graph to everyone uh i only a low of i them you have i i them yeah in kind issue sure that when you an inverse so that title of my presentation is cost says that is taking for audio tag annotation and retrieval so he uh here is a of example one famous uh music taking web that that every N uh we show the uh sound track that T and it's a so your text and this so take provide reach information four oh a and and then is this so for example we can trace and class wise using uh using uh the audio uh and that the audio so your at we use us that then take so in this paper we focus on two important information the first one at a um a which means the number of different users who have annotated this tech uh in this i guess each simple a larger font uh indicates a a higher take on and the second uh information used tech corporation uh as we know that uh sound takes a a uh often cork curve so we propose a a cost is at this they keep for X probably in take a long and take a should joint to the okay we first introduce uh to use the information retrieval task the first is a audio annotation that is given a an audio creep oh we will uh we can make pretty shouldn't use sound take five so we would know that are these audio you be but so at with which takes and the they why is uh take based audio retrieval uh given uh take where carrie we can opt to prediction score using in the take cross fine and they we will have a range in these for the query okay so the first interest is since is the take on uh uh i we know that uh uh show take a a a sign by people with a label musical and knowledge so they inevitable ready concave noisy information oh we think that uh take on information should be can see the rate in all to make all automatic may take it because luck on for X the constants degree of the tech and the higher take on a more reliable and at in yeah were here we can't that uh experiment woman we have to select a at high to take and sound a low can take we uh come and they are uh the prediction performance according to false negative rate as we can see that the force snake you rate on the high can take uh match more more than in the goal con okay so we believe that uh it is uh ever that that uh high context um more reliable way stay at a okay we try to calm be is you by D this it's temple yeah we show the uh not train likely be and it's so text and this a the height how high high cut takes uh include tea six these peters british drastic rock all D so we put believe that these are more reliable the D and and important take so uh is some uh previous work uh that that take on is transformed into one or zero by using a this a whole and then a binary classifier each trend for each take to make prediction but a this may uh at it uh this may have a some problem a first slice that take on information is lost but take a side twice is traded it in the say same way as a take a hunter of time is the second probably used let a it is hard to determine the source all and that's supper so that's there probably is that uh they will be ambiguity in and lower class membership all that is that is nearby by less is all for example you we set take on the so to ten so that is that is moos take on is ten it will be kind see there eyes up as the use simple but but it uh if it's take out is nine they you would be can see there S a negative example i with the are that you is it it is there is strange so i'll question use how to use the take on information for audio tag annotation and retrieval and all i and there is cost is it the name with the take on as cost so in close this at the learning we are given a training in set X Y in C the X is the feature vector and why is the class label and C is the crap and misclassification cost of these it's simple or look for all uh close says that the than the in is to then a class fine which minimize the expected cost and on and thing is and it is a more general state apple all traditional classification problem so in all uh application i'll court is to minimize mays classified take on for audio tag annotation and retrieval so if one hundred you use annotate a an audio clip which is rock but that five years oh force the egg negative then the cost is one hundred so the cost it's of the than in were where we pay more attention on the reliable or at and and important take and so we have it uh we have a it's probably at to close sensitive by binary classifiers the first why is close since that these support vector machine uh is a public the machine the training error wrote ten "'cause" see uh uh uh will uh will be a some shady with a cost to i and the second "'cause" since they class twice uh a close to the end up pose so here we show the update they uh way is that weight update do E do in add up pose and uh though uh weight updating eighteen all and is that is will be proportion to the cost of these is that okay uh the second uh you put "'em" information in is uh take variation i and a is on previous work the take on notation as is separated it into several a binary classification problem so uh les assume that that takes a are independent so the take colouration information use lost for example we know that he have and wrap open call curve for for example we yeah in our database uh we can't all the that they call curve vol one hundred and sixty times and they are only uh seventy and so T six times that they all curve a little or we propose close this at these take into it's probably eight take on and ageing information joint of the so uh in so uh for the uh a so how close is that these in these that uh in this first stage way which change stand close since the D take for a fine for each take and thus thinking class vice use the output put all take class at as it inputs and we use the in yeah class five for that's taking cows five so if the you if we then uh the and so we can then the top you here and if uh W i they is greater than zero then it means take they is positive the core eight at to take i so uh the take or if you information can be a head the read by that's taking cost five okay here you know we discuss uh with these by our experimental state up so uh our baseline is our weenie met the all E matrix two thousand nine audio take in task uh this mess the use cost is sensitive and only use binary class and all uh oh experiments basic the follow the matrix to like to thousand nice it up we use the then forty five take and uh we a little uh audio problem may john mind the with sign which is a web base you C get i we have so many uh a the paper and a it parameters amateurs a they can be select the based on in a course by dish you on a training data and we P cross validation one hundred time okay if you know we show our experiment results uh the the audio annotation is even a by trade but a use the that that is a a a keep we were run the correct uh take to be rank higher and audio retrieval is you rank by take use C and of F major that is given to take we one the correct in is is that is to be dragged higher so we have uh use uh different class Y i different class wise in the first state uh including and uh pose and S yet and the and sample is a combination of the these two uh these two got and we have come for method uh though first slice out matrix baseline line and the second a one is a a it's that the uh close a send the that need only a the sir is staking only and of force is our proposed a a at these taking as we can see that in all cases uh the close is at least expected problem better in a the of uh to other them is the and uh you thus taking only and close to the or learning all only will be better then our matrix baseline okay so i'll cook conclusion she's uh take on a hot and take a if you got two important you formation for so your take pretty should a are time media data and we have first formulate the oh T take should task as a close since T classification problem to minimize the means classified take on and we have uh then for me rate the task as a cost sensitive multi label classification problem and propose close says of these they kid to exploit uh take on and core you formation joint to the and the experiment experiment results show that the new approach oh i'll to prove our matrix two thousand than i we knee have the so uh here we have a a me uh uh a journal paper so please see out the our journal version of this paper of four start uh more details and start it's station walk all this idea okay thank you uh yeah i have a a a a a try to a i i've have used to so of the first mess the is uh transform the uh i'll put all S yeah and and outputs into power would be a T then every average you a proper bit and uh those stick a mess the is to transform the pretty she's goal in into read at rate and uh final decision use the uh every rate thank you