Collaborative Personalization of Image Enhancement

Size: px
Start display at page:

Download "Collaborative Personalization of Image Enhancement"

Transcription

1 Collaborative Personalization of Iage Enhanceent Juan C. Caicedo Universidad Nacional Bogotá, Colobia Ashish Kapoor Microsoft Research Redond, WA. USA Sing Bing Kang Microsoft Research Redond, WA. USA Abstract While ost existing enhanceent tools for photographs have universal auto-enhanceent functionality, recent research [8] shows that users can have personalized preferences. In this paper, we explore whether such personalized preferences in iage enhanceent tend to cluster and whether users can be grouped according to such preferences. To this end, we analyze a coprehensive data set of iage enhanceents collected fro 336 users via Aazon Mechanical Turk. We find that such clusters do exist and can be used to derive ethods to learn statistical preference odels fro a group of users. We also present a probabilistic fraework that exploits the ideas behind collaborative filtering to autoatically enhance novel iages for new users. Experients show that inferring clusters in iage enhanceent preferences results in better prediction of iage enhanceent preferences and outperfors generic auto-correction tools.. Introduction Despite advances in digital photography that iprove caera functionality taking good quality pictures reain a challenge for casual photographers. Probles include incorrect caera settings and poor lighting conditions. Photos ay be iproved using iage enhanceent tools, but anually retouching every single photograph is infeasible. Auto-enhanceent tools are available (e.g., fro Picasa or Windows Live Photo Gallery), but as Kang et al. [8] have shown, users can have significantly different preferences that are not reflected in such tools. We use Kang et al. s observations to explore a new approach for iage enhanceent. We explore if groups of users have siilar preferences. Our hypothesis is that such clusters do exist and that we can utilize that inforation in order to collaboratively enhance iages. Intuitively, we expect that enhanceent preferences should be siilar for users with siilar taste on siilar iages and consequently we can derive ethods that would let us harness the enhanceent efforts of ultiple users to build predictive odels. Such adaptation of the iage enhanceent proble to the collaborative setting is further strengthened by the ever growing use of web-based photo-sharing systes such as Facebook and Flickr. Adding a personalized iage enhanceent coponent to those sites would allow the syste to suggest appropriate enhanceent paraeters for new iages by utilizing prior enhanceents ade by other users with siilar preferences. Our work has three key contributions. First, we describe a coprehensive user study through Aazon Mechanical Turk in order to collect iage enhanceents fro a large nuber of users. Second, we propose a probabilistic odel that explicitly encodes clusterings of user preferences and derive an efficient inference ethod to collaboratively enhance unseen iages for novel users. Finally, we epirically show that explicitly odeling the clustering of user enhanceent preferences leads to better predictions of enhanceent paraeters than the existing one touch-button coercial auto-enhance functionalities. 2. Previous Work The basic idea of iage-derived priors is to exploit inforation in a predefined iage set to autoatically deterine the ost likely paraeters for enhancing an iage. It has been used for a wide range of tasks including denoising [3], color correction [3], and iage restoration [2]. These approaches use generic data sets with a large nuber of iages that do not provide any inforation about user identity or user preferences. Grabler et al. [4] use saple iage anipulations to learn acros to be applied to iages with siilar content. Also, Joshi et al. [7] narrowed the doain of prior iages to a person s favorite photographs to develop a personal iage enhanceent fraework. In principle, these ethods are user-dependent, but no studies were done to establish if their variance across users is significant. The work closest to ours is that of Kang et al. [8], who proposed a odel for personalization of iage enhanceent that learns user preferences by observing her choices in a training set. They conducted a user study that showed 249

2 u u x? x 2? 2 iages x n n n users cluster cluster 2 {, ui, uk} { u, u } Iage Enhanceent Paraeter y y n j, p Enhanceent Paraeter 2 y User in Cluster? User in Cluster 2? Building users database Discovering clusters Enhancing new iages Iage 2 y n Figure. Overview of collaborative personalization of iage enhanceent. A database of users is built, with θi u the observed enhanceent vector for user u and iage x i. Enhanceents are not observed for every cobination of iages and users, which is indicated by the sybol? in the atrix. An algorith is used to discover cluster ebership for users, and to infer enhanceent vectors yi c associated with cluster c for each i th iage. New iages are enhanced based on siilarities to existing iages and cluster to which the user belongs. that different users do have different preferences in iage enhanceent, and deonstrated that personalization is an iportant coponent to iprove the subjective quality of iages. Our ethod builds upon both personalization of iage enhanceent and collaborative filtering techniques. For personalization of iage enhanceent, our approach asks users to enhance a set of training iages using an intuitive interface to collect preference inforation. In this portion of our work, we ipleented ethods siilar to those in Kang et al. [8]. However, our work substantially extends that personalization approach to consider the contribution of other users with siilar preferences. Collaborative filtering is an approach to build recoender systes, which analyzes patterns of user interest in products to provide personalized recoendations [9]. Intuitively, collaborative filtering works by building a database of preferences for ites by users, and atching new users against this database to find neighbors with siilar preferences []. Most existing collaborative filtering ethods attept to predict a scalar quantity, and thus are hard to adapt to our scenario where our goal is to predict a vector of enhanceent paraeters. This is the first work that we know of that proposes the idea of using the wisdo of crowds to enhance iages via collaborative filtering. 3. Overview Figure suarizes the flow of tasks towards building a syste for collaborative iage enhanceent. We first acquire a database of iages that are enhanced by ultiple users, and represent the database as a table (leftost of Figure ) whose rows are individual users and coluns are iages. Each entry in the table corresponds to enhanceent paraeters associated with a different user u and iage x i, and are represented by vector θi u. The iages used for training are a fixed set, and are reasonably representative of iages that need enhanceent. Note that this set of representative iages is large; hence, it is unreasonable to expect that each user will provide enhanceent paraeters for all the iages. Instead, every user enhances only a sall subset of iages, and as such, there are quite a few entries in the table that reain unobserved (denoted as? ). Once this database is acquired, we analyze it to learn a statistical odel that explicitly encodes clustering of users. Specifically, we recover groupings of users and estiate the preference paraeters of each individual group for all the representative iages (iddle of Figure ). Once such a statistical odel is learned, we can enhance a new (i.e., unseen) iage for any cluster by considering its siilarity to the iages in the representative set. 4. Building Blocks Prior to running our user study, we need to pick the knobs required to perfor iage enhanceent and select our representative set of training iages. Also in order to effectively perfor collaborative filtering, we require a distance etric between iages that reflects differences in the respective enhanceent operations required on those iages. Enhanceent Operations: We considered two global enhanceent operations: contrast anipulation (S-curve and Gaa correction) and color correction (tint and teperature), which as with Kang et al. [8], are represented by 5 enhanceent paraeters (two paraeters for S-curve, one each for gaa-correction, tint and teperature). Learning a Distance Metric between Iages: One of the key ingredients in our approach is distance (or siilarity) between iages. In particular, given two iages, we want to estiate the extent to which these iages require siilar enhanceent paraeters. We approach this as a etric learning proble, in which a distance between visual features is adapted to include inforation of enhanceent paraeters. For every iage in the photo collection, we extract 6 color histogras fro the RGB channels and the HSV coponents, which are used to build a unique iage descriptor. Forally, let g(x i ) and g(x j ) be visual feature 250

3 Inforation Gain vectors for two iages x i and x j. We use the Mahalanobis distance in the visual feature space: d A (x i, x j ) = (g(x i ) g(x j )) T A(g(x i ) g(x j )) () Iage Selection via Sensor Placeent Algorith which is paraeterized by a square atrix A to encode weights associated with different features and correlations aongst the in order to reflect the disparity in the enhanceent paraeter space. We learn the atrix A using an online algorith following the ethodology proposed by Jain et al. [6]. The target distance between any two iages is considered to be the distance between their autoenhanceent paraeters. Specifically, we generate constraints of the for (g(x i ), g(x j ), δ ij ), where δ is the distance between the enhanceent paraeters (obtained fro an auto-enhance utility) of the corresponding iages. This specific online algorith is well suited for our task, because our iage collection has approxiately 00 illion constraints, which akes it non-trivial for other distance etric learning ethods to handle. Representative Iage Selection: Selecting a good set of representative iages is critical. We downloaded a set of real photographs fro Flickr, following a sapling through tie strategy, in which the paraeter date taken is randoly changed [, 5]. The list of keywords includes landscape, nature, people, and sports, aong others, to cover a diverse range of photographs without focusing on particular objects. We filtered out iages that are too sall (iages saller than are reoved), blackand-white, and have very little intensity variation (easured by histogra entropy). In the end, our downloaded set consisted of ore than 300,000 photos. Given the set of iages and the distance etric between iages, we need to identify a subset of different iages that represent the original set. This proble was approached as a sensor placeent proble, in which a budget of sensors is available to be located in a space, deciding which positions to chose to axially cover spatial onitoring. In our proble, the sensors are the available iages in the given set where the siilarity (covariance function) is defined by the kernel K ij = exp( d A (x i, x j )/ean(d A (:))). We used the algorith described by Krause et al. [0], in which the eleents that axially reduce the entropy in the original set are iteratively selected. In the Flickr snapshot, we also observed that users tend to upload any siilar iages. We applied the sensor placeent algorith to iages fro the sae user, to keep only the three ost different iages fro each individual. We then cobined these filtered iages and ran the algorith again to select a final photo collection for the user study. We found that with 200 iages, we kept about 84% of the inforation fro a set of approxiately 25,000 candidates. Figure 2 shows how the inforation gain reduces as long as inforative iages are being selected. These iages = 84% inforation (fro 25,000 candidates) Nuber of selected iages Figure 2. The iage selection algorith picks iages that provide axiu increase in utual inforation. This results in a decreasing inforation gain in the selected iage set. 200 iages sufficiently represent the visual feature space defined by the etric that we have learned. 5. User Study The goal of this study was to build the user-iage atrix of enhanceent paraeters. We opted for Aazon Mechanical Turk (MT) to engage people in our iage enhanceent task. Each person had to enhance a sall nuber of iages by selecting the ost preferred enhanced version via an interactive web syste. We anaged to collect enhanceent paraeters fro 336 valid users. Two ain criteria guided the design of the syste for iage enhanceent used in this study: () easy of use and (2) web-oriented. For the first criterion, the user interface does not require paraeter tweaking to apply enhanceents. We ipleented a user interface siilar to the one proposed by Kang et al. [8], in which the syste presents a 3 3 atrix, with an enhanced version of the iage in each cell. In the first iteration, the syste presents the original iage in the center cell surrounded by 8 possible initial enhanceents. When the user clicks one of these possible enhanceents, new siilar enhanceents are coputed to replace the current ones. When the user has identified the ost preferred enhanced version, the syste records the enhanceent paraeters that produce that iage and presents another one. Figure 3 shows a screenshot of this user interface. For the second design criterion, the application is required to be a web-oriented syste, to allow people to enhance iages online through MT. We opted to deploy our application using cloud coputing services, which allows our syste to scale up as needed to copute iage enhanceents online. In our user study, we asked each participant to enhance 20 iages, which are randoly assigned fro the collection of 200 photos. The assignent of iages to users attepts to have approxiately the sae nuber of users enhancing each iage. The axiu tie allowed to coplete this task was hour and the reward was 25

4 Figure 3. Web-based user interface for iage enhanceent Gender F Age older M 40% 30% 20% 0% 0% High School None Basic Education level Soe college Interediate Professional Expertise in Photography Bachelour Masters degree Caera use Ph.D Never Year 3 onths Month Figure 4. Deographics of subjects involved in the MT study. US$.50. The click count, click rate, and tie staps were recorded to identify spaers. Right before starting the task, participants were asked to fill a questionnaire aied to collect deographic inforation, including gender, age bracket, country, and education level. They also have to indicate their level of expertise in photography and frequency of caera use. Figure 4 shows the proportions of these attributes for 336 valid subjects. Most of the subjects are fro USA (43%) and India (4%), with the rest fro ore than 20 other countries. 6. Enhancing Iages Collaboratively Let us denote the collection of all the n iages as X = {x i }. The user study provides us with sets of enhanceent paraeters chosen by all the users. Let θi u be the enhanceent paraeters chosen by user u for iage x i. Because of resource and tie constraints, every user enhances only a sall subset of iages (20 iages out of 200) and consequently there are a lot of iages for which we do not directly observe the enhanceent paraeters corresponding to each user. Our goal is to derive a fraework that can () infer these issing enhanceent paraeters for every user in the study, and ore iportantly, (2) deterine enhanceent paraeters for new iages and users. Our odel is otivated by the ethods in collaborative filtering. In particular, the first key underlying assuption here is that siilar users should have siilar enhanceent Week Daily paraeters for siilar iages. Assuptions like this are also at the heart of any collaborative filtering ethodologies; we could in fact adapt those ethods to infer issing enhanceent paraeters for all the users in the study. However, ost of the work in collaborative filtering focuses on predicting a scalar quantity while our goal is to odel a vector of enhanceent preferences. While off-the-shelf approaches can be individually adapted to each enhanceent paraeter, such siplistic schee ignores the structure of relationships across paraeters and will be sub-optial. In this work, we extend the collaborative filtering to a setting of structured prediction where we jointly predict all the coponents of the enhanceent preferences. Specifically, our odel not only encodes siilarity across iages and users but also odels relationships between different coponents of the enhanceent space. The notion of users siilarity in our syste corresponds to grouping of users into clusters. Thus, users are clustered if they have siilar enhanceent paraeters for all the iages. In the next subsections, we describe () a probabilistic graphical odel for jointly predicting enhanceent preferences that explicitly encodes siilarity across iages and groups users into clusters, (2) an efficient inference algorith, and (3) extensions to do predictions on unseen iages and users that were not part of the user study. 6.. Probabilistic Model for Enhanceents We propose a odel that encodes the dependence of enhanceent paraeters on iage content as well as user preferences. Specifically, given the collection of iages and clustering of users, we assue that there are latent enhanceent preference vectors yi c which correspond to cluster c and iage x i. Further, the enhanceent paraeter vectors we observe in the MT study are siply noisy versions of these latent true enhanceent preferences. Figure 5 illustrates the factor graph corresponding to the proposed odel. The observed enhanceent preferences θi u fro different users are denoted as circles and we introduce a discrete rando variable (squares) h u, u {,.., } for each user that indicates the cluster the user belongs to. We also use Y i to denote collection of all true enhanceent preferences for the i th iage across all the clusterings. The shaded nodes correspond to rando variables that are observed. For exaple, in Figure 5, the enhanceent preferences for user are known for iages 2 and n, while user did not enhance iage 2. This is consistent with our data set where users enhance only a subset of all training iages. The odel also iposes soothness constraints using a GP prior [2] in order to account for the iage content. We use a Gaussian Process prior to enforce the assuption that siilar iages should have siilar preference paraeters. In particular, for each cluster c we induce the GP prior (denoted as GP (c)), where each of the five co- 252

5 User h θ θ 2 Y Y 2 Y n θ n p θ i Y i, h = φ h (θ i, Y i ) p(y Χ) θ θ 2 User h θ n p θ i Y i, h = φ h (θ i, Y i ) Figure 5. Factor graph depicting the proposed odel. Shaded nodes correspond to observed preferences. Not all observations are available for all the iages and users. Square boxes depict latent rando variables corresponding to cluster ebership. ponents of the latent variables yi c and yc j are assued to be jointly Gaussian with zero ean and the covariance specified using a kernel function applied to x i and x j. Forally, GP (c) 5 p= N (yc (p); 0, K). Here y c (p) is the colun vector of p th coponent of enhanceent preference for all iages corresponding to the cluster c and K is a kernel atrix with K ij = k(x i, x j ) and encodes siilarity between pairs of iages. In this paper, we use K ij = exp( d A (x i, x j )/ean(d A (:))). Note, that the GP prior above does not explicitly encode the relationship between the coponents of the paraeters and we can assue that the coponents of yi c have been whitened (zero ean, with unit covariance) beforehand 2. Also, note that all the diensions of the latent variable Y are coupled in the odel and perforing inference will preserve the relationships between different coponents. Let Θ be all the enhanceent preferences fro all the users and for all the iages. Our proposed odel induces a conditional probability distribution p(θ, Y, h X) using the GP prior p(y X), prior probabilities on the cluster ebership p(h), and the potential ters p(θi u x i, Y i, h u ) that link the latent iage preferences to the ones that are observed. Thus, the conditional distribution induced by our odel can be written as p(θ, Y, h X) = p(y X)p(h)p(Θ X, Y, h) Z = k n GP (c) p(h u ) φ hu (θi u, Y i ), Z c= u= i= This kernel atrix is a positive seidefinite atrix and is akin to the kernel atrix used in classifiers such as SVMs. 2 In this work, we whiten the enhanceent preferences before applying the odel and re-project back to orginal space eventulally which results in preservation of the structure of the paraeter space. where Z is the partition function (noralization ter) and the potential φ hu (θ u i, Y i) corresponding to a user u and iage x i takes the following for: φ hu (θ u i, Y i ) e y hu θ i i u 2 2σ 2. (2) Here, y hu i are the hidden rando variable for the sae cluster as the cluster indicated by h u and the iage x i, and σ 2 is the noise paraeter that deterines how tight the relation between the soothness constraint and the final label is. By changing the value of σ we can ephasize or de-ephasize the effect of the GP prior. In suary, the odel provides a powerful fraework for encoding dependence of enhanceent paraeters on iage content (via the GP prior) as well as the clustering of users and allows us to cobine the prior assuptions with the data that is observed in the MT study Inference in the Model Given the observations Θ o fro the MT study the key task is to infer the posterior distribution p(y, h X, Θ o ) over latent true enhanceent preferences and the clustering ebership for all the users. Perforing exact inference is prohibitive as the joint distribution is a product of a Gaussian (GP prior and the φ( ) potentials) and non-gaussian ters (cluster ebership). We resort to approxiate inference techniques in order to get around this proble. In particular, we perfor an approxiate inference by axiizing the variational lower bound with the assuption that the posterior over the unobserved rando variable Y and h can be factorized: F = Y,h log q(y)q(h) log p(y, h X, Θ o) q(y)q(h) Y,h p(y, h X, Θ o ), where q(y) = k 5 c= p= q(yc (p)) is assued to be a Gaussian distribution and q(h) = u= q(h u) is a discrete joint distribution over the unobserved labels. The approxiate inference algorith ais to copute good approxiations q(y) and q(h) to the real posteriors by iteratively optiizing the above described variational bound. Specifically, given the approxiations q t (Y) and q t (h) fro the t th iteration and assuing unifor prior over p(h) the update rules are: q t+ (y c ) GP (c) t (h u=c) q t+ (h u = c) θ u i Θo [φ c (θ u i, Y i )] q θ u i Θo φ c (θ u i, ean(q t+ (Y i ))). Intuitively, the update of iage enhanceent preferences considers the cluster ebership fro the previous iteration and uses it to decide if a data ter should be included 253

6 RMSE RMSE in update for each cluster. Siilarly, the update for the posterior over the cluster ebership considers ean enhanceent preferences fro the previous iteration. Thus, starting fro a rando initialization the paraeters and posterior of the cluster eberships are iteratively updated untill convergence. Upon convergence, we obtain posterior distribution of cluster ebership for each user (q(h)) as well the distribution over the true iage enhanceent preferences (q(y)) approxiated as a Gaussian distribution Handling New Iages and Users Besides inferring about the iages in the database, we are also interested in predicting enhanceent preferences for an iage x test that was not in the original training set as well as for a user who was not part of the user study. The ore straightforward case is where we infer enhanceent preferences for a user u who was a part of the user study and had enhanced training iages for us. Here, we siply need to run the inference algorith where we augent the rando variable Y with Y test, the collection of rando variables that represent true enhanceent paraeters for the new test iage across all the clusterings. Thus, the new iage can be enhanced using the ean of inferred enhanceent preferences corresponding to the cluster the user u belongs to. Note that if we have already perfored variational inference on the training data, inference after the augentation only requires one iteration of update equations. This is because the new iage does not introduce any new inforation about clusterings of the user or enhanceent paraeters, hence, aking it fairly efficient. It is trickier if we need to enhance iages for a user who is not part of the study. However, once we know the cluster ebership of the new user, we can siply use the schee entioned above to estiate the enhanceent preferences. Thus, the proble really reduces down to establishing the cluster ebership of the user. To this end, we can eploy several ethods. For exaple, the new user can enhance soe of the iages fro the training set, which in turn will give us evidence about ebership. The user now can be considered a part of the available corpus and inference can be run as before. Alternatively, we can also engage the user in an introductory dialogue where training iages are used; the resulting enhanceents will indicate ebership. In this work, we use the forer approach of asking users to enhance a subset of training iages to test the extension of the syste to new users and iages. 7. Experients and Results We first perfor experients to deterine the correct nuber of clusters using the data collected in the MT study. Train-test splits were created by randoly choosing 0% of the observed enhanceent vectors in the user-iage atrix as test exaples. The other 90% is used to run the in Error in Paraeter Space Nuber of Clusters Error in Intensity Space Nuber of Clusters Figure 6. Average estiation error on training for different nuber of clusters. ference algorith described in Section 6.2 with σ 2 fixed to 0 3. Evaluating the quality of user adjustents is difficult, since there is no universal agreeent on the perceptual etric. We resorted to the objective easure of intensity difference looking at average Root Mean Square Error (RMSE) in paraeter space (enhanceent vectors) as well as iage intensity space (resulting enhanced iages). Figure 6 highlights the variation of RMSE as we change the nuber of clusters in the odel. Note, that considering only one cluster is equivalent to assuing that all users follow the sae preferences for enhancing iages. We observe that error is highest in both paraeter and intensity spaces when the nuber of clusters is fixed to one and reduces significantly as we start incorporating additional nuber of clusters. This result is consistent with the observations reported by Kang et al. [8], where it was shown that different users do have significanlty different preferences. We also find that the error was iniu in the paraeter space for 3 clusters and pretty close to iniu in the intensity space. Thus, we perfor rest of the analysis and experients by fixing the nuber of clusters to 3. The proposed ethod can be thought of as a clustering algorith with constraints iposed by siilarity in the iage appearance space. Consequently, it is in principle the sae as an EM algorith for learning a Gaussian ixture odel, and it does not prefer that ost users have their own distinct cluster. We think that the increase in error due to further addition of clusters ight be due to the sall nuber of enhanceents operations allowed and we can expect to see different clusterings if additional operations are peritted. Figure 7 shows exaple iages enhanced according to the preferences of users in each of the three clusters, giving a general indication of the preferred appearance in each cluster. While the corrected iages in cluster 2 are slightly ore saturated than the others, iages in cluster 3 see to have ore contrast. A ore quantitative description of the preferred enhanceents in each cluster can be seen in Figure 8, which presents the distribution of the five paraeters as box plots. The box plots are over the eans of the inferred distribution over Y (all 200 iages for the 3 clusters), where the red line in each colun denotes the edian. Note that the largest difference across clusters is for the Power Curve and the S-Curve Inflection Point, suggest- 254

7 RMSE (Intensity) on Test Set Cluster Cluster 2 Cluster 3 Figure 9. Scatter plot of all iages according to paraeters Power- Curve and S-Curve Inflection Point (which partially control global iage contrast). Cluster in red squares, Cluster 2 in green circles and Cluster 3 in blue triangles. Coparison with Existing Systes Figure 7. Exaple iages enhanced according to the preferences discovered for each cluster. ing that variations in contrast is a doinating factor across the clusters. There are soe differences across the rest of the three paraeters as well suggesting a difference in paraeter choice aongst the users. In the scatter plot shown in 9, each point corresponds to an enhanceent preference color-coded by the cluster to which it was assigned. Note that the points are clustered fairly evenly in this 2D projection of the space, further ephasizing the latent structure which is successfully discovered by the inference algorith. We also analyzed statistics of deographic data across the clusters but the statistics were fairly siilar across the clusters. Next, we evaluate the ability of the proposed odel to predict enhanceent preferences for a new user on unseen iages in a collaborative environent. In this test, 0% of all users were randoly chosen as test subjects and inference was perfored using only the data observed for rest of the users. We siulate the real-life situation where every test user provides evidence about its cluster ebership by enhancing soe iages. Since this work is about collaboratively using inforation about iage enhanceent available fro any different users, we show results on scenarios where a test user has provided such little inforation that personalized odel of [8] cannot learn. We look at the RMS error in intensity space as each user enhances one iage at a tie (chosen randoly) and copare the with results of auto-enhanceent fro Picassa and Windows Live Photogallery. Figure 0 presents the plot of RMSE obtained by the Our Syste (3 Clusters) Picasa Photogallery Nuber of Training Iages Enhanced by the Test User Figure 0. Average error of predicting personalized enhanceent paraeters. Auto-enhanceent tools do not learn user preferences. Our approach predicts personalized enhanceents with increasing accuracy as users provide new exaples. syste with increasing nuber of enhanceents available fro the user. We find a consistent reduction in estiation errors as ore enhanceents are added to the personal profile. This is due to the fact that additional enhanced iages provides ore inforation about the cluster ebership of the user, enabling the inference procedure to estiate a better recoendation. We also observe that the perforance of the collaborative approach is far better than the two autoenhanceent tools. While all the three approaches perfor siilarly when there is no evidence about the cluster ebership of the user, we see that the perforance of collaborative strategy greatly iproves as user starts enhancing iages. Also, note that the gains start showing up with as few as 2 iages indicating that strong gains can be obtained with the collaborative enhanceent of iages. Finally, Figure shows several iage exaples with the corresponding enhanced versions produced by Photo- 255

8 Power S Curve Labda S Curve Inflection Teperature Tint Figure 8. Box plots of values for each paraeter (first three control global contrast, last two control color correction). The x-axis indicates cluster nuber and y-axis corresponds to range of values. Red lines denote the edian, end lines are at the lower and upper quartile values, and crosses beyond the ends of the whiskers are outliers. Original Iage PhotoGallery Picassa Cluster 3 We do not clai that our results extend to a uch larger scale with any ore users and a uch richer set of enhanceent knobs; ore experients are required to substantiate such a clai. Future work includes active selection of iages that would enable better estiation of cluster ebership with fewest iage enhanceents by the user. References Figure. Exaple iages with the corresponding enhanced versions generated by auto-enhanceent tools and the proposed approach for a subject that belongs to cluster 3. Gallery and Picassa. Also, the enhanced version of our ethod, for a subject of the cluster 3 is presented. In general, each tool produces a different enhanceent resulting in a different final look and feel. Notice that the suggestion ade by our ethod has been inferred using the subjective preferences learned fro people in cluster 3, thus it is ore likely to be preferred by a user belonging to that population. 8. Conclusion and Future Work We present a novel approach for iage enhanceent that follows a collaborative fraework to recoend corrections based on user preferences. Instead of building a syste that is trained to enhance iages for a specific user, our technique is a principled and practical way for collaboration at a web scale, and can encopass alternative paraeterizations and error criterion. The ain idea is we prediscover the clusters of personalized enhanceent paraeters through collaborative learning on large iage collections. Then, for a new user, we need only figure out which cluster he/she belongs to. Results of the cluster analysis showed the existence of three ain groups, ainly characterized by differences in contrast preferences. Experiental results indicate that the collaborative enhanceent strategy significantly helps in aking better predictions of enhanceent paraeters than existing one touch-button coercial auto-enhance tools. [] T. S. Chua, J. Tang, R. Hong, H. Li, Z. Luo, and Y. Zheng. Nus-wide: a real-world web iage database fro national university of singapore. In CIVR 09. ACM, [2] K. Dale, M. K. Johnson, K. Sunkavalli, W. Matusik, and H. Pfister. Iage restoration using online photo collections. In ICCV. IEEE, [3] R. Fergus, B. Singh, A. Hertzann, S. T. Roweis, and W. T. Freean. Reoving caera shake fro a single photograph. ACM Trans. Graph., 25(3): , [4] F. Grabler, M. Agrawala, W. Li, M. Dontcheva, and T. Igarashi. Generating photo anipulation tutorials by deonstration. ACM Trans. Graph., 28(3): 9, [5] J. Hays and A. A. Efros. Scene copletion using illions of photographs. Coun. ACM, 5(0):87 94, [6] P. Jain, B. Kulis, I. Dhillon, and K. Grauan. Online etric learning and fast siilarity search. In NIPS, [7] N. Joshi, W. Matusik, E. H. Adelson, and D. J. Kriegan. Personal photo enhanceent using exaple iages. ACM Trans. Graph., 29(2): 5, 200. [8] S. B. Kang, A. Kapoor, and D. Lischinski. Personalization of iage enhanceent. Coputer Vision and Pattern Recognition Conference, CVPR, 200. [9] Y. Koren, R. Bell, and C. Volinsky. Matrix factorization techniques for recoender systes. IEEE Coputer Journal, pages 30 37, [0] A. Krause, A. Singh, and C. Guestrin. Near-optial sensor placeents in gaussian processes: Theory, efficient algoriths and epirical studies. J. Mach. Learn. Res., 9: , [] B. Sarwar, G. Karypis, J. Konstan, and J. Riedl. Ite-based collaborative filtering recoendation algoriths. ACM, WWW, 200. [2] M. Seeger. Gaussian Processes for achine learning. International Journal of Neural Systes, 4(2), [3] R. Stanikunas. Investigation of color constancy with a neural network. Neural Networks, 7(3): , April

Media Adaptation Framework in Biofeedback System for Stroke Patient Rehabilitation

Media Adaptation Framework in Biofeedback System for Stroke Patient Rehabilitation Media Adaptation Fraework in Biofeedback Syste for Stroke Patient Rehabilitation Yinpeng Chen, Weiwei Xu, Hari Sundara, Thanassis Rikakis, Sheng-Min Liu Arts, Media and Engineering Progra Arizona State

More information

Extended-Horizon Analysis of Pressure Sensitivities for Leak Detection in Water Distribution Networks: Application to the Barcelona Network

Extended-Horizon Analysis of Pressure Sensitivities for Leak Detection in Water Distribution Networks: Application to the Barcelona Network 2013 European Control Conference (ECC) July 17-19, 2013, Zürich, Switzerland. Extended-Horizon Analysis of Pressure Sensitivities for Leak Detection in Water Distribution Networks: Application to the Barcelona

More information

Applying Multiple Neural Networks on Large Scale Data

Applying Multiple Neural Networks on Large Scale Data 0 International Conference on Inforation and Electronics Engineering IPCSIT vol6 (0) (0) IACSIT Press, Singapore Applying Multiple Neural Networks on Large Scale Data Kritsanatt Boonkiatpong and Sukree

More information

Online Bagging and Boosting

Online Bagging and Boosting Abstract Bagging and boosting are two of the ost well-known enseble learning ethods due to their theoretical perforance guarantees and strong experiental results. However, these algoriths have been used

More information

Searching strategy for multi-target discovery in wireless networks

Searching strategy for multi-target discovery in wireless networks Searching strategy for ulti-target discovery in wireless networks Zhao Cheng, Wendi B. Heinzelan Departent of Electrical and Coputer Engineering University of Rochester Rochester, NY 467 (585) 75-{878,

More information

Machine Learning Applications in Grid Computing

Machine Learning Applications in Grid Computing Machine Learning Applications in Grid Coputing George Cybenko, Guofei Jiang and Daniel Bilar Thayer School of Engineering Dartouth College Hanover, NH 03755, USA gvc@dartouth.edu, guofei.jiang@dartouth.edu

More information

Software Quality Characteristics Tested For Mobile Application Development

Software Quality Characteristics Tested For Mobile Application Development Thesis no: MGSE-2015-02 Software Quality Characteristics Tested For Mobile Application Developent Literature Review and Epirical Survey WALEED ANWAR Faculty of Coputing Blekinge Institute of Technology

More information

Performance Evaluation of Machine Learning Techniques using Software Cost Drivers

Performance Evaluation of Machine Learning Techniques using Software Cost Drivers Perforance Evaluation of Machine Learning Techniques using Software Cost Drivers Manas Gaur Departent of Coputer Engineering, Delhi Technological University Delhi, India ABSTRACT There is a treendous rise

More information

Modeling Parallel Applications Performance on Heterogeneous Systems

Modeling Parallel Applications Performance on Heterogeneous Systems Modeling Parallel Applications Perforance on Heterogeneous Systes Jaeela Al-Jaroodi, Nader Mohaed, Hong Jiang and David Swanson Departent of Coputer Science and Engineering University of Nebraska Lincoln

More information

PERFORMANCE METRICS FOR THE IT SERVICES PORTFOLIO

PERFORMANCE METRICS FOR THE IT SERVICES PORTFOLIO Bulletin of the Transilvania University of Braşov Series I: Engineering Sciences Vol. 4 (53) No. - 0 PERFORMANCE METRICS FOR THE IT SERVICES PORTFOLIO V. CAZACU I. SZÉKELY F. SANDU 3 T. BĂLAN Abstract:

More information

An improved TF-IDF approach for text classification *

An improved TF-IDF approach for text classification * Zhang et al. / J Zheiang Univ SCI 2005 6A(1:49-55 49 Journal of Zheiang University SCIECE ISS 1009-3095 http://www.zu.edu.cn/zus E-ail: zus@zu.edu.cn An iproved TF-IDF approach for text classification

More information

Image restoration for a rectangular poor-pixels detector

Image restoration for a rectangular poor-pixels detector Iage restoration for a rectangular poor-pixels detector Pengcheng Wen 1, Xiangjun Wang 1, Hong Wei 2 1 State Key Laboratory of Precision Measuring Technology and Instruents, Tianjin University, China 2

More information

Analyzing Spatiotemporal Characteristics of Education Network Traffic with Flexible Multiscale Entropy

Analyzing Spatiotemporal Characteristics of Education Network Traffic with Flexible Multiscale Entropy Vol. 9, No. 5 (2016), pp.303-312 http://dx.doi.org/10.14257/ijgdc.2016.9.5.26 Analyzing Spatioteporal Characteristics of Education Network Traffic with Flexible Multiscale Entropy Chen Yang, Renjie Zhou

More information

Exploiting Hardware Heterogeneity within the Same Instance Type of Amazon EC2

Exploiting Hardware Heterogeneity within the Same Instance Type of Amazon EC2 Exploiting Hardware Heterogeneity within the Sae Instance Type of Aazon EC2 Zhonghong Ou, Hao Zhuang, Jukka K. Nurinen, Antti Ylä-Jääski, Pan Hui Aalto University, Finland; Deutsch Teleko Laboratories,

More information

Fuzzy Sets in HR Management

Fuzzy Sets in HR Management Acta Polytechnica Hungarica Vol. 8, No. 3, 2011 Fuzzy Sets in HR Manageent Blanka Zeková AXIOM SW, s.r.o., 760 01 Zlín, Czech Republic blanka.zekova@sezna.cz Jana Talašová Faculty of Science, Palacký Univerzity,

More information

Evaluating Inventory Management Performance: a Preliminary Desk-Simulation Study Based on IOC Model

Evaluating Inventory Management Performance: a Preliminary Desk-Simulation Study Based on IOC Model Evaluating Inventory Manageent Perforance: a Preliinary Desk-Siulation Study Based on IOC Model Flora Bernardel, Roberto Panizzolo, and Davide Martinazzo Abstract The focus of this study is on preliinary

More information

High Performance Chinese/English Mixed OCR with Character Level Language Identification

High Performance Chinese/English Mixed OCR with Character Level Language Identification 2009 0th International Conference on Docuent Analysis and Recognition High Perforance Chinese/English Mixed OCR with Character Level Language Identification Kai Wang Institute of Machine Intelligence,

More information

Data Set Generation for Rectangular Placement Problems

Data Set Generation for Rectangular Placement Problems Data Set Generation for Rectangular Placeent Probles Christine L. Valenzuela (Muford) Pearl Y. Wang School of Coputer Science & Inforatics Departent of Coputer Science MS 4A5 Cardiff University George

More information

Dynamic Placement for Clustered Web Applications

Dynamic Placement for Clustered Web Applications Dynaic laceent for Clustered Web Applications A. Karve, T. Kibrel, G. acifici, M. Spreitzer, M. Steinder, M. Sviridenko, and A. Tantawi IBM T.J. Watson Research Center {karve,kibrel,giovanni,spreitz,steinder,sviri,tantawi}@us.ib.co

More information

Real Time Target Tracking with Binary Sensor Networks and Parallel Computing

Real Time Target Tracking with Binary Sensor Networks and Parallel Computing Real Tie Target Tracking with Binary Sensor Networks and Parallel Coputing Hong Lin, John Rushing, Sara J. Graves, Steve Tanner, and Evans Criswell Abstract A parallel real tie data fusion and target tracking

More information

CRM FACTORS ASSESSMENT USING ANALYTIC HIERARCHY PROCESS

CRM FACTORS ASSESSMENT USING ANALYTIC HIERARCHY PROCESS 641 CRM FACTORS ASSESSMENT USING ANALYTIC HIERARCHY PROCESS Marketa Zajarosova 1* *Ph.D. VSB - Technical University of Ostrava, THE CZECH REPUBLIC arketa.zajarosova@vsb.cz Abstract Custoer relationship

More information

A Scalable Application Placement Controller for Enterprise Data Centers

A Scalable Application Placement Controller for Enterprise Data Centers W WWW 7 / Track: Perforance and Scalability A Scalable Application Placeent Controller for Enterprise Data Centers Chunqiang Tang, Malgorzata Steinder, Michael Spreitzer, and Giovanni Pacifici IBM T.J.

More information

Modeling operational risk data reported above a time-varying threshold

Modeling operational risk data reported above a time-varying threshold Modeling operational risk data reported above a tie-varying threshold Pavel V. Shevchenko CSIRO Matheatical and Inforation Sciences, Sydney, Locked bag 7, North Ryde, NSW, 670, Australia. e-ail: Pavel.Shevchenko@csiro.au

More information

Energy Proportionality for Disk Storage Using Replication

Energy Proportionality for Disk Storage Using Replication Energy Proportionality for Disk Storage Using Replication Jinoh Ki and Doron Rote Lawrence Berkeley National Laboratory University of California, Berkeley, CA 94720 {jinohki,d rote}@lbl.gov Abstract Energy

More information

Cross-Domain Metric Learning Based on Information Theory

Cross-Domain Metric Learning Based on Information Theory Proceedings of the Twenty-Eighth AAAI Conference on Artificial Intelligence Cross-Doain Metric Learning Based on Inforation Theory Hao Wang,2, Wei Wang 2,3, Chen Zhang 2, Fanjiang Xu 2. State Key Laboratory

More information

AN ALGORITHM FOR REDUCING THE DIMENSION AND SIZE OF A SAMPLE FOR DATA EXPLORATION PROCEDURES

AN ALGORITHM FOR REDUCING THE DIMENSION AND SIZE OF A SAMPLE FOR DATA EXPLORATION PROCEDURES Int. J. Appl. Math. Coput. Sci., 2014, Vol. 24, No. 1, 133 149 DOI: 10.2478/acs-2014-0011 AN ALGORITHM FOR REDUCING THE DIMENSION AND SIZE OF A SAMPLE FOR DATA EXPLORATION PROCEDURES PIOTR KULCZYCKI,,

More information

The Research of Measuring Approach and Energy Efficiency for Hadoop Periodic Jobs

The Research of Measuring Approach and Energy Efficiency for Hadoop Periodic Jobs Send Orders for Reprints to reprints@benthascience.ae 206 The Open Fuels & Energy Science Journal, 2015, 8, 206-210 Open Access The Research of Measuring Approach and Energy Efficiency for Hadoop Periodic

More information

Energy Efficient VM Scheduling for Cloud Data Centers: Exact allocation and migration algorithms

Energy Efficient VM Scheduling for Cloud Data Centers: Exact allocation and migration algorithms Energy Efficient VM Scheduling for Cloud Data Centers: Exact allocation and igration algoriths Chaia Ghribi, Makhlouf Hadji and Djaal Zeghlache Institut Mines-Téléco, Téléco SudParis UMR CNRS 5157 9, Rue

More information

An Integrated Approach for Monitoring Service Level Parameters of Software-Defined Networking

An Integrated Approach for Monitoring Service Level Parameters of Software-Defined Networking International Journal of Future Generation Counication and Networking Vol. 8, No. 6 (15), pp. 197-4 http://d.doi.org/1.1457/ijfgcn.15.8.6.19 An Integrated Approach for Monitoring Service Level Paraeters

More information

SAMPLING METHODS LEARNING OBJECTIVES

SAMPLING METHODS LEARNING OBJECTIVES 6 SAMPLING METHODS 6 Using Statistics 6-6 2 Nonprobability Sapling and Bias 6-6 Stratified Rando Sapling 6-2 6 4 Cluster Sapling 6-4 6 5 Systeatic Sapling 6-9 6 6 Nonresponse 6-2 6 7 Suary and Review of

More information

Data Streaming Algorithms for Estimating Entropy of Network Traffic

Data Streaming Algorithms for Estimating Entropy of Network Traffic Data Streaing Algoriths for Estiating Entropy of Network Traffic Ashwin Lall University of Rochester Vyas Sekar Carnegie Mellon University Mitsunori Ogihara University of Rochester Jun (Ji) Xu Georgia

More information

Construction Economics & Finance. Module 3 Lecture-1

Construction Economics & Finance. Module 3 Lecture-1 Depreciation:- Construction Econoics & Finance Module 3 Lecture- It represents the reduction in arket value of an asset due to age, wear and tear and obsolescence. The physical deterioration of the asset

More information

Physics 211: Lab Oscillations. Simple Harmonic Motion.

Physics 211: Lab Oscillations. Simple Harmonic Motion. Physics 11: Lab Oscillations. Siple Haronic Motion. Reading Assignent: Chapter 15 Introduction: As we learned in class, physical systes will undergo an oscillatory otion, when displaced fro a stable equilibriu.

More information

6. Time (or Space) Series Analysis

6. Time (or Space) Series Analysis ATM 55 otes: Tie Series Analysis - Section 6a Page 8 6. Tie (or Space) Series Analysis In this chapter we will consider soe coon aspects of tie series analysis including autocorrelation, statistical prediction,

More information

Research Article Performance Evaluation of Human Resource Outsourcing in Food Processing Enterprises

Research Article Performance Evaluation of Human Resource Outsourcing in Food Processing Enterprises Advance Journal of Food Science and Technology 9(2): 964-969, 205 ISSN: 2042-4868; e-issn: 2042-4876 205 Maxwell Scientific Publication Corp. Subitted: August 0, 205 Accepted: Septeber 3, 205 Published:

More information

Cooperative Caching for Adaptive Bit Rate Streaming in Content Delivery Networks

Cooperative Caching for Adaptive Bit Rate Streaming in Content Delivery Networks Cooperative Caching for Adaptive Bit Rate Streaing in Content Delivery Networs Phuong Luu Vo Departent of Coputer Science and Engineering, International University - VNUHCM, Vietna vtlphuong@hciu.edu.vn

More information

Audio Engineering Society. Convention Paper. Presented at the 119th Convention 2005 October 7 10 New York, New York USA

Audio Engineering Society. Convention Paper. Presented at the 119th Convention 2005 October 7 10 New York, New York USA Audio Engineering Society Convention Paper Presented at the 119th Convention 2005 October 7 10 New York, New York USA This convention paper has been reproduced fro the authors advance anuscript, without

More information

Factored Models for Probabilistic Modal Logic

Factored Models for Probabilistic Modal Logic Proceedings of the Twenty-Third AAAI Conference on Artificial Intelligence (2008 Factored Models for Probabilistic Modal Logic Afsaneh Shirazi and Eyal Air Coputer Science Departent, University of Illinois

More information

INTEGRATED ENVIRONMENT FOR STORING AND HANDLING INFORMATION IN TASKS OF INDUCTIVE MODELLING FOR BUSINESS INTELLIGENCE SYSTEMS

INTEGRATED ENVIRONMENT FOR STORING AND HANDLING INFORMATION IN TASKS OF INDUCTIVE MODELLING FOR BUSINESS INTELLIGENCE SYSTEMS Artificial Intelligence Methods and Techniques for Business and Engineering Applications 210 INTEGRATED ENVIRONMENT FOR STORING AND HANDLING INFORMATION IN TASKS OF INDUCTIVE MODELLING FOR BUSINESS INTELLIGENCE

More information

This paper studies a rental firm that offers reusable products to price- and quality-of-service sensitive

This paper studies a rental firm that offers reusable products to price- and quality-of-service sensitive MANUFACTURING & SERVICE OPERATIONS MANAGEMENT Vol., No. 3, Suer 28, pp. 429 447 issn 523-464 eissn 526-5498 8 3 429 infors doi.287/so.7.8 28 INFORMS INFORMS holds copyright to this article and distributed

More information

Managing Complex Network Operation with Predictive Analytics

Managing Complex Network Operation with Predictive Analytics Managing Coplex Network Operation with Predictive Analytics Zhenyu Huang, Pak Chung Wong, Patrick Mackey, Yousu Chen, Jian Ma, Kevin Schneider, and Frank L. Greitzer Pacific Northwest National Laboratory

More information

Use of extrapolation to forecast the working capital in the mechanical engineering companies

Use of extrapolation to forecast the working capital in the mechanical engineering companies ECONTECHMOD. AN INTERNATIONAL QUARTERLY JOURNAL 2014. Vol. 1. No. 1. 23 28 Use of extrapolation to forecast the working capital in the echanical engineering copanies A. Cherep, Y. Shvets Departent of finance

More information

Adaptive Modulation and Coding for Unmanned Aerial Vehicle (UAV) Radio Channel

Adaptive Modulation and Coding for Unmanned Aerial Vehicle (UAV) Radio Channel Recent Advances in Counications Adaptive odulation and Coding for Unanned Aerial Vehicle (UAV) Radio Channel Airhossein Fereidountabar,Gian Carlo Cardarilli, Rocco Fazzolari,Luca Di Nunzio Abstract In

More information

COMBINING CRASH RECORDER AND PAIRED COMPARISON TECHNIQUE: INJURY RISK FUNCTIONS IN FRONTAL AND REAR IMPACTS WITH SPECIAL REFERENCE TO NECK INJURIES

COMBINING CRASH RECORDER AND PAIRED COMPARISON TECHNIQUE: INJURY RISK FUNCTIONS IN FRONTAL AND REAR IMPACTS WITH SPECIAL REFERENCE TO NECK INJURIES COMBINING CRASH RECORDER AND AIRED COMARISON TECHNIQUE: INJURY RISK FUNCTIONS IN FRONTAL AND REAR IMACTS WITH SECIAL REFERENCE TO NECK INJURIES Anders Kullgren, Maria Krafft Folksa Research, 66 Stockhol,

More information

ASIC Design Project Management Supported by Multi Agent Simulation

ASIC Design Project Management Supported by Multi Agent Simulation ASIC Design Project Manageent Supported by Multi Agent Siulation Jana Blaschke, Christian Sebeke, Wolfgang Rosenstiel Abstract The coplexity of Application Specific Integrated Circuits (ASICs) is continuously

More information

An Innovate Dynamic Load Balancing Algorithm Based on Task

An Innovate Dynamic Load Balancing Algorithm Based on Task An Innovate Dynaic Load Balancing Algorith Based on Task Classification Hong-bin Wang,,a, Zhi-yi Fang, b, Guan-nan Qu,*,c, Xiao-dan Ren,d College of Coputer Science and Technology, Jilin University, Changchun

More information

The AGA Evaluating Model of Customer Loyalty Based on E-commerce Environment

The AGA Evaluating Model of Customer Loyalty Based on E-commerce Environment 6 JOURNAL OF SOFTWARE, VOL. 4, NO. 3, MAY 009 The AGA Evaluating Model of Custoer Loyalty Based on E-coerce Environent Shaoei Yang Econoics and Manageent Departent, North China Electric Power University,

More information

An Improved Decision-making Model of Human Resource Outsourcing Based on Internet Collaboration

An Improved Decision-making Model of Human Resource Outsourcing Based on Internet Collaboration International Journal of Hybrid Inforation Technology, pp. 339-350 http://dx.doi.org/10.14257/hit.2016.9.4.28 An Iproved Decision-aking Model of Huan Resource Outsourcing Based on Internet Collaboration

More information

Design of Model Reference Self Tuning Mechanism for PID like Fuzzy Controller

Design of Model Reference Self Tuning Mechanism for PID like Fuzzy Controller Research Article International Journal of Current Engineering and Technology EISSN 77 46, PISSN 347 56 4 INPRESSCO, All Rights Reserved Available at http://inpressco.co/category/ijcet Design of Model Reference

More information

Implementation of Active Queue Management in a Combined Input and Output Queued Switch

Implementation of Active Queue Management in a Combined Input and Output Queued Switch pleentation of Active Queue Manageent in a obined nput and Output Queued Switch Bartek Wydrowski and Moshe Zukeran AR Special Research entre for Ultra-Broadband nforation Networks, EEE Departent, The University

More information

A quantum secret ballot. Abstract

A quantum secret ballot. Abstract A quantu secret ballot Shahar Dolev and Itaar Pitowsky The Edelstein Center, Levi Building, The Hebrerw University, Givat Ra, Jerusale, Israel Boaz Tair arxiv:quant-ph/060087v 8 Mar 006 Departent of Philosophy

More information

Evaluating the Effectiveness of Task Overlapping as a Risk Response Strategy in Engineering Projects

Evaluating the Effectiveness of Task Overlapping as a Risk Response Strategy in Engineering Projects Evaluating the Effectiveness of Task Overlapping as a Risk Response Strategy in Engineering Projects Lucas Grèze Robert Pellerin Nathalie Perrier Patrice Leclaire February 2011 CIRRELT-2011-11 Bureaux

More information

Capacity of Multiple-Antenna Systems With Both Receiver and Transmitter Channel State Information

Capacity of Multiple-Antenna Systems With Both Receiver and Transmitter Channel State Information IEEE TRANSACTIONS ON INFORMATION THEORY, VOL. 49, NO., OCTOBER 23 2697 Capacity of Multiple-Antenna Systes With Both Receiver and Transitter Channel State Inforation Sudharan K. Jayaweera, Student Meber,

More information

Entity Search Engine: Towards Agile Best-Effort Information Integration over the Web

Entity Search Engine: Towards Agile Best-Effort Information Integration over the Web Entity Search Engine: Towards Agile Best-Effort Inforation Integration over the Web Tao Cheng, Kevin Chen-Chuan Chang University of Illinois at Urbana-Chapaign {tcheng3, kcchang}@cs.uiuc.edu. INTRODUCTION

More information

Generating Certification Authority Authenticated Public Keys in Ad Hoc Networks

Generating Certification Authority Authenticated Public Keys in Ad Hoc Networks SECURITY AND COMMUNICATION NETWORKS Published online in Wiley InterScience (www.interscience.wiley.co). Generating Certification Authority Authenticated Public Keys in Ad Hoc Networks G. Kounga 1, C. J.

More information

Reliability Constrained Packet-sizing for Linear Multi-hop Wireless Networks

Reliability Constrained Packet-sizing for Linear Multi-hop Wireless Networks Reliability Constrained acket-sizing for inear Multi-hop Wireless Networks Ning Wen, and Randall A. Berry Departent of Electrical Engineering and Coputer Science Northwestern University, Evanston, Illinois

More information

An Approach to Combating Free-riding in Peer-to-Peer Networks

An Approach to Combating Free-riding in Peer-to-Peer Networks An Approach to Cobating Free-riding in Peer-to-Peer Networks Victor Ponce, Jie Wu, and Xiuqi Li Departent of Coputer Science and Engineering Florida Atlantic University Boca Raton, FL 33431 April 7, 2008

More information

Gaussian Processes for Regression: A Quick Introduction

Gaussian Processes for Regression: A Quick Introduction Gaussian Processes for Regression A Quick Introduction M Ebden, August 28 Coents to arkebden@engoacuk MOTIVATION Figure illustrates a typical eaple of a prediction proble given soe noisy observations of

More information

130: Rule-based Expert Systems. Ajith Abraham Oklahoma State University, Stillwater, OK, USA 1 PROBLEM SOLVING USING HEURISTICS

130: Rule-based Expert Systems. Ajith Abraham Oklahoma State University, Stillwater, OK, USA 1 PROBLEM SOLVING USING HEURISTICS : Rule-based Expert Systes Ajith Abraha Oklahoa State University, Stillwater, OK, USA Proble Solving Using Heuristics 99 What are Rule-based Systes? 9 Inference Engine in Rule-based Systes 9 Expert Syste

More information

arxiv:0805.1434v1 [math.pr] 9 May 2008

arxiv:0805.1434v1 [math.pr] 9 May 2008 Degree-distribution stability of scale-free networs Zhenting Hou, Xiangxing Kong, Dinghua Shi,2, and Guanrong Chen 3 School of Matheatics, Central South University, Changsha 40083, China 2 Departent of

More information

Preference-based Search and Multi-criteria Optimization

Preference-based Search and Multi-criteria Optimization Fro: AAAI-02 Proceedings. Copyright 2002, AAAI (www.aaai.org). All rights reserved. Preference-based Search and Multi-criteria Optiization Ulrich Junker ILOG 1681, route des Dolines F-06560 Valbonne ujunker@ilog.fr

More information

Quality evaluation of the model-based forecasts of implied volatility index

Quality evaluation of the model-based forecasts of implied volatility index Quality evaluation of the odel-based forecasts of iplied volatility index Katarzyna Łęczycka 1 Abstract Influence of volatility on financial arket forecasts is very high. It appears as a specific factor

More information

Budget-optimal Crowdsourcing using Low-rank Matrix Approximations

Budget-optimal Crowdsourcing using Low-rank Matrix Approximations Budget-optial Crowdsourcing using Low-rank Matrix Approxiations David R. Karger, Sewoong Oh, and Devavrat Shah Departent of EECS, Massachusetts Institute of Technology Eail: {karger, swoh, devavrat}@it.edu

More information

How To Balance Over Redundant Wireless Sensor Networks Based On Diffluent

How To Balance Over Redundant Wireless Sensor Networks Based On Diffluent Load balancing over redundant wireless sensor networks based on diffluent Abstract Xikui Gao Yan ai Yun Ju School of Control and Coputer Engineering North China Electric ower University 02206 China Received

More information

The Velocities of Gas Molecules

The Velocities of Gas Molecules he Velocities of Gas Molecules by Flick Colean Departent of Cheistry Wellesley College Wellesley MA 8 Copyright Flick Colean 996 All rights reserved You are welcoe to use this docuent in your own classes

More information

Botnets Detection Based on IRC-Community

Botnets Detection Based on IRC-Community Botnets Detection Based on IRC-Counity Wei Lu and Ali A. Ghorbani Network Security Laboratory, Faculty of Coputer Science University of New Brunswick, Fredericton, NB E3B 5A3, Canada {wlu, ghorbani}@unb.ca

More information

ENZYME KINETICS: THEORY. A. Introduction

ENZYME KINETICS: THEORY. A. Introduction ENZYME INETICS: THEORY A. Introduction Enzyes are protein olecules coposed of aino acids and are anufactured by the living cell. These olecules provide energy for the organis by catalyzing various biocheical

More information

Stable Learning in Coding Space for Multi-Class Decoding and Its Extension for Multi-Class Hypothesis Transfer Learning

Stable Learning in Coding Space for Multi-Class Decoding and Its Extension for Multi-Class Hypothesis Transfer Learning Stable Learning in Coding Space for Multi-Class Decoding and Its Extension for Multi-Class Hypothesis Transfer Learning Bang Zhang, Yi Wang 2, Yang Wang, Fang Chen 2 National ICT Australia 2 School of

More information

The Benefit of SMT in the Multi-Core Era: Flexibility towards Degrees of Thread-Level Parallelism

The Benefit of SMT in the Multi-Core Era: Flexibility towards Degrees of Thread-Level Parallelism The enefit of SMT in the Multi-Core Era: Flexibility towards Degrees of Thread-Level Parallelis Stijn Eyeran Lieven Eeckhout Ghent University, elgiu Stijn.Eyeran@elis.UGent.be, Lieven.Eeckhout@elis.UGent.be

More information

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, ACCEPTED FOR PUBLICATION 1. Secure Wireless Multicast for Delay-Sensitive Data via Network Coding

IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, ACCEPTED FOR PUBLICATION 1. Secure Wireless Multicast for Delay-Sensitive Data via Network Coding IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS, ACCEPTED FOR PUBLICATION 1 Secure Wireless Multicast for Delay-Sensitive Data via Network Coding Tuan T. Tran, Meber, IEEE, Hongxiang Li, Senior Meber, IEEE,

More information

A framework for performance monitoring, load balancing, adaptive timeouts and quality of service in digital libraries

A framework for performance monitoring, load balancing, adaptive timeouts and quality of service in digital libraries Int J Digit Libr (2000) 3: 9 35 INTERNATIONAL JOURNAL ON Digital Libraries Springer-Verlag 2000 A fraework for perforance onitoring, load balancing, adaptive tieouts and quality of service in digital libraries

More information

Optimal Resource-Constraint Project Scheduling with Overlapping Modes

Optimal Resource-Constraint Project Scheduling with Overlapping Modes Optial Resource-Constraint Proect Scheduling with Overlapping Modes François Berthaut Lucas Grèze Robert Pellerin Nathalie Perrier Adnène Hai February 20 CIRRELT-20-09 Bureaux de Montréal : Bureaux de

More information

Fuzzy approach for searching in CRM systems

Fuzzy approach for searching in CRM systems Fuzzy approach for searching in CRM systes BOGDAN WALEK, JIŘÍ BARTOŠ, CYRIL KLIMEŠ Departent of Inforatics and Coputers University of Ostrava 30. dubna 22, 701 03 Ostrava CZECH REPUBLIC bogdan.walek@osu.cz,

More information

Modeling Cooperative Gene Regulation Using Fast Orthogonal Search

Modeling Cooperative Gene Regulation Using Fast Orthogonal Search 8 The Open Bioinforatics Journal, 28, 2, 8-89 Open Access odeling Cooperative Gene Regulation Using Fast Orthogonal Search Ian inz* and ichael J. Korenberg* Departent of Electrical and Coputer Engineering,

More information

ADJUSTING FOR QUALITY CHANGE

ADJUSTING FOR QUALITY CHANGE ADJUSTING FOR QUALITY CHANGE 7 Introduction 7.1 The easureent of changes in the level of consuer prices is coplicated by the appearance and disappearance of new and old goods and services, as well as changes

More information

AUC Optimization vs. Error Rate Minimization

AUC Optimization vs. Error Rate Minimization AUC Optiization vs. Error Rate Miniization Corinna Cortes and Mehryar Mohri AT&T Labs Research 180 Park Avenue, Florha Park, NJ 0793, USA {corinna, ohri}@research.att.co Abstract The area under an ROC

More information

Equivalent Tapped Delay Line Channel Responses with Reduced Taps

Equivalent Tapped Delay Line Channel Responses with Reduced Taps Equivalent Tapped Delay Line Channel Responses with Reduced Taps Shweta Sagari, Wade Trappe, Larry Greenstein {shsagari, trappe, ljg}@winlab.rutgers.edu WINLAB, Rutgers University, North Brunswick, NJ

More information

Towards Change Management Capability Assessment Model for Contractors in Building Project

Towards Change Management Capability Assessment Model for Contractors in Building Project Middle-East Journal of Scientific Research 23 (7): 1327-1333, 2015 ISSN 1990-9233 IDOSI Publications, 2015 DOI: 10.5829/idosi.ejsr.2015.23.07.120 Towards Change Manageent Capability Assessent Model for

More information

Multi-Class Deep Boosting

Multi-Class Deep Boosting Multi-Class Deep Boosting Vitaly Kuznetsov Courant Institute 25 Mercer Street New York, NY 002 vitaly@cis.nyu.edu Mehryar Mohri Courant Institute & Google Research 25 Mercer Street New York, NY 002 ohri@cis.nyu.edu

More information

Partitioned Elias-Fano Indexes

Partitioned Elias-Fano Indexes Partitioned Elias-ano Indexes Giuseppe Ottaviano ISTI-CNR, Pisa giuseppe.ottaviano@isti.cnr.it Rossano Venturini Dept. of Coputer Science, University of Pisa rossano@di.unipi.it ABSTRACT The Elias-ano

More information

International Journal of Management & Information Systems First Quarter 2012 Volume 16, Number 1

International Journal of Management & Information Systems First Quarter 2012 Volume 16, Number 1 International Journal of Manageent & Inforation Systes First Quarter 2012 Volue 16, Nuber 1 Proposal And Effectiveness Of A Highly Copelling Direct Mail Method - Establishent And Deployent Of PMOS-DM Hisatoshi

More information

Enrolment into Higher Education and Changes in Repayment Obligations of Student Aid Microeconometric Evidence for Germany

Enrolment into Higher Education and Changes in Repayment Obligations of Student Aid Microeconometric Evidence for Germany Enrolent into Higher Education and Changes in Repayent Obligations of Student Aid Microeconoetric Evidence for Gerany Hans J. Baugartner *) Viktor Steiner **) *) DIW Berlin **) Free University of Berlin,

More information

SOME APPLICATIONS OF FORECASTING Prof. Thomas B. Fomby Department of Economics Southern Methodist University May 2008

SOME APPLICATIONS OF FORECASTING Prof. Thomas B. Fomby Department of Economics Southern Methodist University May 2008 SOME APPLCATONS OF FORECASTNG Prof. Thoas B. Foby Departent of Econoics Southern Methodist University May 8 To deonstrate the usefulness of forecasting ethods this note discusses four applications of forecasting

More information

Binary Embedding: Fundamental Limits and Fast Algorithm

Binary Embedding: Fundamental Limits and Fast Algorithm Binary Ebedding: Fundaental Liits and Fast Algorith Xinyang Yi The University of Texas at Austin yixy@utexas.edu Eric Price The University of Texas at Austin ecprice@cs.utexas.edu Constantine Caraanis

More information

Comment on On Discriminative vs. Generative Classifiers: A Comparison of Logistic Regression and Naive Bayes

Comment on On Discriminative vs. Generative Classifiers: A Comparison of Logistic Regression and Naive Bayes Coent on On Discriinative vs. Generative Classifiers: A Coparison of Logistic Regression and Naive Bayes Jing-Hao Xue (jinghao@stats.gla.ac.uk) and D. Michael Titterington (ike@stats.gla.ac.uk) Departent

More information

ESTIMATING LIQUIDITY PREMIA IN THE SPANISH GOVERNMENT SECURITIES MARKET

ESTIMATING LIQUIDITY PREMIA IN THE SPANISH GOVERNMENT SECURITIES MARKET ESTIMATING LIQUIDITY PREMIA IN THE SPANISH GOVERNMENT SECURITIES MARKET Francisco Alonso, Roberto Blanco, Ana del Río and Alicia Sanchis Banco de España Banco de España Servicio de Estudios Docuento de

More information

Driving Behavior Analysis Based on Vehicle OBD Information and AdaBoost Algorithms

Driving Behavior Analysis Based on Vehicle OBD Information and AdaBoost Algorithms , March 18-20, 2015, Hong Kong Driving Behavior Analysis Based on Vehicle OBD Inforation and AdaBoost Algoriths Shi-Huang Chen, Jeng-Shyang Pan, and Kaixuan Lu Abstract This paper proposes a novel driving

More information

Online Methods for Multi-Domain Learning and Adaptation

Online Methods for Multi-Domain Learning and Adaptation Online Methods for Multi-Doain Learning and Adaptation Mark Dredze and Koby Craer Departent of Coputer and Inforation Science University of Pennsylvania Philadelphia, PA 19104 USA {dredze,craer}@cis.upenn.edu

More information

Local Area Network Management

Local Area Network Management Technology Guidelines for School Coputer-based Technologies Local Area Network Manageent Local Area Network Manageent Introduction This docuent discusses the tasks associated with anageent of Local Area

More information

REQUIREMENTS FOR A COMPUTER SCIENCE CURRICULUM EMPHASIZING INFORMATION TECHNOLOGY SUBJECT AREA: CURRICULUM ISSUES

REQUIREMENTS FOR A COMPUTER SCIENCE CURRICULUM EMPHASIZING INFORMATION TECHNOLOGY SUBJECT AREA: CURRICULUM ISSUES REQUIREMENTS FOR A COMPUTER SCIENCE CURRICULUM EMPHASIZING INFORMATION TECHNOLOGY SUBJECT AREA: CURRICULUM ISSUES Charles Reynolds Christopher Fox reynolds @cs.ju.edu fox@cs.ju.edu Departent of Coputer

More information

Position Auctions and Non-uniform Conversion Rates

Position Auctions and Non-uniform Conversion Rates Position Auctions and Non-unifor Conversion Rates Liad Blurosen Microsoft Research Mountain View, CA 944 liadbl@icrosoft.co Jason D. Hartline Shuzhen Nong Electrical Engineering and Microsoft AdCenter

More information

Leak detection in open water channels

Leak detection in open water channels Proceedings of the 17th World Congress The International Federation of Autoatic Control Seoul, Korea, July 6-11, 28 Leak detection in open water channels Erik Weyer Georges Bastin Departent of Electrical

More information

Markovian inventory policy with application to the paper industry

Markovian inventory policy with application to the paper industry Coputers and Cheical Engineering 26 (2002) 1399 1413 www.elsevier.co/locate/copcheeng Markovian inventory policy with application to the paper industry K. Karen Yin a, *, Hu Liu a,1, Neil E. Johnson b,2

More information

The Model of Lines for Option Pricing with Jumps

The Model of Lines for Option Pricing with Jumps The Model of Lines for Option Pricing with Jups Claudio Albanese, Sebastian Jaiungal and Ditri H Rubisov January 17, 21 Departent of Matheatics, University of Toronto Abstract This article reviews a pricing

More information

Airline Yield Management with Overbooking, Cancellations, and No-Shows JANAKIRAM SUBRAMANIAN

Airline Yield Management with Overbooking, Cancellations, and No-Shows JANAKIRAM SUBRAMANIAN Airline Yield Manageent with Overbooking, Cancellations, and No-Shows JANAKIRAM SUBRAMANIAN Integral Developent Corporation, 301 University Avenue, Suite 200, Palo Alto, California 94301 SHALER STIDHAM

More information

The individual neurons are complicated. They have a myriad of parts, subsystems and control mechanisms. They convey information via a host of

The individual neurons are complicated. They have a myriad of parts, subsystems and control mechanisms. They convey information via a host of CHAPTER 4 ARTIFICIAL NEURAL NETWORKS 4. INTRODUCTION Artificial Neural Networks (ANNs) are relatively crude electronic odels based on the neural structure of the brain. The brain learns fro experience.

More information

Information Processing Letters

Information Processing Letters Inforation Processing Letters 111 2011) 178 183 Contents lists available at ScienceDirect Inforation Processing Letters www.elsevier.co/locate/ipl Offline file assignents for online load balancing Paul

More information

Presentation Safety Legislation and Standards

Presentation Safety Legislation and Standards levels in different discrete levels corresponding for each one to a probability of dangerous failure per hour: > > The table below gives the relationship between the perforance level (PL) and the Safety

More information

On Computing Nearest Neighbors with Applications to Decoding of Binary Linear Codes

On Computing Nearest Neighbors with Applications to Decoding of Binary Linear Codes On Coputing Nearest Neighbors with Applications to Decoding of Binary Linear Codes Alexander May and Ilya Ozerov Horst Görtz Institute for IT-Security Ruhr-University Bochu, Gerany Faculty of Matheatics

More information

Halloween Costume Ideas for the Wii Game

Halloween Costume Ideas for the Wii Game Algorithica 2001) 30: 101 139 DOI: 101007/s00453-001-0003-0 Algorithica 2001 Springer-Verlag New York Inc Optial Search and One-Way Trading Online Algoriths R El-Yaniv, 1 A Fiat, 2 R M Karp, 3 and G Turpin

More information