Online Course Management System

ABSTRACT:

We study the problem of online multitask learning for solving multiple related classification tasks in parallel, aiming atclassifying every sequence of data received by each task accurately and efficiently. One practical example of online multitask learningis the micro-blog sentiment detection on a group of users, which classifies micro-blog posts generated by each user into emotional ornon-emotional categories. This particular online learning task is challenging for a number of reasons. First of all, to meet the criticalrequirements of online applications, a highly efficient and scalable classification solution that can make immediate predictions withlow learning cost is needed. This requirement leaves conventional batch learning algorithms out of consideration. Second, classicalclassification methods, be it batch or online, often encounter a dilemma when applied to a group of tasks, i.e., on one hand, a singleclassification model trained on the entire collection of data from all tasks may fail to capture characteristics of individual task; on theother hand, a model trained independently on individual tasks may suffer from insufficient training data. To overcome thesechallenges, in this paper, we propose a collaborative online multitask learning method, which learns a global model over the entiredata of all tasks. At the same time, individual models for multiple related tasks are jointly inferred by leveraging the global modelthrough a collaborative online learning approach. We illustrate the efficacy of the proposed technique on a synthetic dataset. We alsoevaluate it on three real-life problems—spam email filtering, bioinformatics data classification, and micro-blog sentiment detection.

Experimental results show that our method is effective and scalable at the online classification of multiple related tasks.

PROJECT OUTPUT VIDEO:

EXISTING SYSTEM:

Unlike batch learning methods, which assume all trainingsamples to be available before the learning task begins;online learning algorithms incrementally build a modelfrom a stream of samples, allowing for simple and fastmodel updates. They are thus naturally capable of dealing with large datasets and problems whose data arrivesequentially. The origin of online learning can date back tothe well-known Perceptron algorithm, which updatesthe model weight by moving it a tad closer to each misclassified sample. Descendant Perceptron-like methods employmore sophisticated update strategies.

DISADVANTAGES OF EXISTING SYSTEM:

• The classical multi-task learning approach focuses on learning the primary task without caring how the other tasks are learned.
• The classical multitask learning problem is often studied in a batch learning setting, which assumes that the training data of all tasks are available.
• On one hand, this assumption is not realistic for many real-world problems where data arrives sequentially. On the other hand, the batch multitask learning algorithms usually have fairly intensive training cost and poor scalability performance, as far as large real applications are concerned

PROPOSED SYSTEM:

In this paper, we investigate the problem of online multitask learning, which differs from the classical multitasklearning in two aspects. First, our goal is to improve thelearning performance of all tasks instead of focusing on asingle primary task. Second, we frame the multitask learning problem in an online learning setting by assuming thatthe data for each task arrives sequentially, which is a morerealistic scenario for real-world applications. Unlike batchlearning techniques, online learning methods learn over asequence of data by processing each sample upon arrival.At each round, the learner first receives one instance,makes a prediction, and receives the true label. The errorinformation is then used to update the learning model.

ADVANTAGES OF PROPOSED SYSTEM:

The basic idea is to first build a genericglobalmodelfrom large amount of data gathered from all users, and thensubsequently leverage the global model to build the personalizedclassification models for individual users through acollaborative learning process. We formulate this idea intoan optimization problem under an online learning setting,and propose two different COML algorithms by exploringdifferent kinds of online learning methodologies.

(1) A single taskonline learning approach that simply learns a global modelover the entire collection of data gathered from all thetasks, (2) a single task online learning approach that solveseach task independently, and (3) a state-of-the-art onlinemultitask learning approach.

MODULES:

1. Two fold motivation
2. Formulation
3. Confidence Weighted Learning

MODULES DESCSRIPTION:

1. Two fold motivation:

First, as tasksoften exhibit varying patterns, it is neither practical noreffective to learn a single global model for classification.Second, it is also not always possible to learn a good classification model for each task since training data availablefor a single task are often limited. For such case, it is reasonable to pool together data across many related tasks.Hence, a better solution is to combine these two approachescoherently.Specifically,the collaborative online multitask learning operates in asequential manner. At each learning round, it collects thecurrent global set of data; one from each of the engagedusers/tasks, which are employed to update the global classification model. At the same time, a collaborative personalized model is maintained for each user/task. The individualcollaborative classification model is subsequently updatedusing the latest individual data and the global modelparameters. Therefore, our approach can leverage globalknowledge for classification, while adapting to individualnuances via the collaborative learning way.

2. Formulation:

Online multitask classification proceeds in rounds byobserving a sequence of examples, each belonging to someuser/task from set ofKusers/tasks. On each round, thereareKseparate online binary classification problems beingsolved jointly. We assume that data from all users/taskscan be represented in the same global feature space, so thatit is possible to use the shared information between tasksto enhance each learning task. The first step of the collaborative online multitask learningbuilds a global classification model to exploit the commonality among tasks. We adopt the online passive aggressive(PA) framework to build a global model using datacollected fromallusersatroundt.The critical step of our collaborative online multitask learning is to apply the existing global model to collaborativelylearn the each of the Kindividual user models. Using thesame PA formulation, the goal is to learn a classificationmodel for thek^thuser. The next step is to use the shared information learnedby the global model to enhance each individual learningmodel. We formulate the collaborative learning model as aconvex optimization problem that minimizes the deviationof the new weight vector from the prior collaborative oneand the global one.

3. Confidence Weighted Learning:

A current trend in online learning research is to useparameter confidence information to guide online learning process. Confidence weighted learning, proposed byCrammeret al., models the linear classifierhypotheses uncertainty with a multivariate Gaussian distribution over weight vectors, which is then used to controlthe direction and scale of parameter updates. Conceptually,to classify an instance x, a confidence weighted classifierdraws a parameter vectorw∼N(μ,∑) and predicts thelabel according to sign(w·x). In practice, however, the aver-age weight vectorE(w)=μis used to make the prediction.Confidence weighted learning algorithms have been shownto perform well on many tasks. In this section, we extendthe proposed collaborative online multitask learning withthe confidence weighted hypothesis.

SYSTEM REQUIREMENTS:

HARDWARE REQUIREMENTS:

• System           : Pentium IV 2.4 GHz.
• Hard Disk      : 40 GB.
• Floppy Drive : 44 Mb.
• Monitor          : 15 VGA Colour.
• Mouse             : Logitech
• Ram                 : 512 Mb.

SOFTWARE REQUIREMENTS:

• Operating system : Windows XP/7.
• Coding Language : net, C#.net
• Tool                         : Visual Studio 2010
• Database                : SQL SERVER 2008