In this poster, we present iCAP (In Class Activity Participation), an Audience Response System (ARS) for student engagement during introductory programming lectures. iCAP is a web-based solution, allowing student access from any laptop, tablet, or phone with access to a modern browser. Therefore, iCAP increases the flexibility and convenience of educator-student interaction. This poster introduces iCAP's functionalities and compares student mastery of learning outcomes among students taught with and without question-based learning methodology.

Ordinal regression has been widely used in applications, such as credit portfolio management, recommendation systems, and ecology, where the core task is to predict the labels on ordinal scales. Due to its learning efficiency, online ordinal regression using passive aggressive (PA) algorithms has gained a much attention for solving large-scale ranking problems. However, the PA method is sensitive to noise especially in the scenario of streaming data, where the ranking of data samples may change dramatically. In this paper, we propose a noise-resilient online learning algorithm using the Ramp loss function, called PA-RAMP, to improve the performance of PA method for noisy data streams. Also, we validate the order preservation of thresholds of the proposed algorithm. Experiments on real-world data sets demonstrate that the proposed noise-resilient online ordinal regression algorithm is more robust and efficient than state-of-the-art online ordinal regression algorithms.

Post-database searching is a key procedure for peptide spectrum matches (PSMs) in protein identification with mass spectrometry-based strategies. Although many machine learning-based approaches have been developed to improve the accuracy of peptide identification, the challenge remains for improvement due to the poor quality of data samples. CRanker has shown its effectiveness and efficiency in terms of the number of identified PSMs compared with benchmark algorithms. However, it has two weaknesses: overfitting and instability on small-sized datasets. In this paper, we incorporate two new strategies into CRanker to tackle its weaknesses. First of all, we modify the CRanker model by using different weight parameters for the learning losses of decoy and target PSMs. Moreover, we employ self-paced learning in training process to help the classifier getting avoid of those incorrect PSMs. Experimental studies show the modified CRanker with new strategies is more stable than the original one and outperforms benchmark methods in terms of the number of identified PSMs at the same false discovery rates (FDRs).

SEQUEST is a database-searching engine, which calculates the correlation score between observed spectrum and theoretical spectrum deduced from protein sequences stored in a flat text file, even though it is not a relational and object-oriental repository. Nevertheless, the SEQUEST score functions fail to discriminate between true and false PSMs accurately. Some approaches, such as PeptideProphet and Percolator, have been proposed to address the task of distinguishing true and false PSMs. However, most of these methods employ time-consuming learning algorithms to validate peptide assignments [1] . In this paper, we propose a fast algorithm for validating peptide identification by incorporating heterogeneous information from SEQUEST scores and peptide digested knowledge. To automate the peptide identification process and incorporate additional information, we employ l2 multiple kernel learning (MKL) to implement the current peptide identification task. Results on experimental datasets indicate that compared with state-of-the-art methods, i.e., PeptideProphet and Percolator, our data fusing strategy has comparable performance but reduces the running time significantly.

Peptide sequence assignment is the central task in protein identification with MS/MS-based strategies. Sequence database searching routinely generate a large number of peptide spectrum matches (PSMs). Due to either the poor quality of the experimental MS/MS data or unexpected amino acid modifications, there are a large number of incorrect target PSMs. CRanker has shown its efficiency and accuracy in discrimination between correct and incorrect PSMs. However, it costs CRanker too much time on large PSM datasets as a built-in matlab optimization solver needs to be called for training the model. In this work, we exploit the bi-convex structure of the CRanker model and develop an alternate convex search (ACS) algorithm to reduce its total running time. At each iteration, ACS alternately solves one part of the problem when the other part of variables fixed. Compared with Matlab optimization tools, ACS is one order of magnitude faster on most datasets.

BACKGROUND: Identifying gene regulatory network (GRN) from time course gene expression data has attracted more and more attentions. Due to the computational complexity, most approaches for GRN reconstruction are limited on a small number of genes and low connectivity of the underlying networks. These approaches can only identify a single network for a given set of genes. However, for a large-scale gene network, there might exist multiple potential sub-networks, in which genes are only functionally related to others in the sub-networks. RESULTS: We propose the network and community identification (NCI) method for identifying multiple subnetworks from gene expression data by incorporating community structure information into GRN inference. The proposed algorithm iteratively solves two optimization problems, and can promisingly be applied to large-scale GRNs. Furthermore, we present the efficient Block PCA method for searching communities in GRNs. CONCLUSIONS: The NCI method is effective in identifying multiple subnetworks in a large-scale GRN. With the splitting algorithm, the Block PCA method shows a promosing attempt for exploring communities in a large-scale GRN.

This paper constructs the framework of the reproducing kernel Hilbert space for multiple kernel learning, which provides clear insights into the reason that multiple kernel support vector machines (SVM) outperform single kernel SVM. These results can serve as a fundamental guide to account for the superiority of multiple kernel to single kernel learning. Subsequently, the constructed multiple kernel learning algorithms are applied to model a nonlinear blast furnace system only based on its input-output signals. The experimental results not only confirm the superiority of multiple kernel learning algorithms, but also indicate that multiple kernel SVM is a kind of highly competitive data-driven modeling method for the blast furnace system and can provide reliable indication for blast furnace operators to take control actions.

Identifying gene regulatory network (GRN) from time course gene expression data has attracted more and more attentions. Due to the computational complexity, most approaches for GRN reconstruction are limited on a small number of genes and low connectivity of the underlying networks. These approaches can only identify a single network for a given set of genes. However, for a large-scale gene network, there might exist multiple potential sub-networks, in which genes are only functionally related to others in the sub-networks. In this paper, we propose an efficient algorithm for identifying multiple sub-networks from gene expression data by incorporating community structure information into GRN inference. The proposed algorithm iteratively solves two optimization problems, and thus promisingly applies to large- scale GRNs. Experimental studies on synthetic datasets validate the effectiveness of the proposed algorithm in the inference of sub-networks.

Availability is an important issue in distributed systems providing quality of services. Data or service availability modeling is very important to ensure such quality of services. Many factors, such as node availability and network link availability, could affect the data or service availability in a distributed system, and replication schemes have been widely used to improve data or service availability. In this paper, we consider node and network link failures, as well as the impact of multiple replicas on data availability. Efficient availability computing algorithms are developed to model data availability for a system with different topologies. Finally, experimental studies have been conducted to see how well our modeling approach is for the availability modeling of distributed environment with data replication.

In this paper, we propose a binocular framework for face liveness verification under unconstrained localization. The proposed framework contains two components: the first component localizes imbalanced points in face regions of an input pair of stereo images and the second component detects whether an imaging face is a 2D object or a 3D object. We test the propose framework on a publicly available stereo face database, which demonstrated its potential.

In a large-scale collaborative filtering system, pair wise similarity between users is usually measured by users' ratings on the whole set of items. However, this measurement may not be well defined due to the sparsity problem, i.e., the lack of adequate ratings on items for calculating accurate predictions. In fact, most correlated users have similar ratings only on a subset of items. In this paper, we consider a kernel-based classification approach for collaborative filtering and propose several kernel matrix construction methods by using biclusters to capture pair wise similarity between users. In order to characterize accurate correlation among users, we embed both local information and global information into the similarity matrix. However, this similarity matrix may not be a kernel matrix. Our solution is to approximate it with the matrix close to it and use low rank constraints to control the complexity of the matrix.

Imbalanced points are image points whose first-order intensity can be clustered into two imbalanced classes. An important property of imbalanced points is that they can be contiguous to each other. The property helps improve the localization accuracy of imbalanced points across imaging variations. Based on this local geometric coherency property, we propose a global-to-local scheme for imbalanced point matching. The proposed matching scheme first builds correspondence between components of coherent imbalanced points and then refines point correspondence within corresponding components. We test the global-to-local matching scheme, compared with several other well-known methods, on a set of groundtruth stereo images. Furthermore, we present a case study of the proposed scheme in face liveness detection. Our results show the promise of the global-to-local matching scheme.

Edge server replication is an effective solution to achieve high performance in dynamic web applications, such as Web services. Many Web services involve frequent accesses to large-scale backend databases. Current database replication techniques are not directly applicable to edge server architectures. There is no algorithm to dynamically and automatically select the tables for replication. Also, most of the solutions do not consider the potentially limited disk sizes at the edge servers and the needs for them to serve many application sites. In this paper, we present a novel weighted table graph based database replication approach for edge servers to address these problems. Every step in our approach is based on quantitative computation, so it can generate an accurate result. Experimental studies show that our database replication approach significantly improves the performance of Web systems in terms of client response latency, Web application server off loading, and network bandwidth saving.

Imbalance oriented selection scheme was recently proposed to select good candidates of interest points [1]. In this paper, we propose a method to quantify the local diversity of imbalance of an image point, which provides us a new interest strength assignment scheme. We test the proposed approach by repeatability evaluation and stereo matching and obtain promising results.

The quality of large-scale recommendation systems has been insufficient in terms of the accuracy of prediction. One of the major reasons is caused by the sparsity of the samples, usually represented by vectors of userspsila ratings on a set of items. Combining information other than userspsila ratings can provide the learning model complementary views of the data and, thus, a more accurate prediction. In this paper, we propose efficient methods for finding the best combination weights among single kernels. The weight parameters are optimized by aligning the combination kernel to ideal kernels. We solve the kernel alignment problem by linear programming techniques.

Edge server replication is an effective solution to achieve high performance in dynamic Web applications, such as Web services. It can replicate the application logic and the data close to the clients to reduce the response latency and network traffic and to offload the servers. Many Web services involve frequent accesses to large-scale backend databases. Current database replication techniques are not directly applicable to edge server architectures. Most of the solutions do not consider the potentially limited disk sizes at the edge servers and the needs for them to serve many application sites. In this paper, we present a novel database replication approach, PSO-MP, for edge servers to address the problems. Experimental studies show that the PSO-MP approach significantly improves the performance of Web systems in terms of client response latency, Web application server offloading, and network bandwidth saving.

It has been shown that the cost parameters and kernel parameters are critical in the performance of support vector machines (SVMs). A standard parameter selection method compares parameters among a discrete set of values, called the candidate set, and picks the one which has the best classification accuracy. As a result, the choice of parameters strongly depends on the pre-defined candidate set. In this paper, we formulate the selection of the cost parameter and kernel parameter as a two-level optimization problem, in which the values of parameters vary continuously and thus optimization techniques can be applied to select ideal parameters. Due to the non-smoothness of the objective function in our model, a genetic algorithm has been presented. Numerical results show that the two-level approach can significantly improve the performance of SVM classifier in terms of classification accuracy.

Conventional admission control models incur some performance penalty. First, admission control computation can overload a server that is already heavily loaded. Also, in large-scale media systems with geographically distributed server clusters, performing admission control on each cluster can result in long response latency, if the client request is denied at one site and has to be forwarded to another site. Furthermore, in prefix caching, initial frames cached at the proxy are delivered to the client before the admission decisions are made. If the media server is heavily loaded and, finally, has to deny the client request, forwarding a large number of initial frames is a waste of critical network resources. In this paper, a novel distributed admission control model is presented. We make use of proxy servers to perform the admission control tasks. Each proxy hosts an agent to coordinate the effort. Agents reserve media server's disk bandwidth and make admission decisions autonomously based on the allocated disk bandwidth. We develop an effective game theoretic framework to achieve fairness in the bandwidth allocation among the agents. To improve the overall bandwidth utilization, we also consider an aggressive admission control policy where each agent may admit more requests than its allocated bandwidth allows. The distributed admission control approach provides the solution to the stated problems incurred in conventional admission control models. Experimental studies show that our algorithms significantly reduce the response latency and the media server load.

In this paper, we propose a comprehensive request processing approach at the proxy server to preprocess media requests with the goal of quality of service (QoS) enhancement. This approach integrates distributed admission control, batching and prefix caching schemes. A proxy assistant is designed to realize the proxy site request processing approach. Also, to fully utilize prefix caching, we allow a request to stay at the proxy to wait for the disk bandwidth of the media server to become available before the cached contents are exhausted. Performance studies show that our request processing approach significantly improves the admission rates, and response time

Summary form only given. Application level multicast approaches have attracted interests in both research and industrial communities. Among various approaches, constructing the overlay multicast tree based on mesh network is very promising for a large multicast group size. This approach is efficient in bandwidth utilization but suffer from high computation overhead. We present a new approach for constructing an overlay multicast tree in a large scale media streaming system using neural networks. The media server first constructs an overlay multicast tree for requests arriving in a relatively short time interval. Then the subsequent requests can join the multicast tree in a distributed way. To utilize the interface bandwidth efficiently, we formulate the original tree construction problem as a balanced multicast tree problem and propose a new SOFM-like neural network algorithm to obtain the solution. For the requests arriving at a later time, the proxy servers process the requests and join the multicast tree using a decentralized prediction based on a multilayered neural network. The tree is kept balanced in the tree joining process by having proxy servers predict the traffic load on its neighbors.

We present a novel distributed admission control model for multimedia delivery. Proxy servers form a large-scale distributed infrastructure and can be used to perform the admission control tasks. In our distributed admission control algorithm, an agent first characterizes its disk bandwidth requirement based on the distribution of requests arrived at the proxies. Then, it reserves the disk bandwidth from the server by maximizing its utility function. The server grants the bandwidth to the agents according to its capacity. Based on the granted bandwidth, each agent executes the admission control algorithm for its local requests. The distributed admission control model provides the admission control service close to the users and, hence, reduces the network load, media server load, and, in some cases, users' response latency.

Prefix caching has significant benefits in media delivery systems in terms of reduced startup latency and network traffic. However, with the conventional admission control model, prefix caching may incur some performance penalty. When the media server is heavily loaded and has to deny the client request finally, forwarding a large number of initial frames from proxy to clients can result in dramatic waste of critical network resource. An ideal solution to this problem is to let proxy servers perform the admission control on behalf of the media servers. In this paper, we present a distributed admission control policy. We consider an aggressive disk bandwidth allocation scheme where an agent may, with a certain probability, admit a new request when the current bandwidth used at the agent has reached its limit. We formulate the problem as a min-max optimization problem and provide a MINMAX algorithm to find the solutions. The effectiveness of our algorithm is demonstrated through simulation studies.