a sequential
0:18:23	and the in the
0:18:24	a a good i'm from station the
0:18:27	the profile is a the to the different
0:18:30	the input in uh in separation pot is wrong slightly faster
0:18:34	but and the though each mark tape to price size
0:18:38	and the man is we pretty that in the present prize
0:18:42	uh is
0:18:43	i suitable for the message powers
0:18:45	that's
0:18:46	because they did not have a a um there's something in uh there's on the as
0:18:51	but should at takes so you have a uh a a a that the sorry
0:18:55	each are how you passed
0:18:57	because uh
0:18:59	because you cannot not fit the
0:19:01	uh a the the where
0:19:04	where they extra our point where happen
0:19:07	so last that they'd had a and then you regular T
0:19:12	to some rice
0:19:13	and this one provide tools to there right uh to their label power or or i me and the from
0:19:18	station four
0:19:20	uh
0:19:20	for cheap you C if you
0:19:22	uh
0:19:23	our program model is screwed i and with zero
0:19:26	i we got the pretty good speed
0:19:28	and just get up
0:19:32	and
0:19:32	this what
0:19:33	uh
0:19:34	the call will be released and didn't in this way say
0:19:37	and also this work this
0:19:40	and
0:19:41	in this library
0:19:43	uh power em D N E and the also all support
0:19:46	i also support
0:19:49	but
0:19:49	uh is not
0:19:51	how uh how we to light is not show in the this uh
0:19:54	uh this paper we published to
0:19:56	another that france
0:19:58	and
0:19:59	for the future work we will provide uh
0:20:02	still wrote true you can support a a a a and D if you and and the ad dot texture
0:20:07	and a
0:20:09	you also my the hard to pin of you you can only
0:20:12	use a one still scroll to long on the uh to on and you have to take
0:20:17	and
0:20:18	i put that's all
0:20:19	that's all i have
0:20:21	and question
0:20:26	a questions only
0:20:49	oh we include
0:20:51	we
0:20:53	and in the paper we showed the
0:20:55	the uh chance uh
0:20:57	that actions run times the make uh less than one person
0:21:01	right
0:21:02	on percent of the total i'm person of the took place cushion
0:21:11	or
0:21:16	like the speaker
0:21:17	and that's C and uh
0:21:19	today

PARALLEL IMPLEMENTATION OF MULTI-DIMENSIONAL ENSEMBLE EMPIRICAL MODE DECOMPOSITION

Parallel Software Implementation of DSP Algorithms