Parag Kulkarni

Self Organizing Maps to Build Intrusion Detection System

With the rapid expansion of computer usage and computer network the security of the computer system has became very important. Every day new kind of attacks are being faced by industries. Many methods have been proposed for the development of intrusion detection system using artificial intelligence technique. In this paper we will have a look at an algorithm based on neural networks that are suitable for Intrusion Detection Systems (IDS) [1] [2]. The name of this algorithm is ―Self Organizing Maps‖ (SOM). Neural networks method is a promising technique which has been used in many classification problems. The neural network component will implement the neural approach, which is based on the assumption that each user is unique and leaves a unique footprint on a computer system when using it. If a user’s footprint does not match his/her reference footprint based on normal system activities, the system administrator or security officer can be alerted to a possible security breach. At the end of the paper we will figure out the advantages and disadvantages of Self Organizing Maps and explain how it is useful for building an Intrusion Detection System.

A Performance based Routing Classification in Cognitive Radio Networks

Cognitive Radio Networks (CRN) is offering tremendous performance and operational benefit by providing high bandwidth to mobile users via dynamic spectrum access techniques. In this paper, we address the problem of routing in CRN which concerns about identification and maintenance of the optimal path from source to destination through intermediate relay nodes and spectrum on each link using available common channels. In this survey, the characteristics features and limiting factors of existing routing protocols are thoroughly investigated with its performance evaluation criterias. First, the overview of the routing with its unique challenges is given under the restriction of interference and fairness to increase overall network throughput. Next, a detailed classification of the routing strategies is given according to performance evaluation matrices which are considered according to specific demand and requirements of network users. A representative selection of these strategies is discussed in detail in this paper together with services given to unique challenges of CRN. Important issues and future directions are also discussed, while highlighting the need of close coupling between interaction of network users and dynamic decision theories.

Multi-Agent Reinforcement Learning Based Opportunistic Routing and Channel Assignment for Mobile Cognitive Radio Ad Hoc Network

Opportunistic spectrum access using cognitive radio technology enables exploring vacant licensed spectrum bands and thereby improving the spectrum utilization. However, it will have a significant impact on upper layer performance like routing as the reliable knowledge of topology and channel statistics are not available, especially in Mobile Cognitive Radio Ad hoc Network (MCRAN). To address specific requirements of MCRAN, this paper is proposing online opportunistic routing algorithm using multi-agent reinforcement learning. The proposed routing scheme jointly addresses, link and relay selection based on transmission success probabilities. This sophisticated learning mechanism successfully explores opportunities in partially observable and non-stationary environment of MCRAN. Simulation results show the effectiveness of this algorithm.

Dynamic channel selection and routing through reinforcement learning in Cognitive Radio Networks

Recent exploration in Cognitive Radio Network proved itself as emerging paradigm to attempt the underutilization of wireless spectrum. Routing is challenging problem due to intermittent spectrum availability and incomplete knowledge of environment. This paper proposes reinforcement learning based combined framework of channel selection and routing for multi-hop cognitive radio network. Reinforcement learning is generic method for resource utilization in a partially observable and non-stationary environment. In this paper, channel selection and routing is modeled as Markao Decision Process to design the methodology of learning the best resource allocation policies adopted in the process state, based on the feedback received from the environment. First the design of the reward, transition and value function is described which helps in evolving the policy for selecting channel which results in increased spectrum utilization. The routing strategy is described which is exploring different state-action pair to come up with various routing solution which are ranked according to their reinforcement signal. Overhead of rerouting is also minimized by providing backup routes. Agent experiences in the form of reinforcement signal can be used by each cognitive node to further refine the routing strategies.

Emotional Recognition and towards Context based Decision

verbal communication may be used to enhance verbal communication or even provide developers with an alternative for communicating information. Emotion or Gesture recognition is been highlighted in the area of Artificial Intelligence and advanced machine learning. Emotion or gesture is an important feature for an intelligent Human Computer Interaction. This paper basically is a literature survey paper which reveals with the research work already dealt with in this area. Facial expression has been concluded as the most important part involved in it. Even Facial features are also distinguished out of which eyes and mouth is probably more prominent. Neural networks are the widely used. Approaches towards Rough Fuzzy definition can be probably resolve the complexity. Context based recognition can be added so as to resolve the ambiguity involved in different scenarios.

A Novel Approach for Clustering based on Pattern Analysis

Clustering aims at grouping of data into clusters based on the similarity between them. It is the pattern of the data that governs grouping. In this paper, we propose method for clustering that is based on finding closeness between the data series. A novel method referred as Clustering with Closeness factor (CCF) is proposed that works in two phases and is not pre-bound with clusters numbers. The method identifies the pattern of data and performs clustering. With proper selection of threshold value, the approach can prove to be a big step for decision making. General Terms Machine intelligence, machine learning and pattern analysis

An effective content based video analysis and retrieval using pattern indexing techniques

To retrieve videos from database, effective video analyzing, indexing and retrieval techniques are required. Video retrieval using query-by-image is not successful as it gives result of videos with less relevancy and accuracy. A method is proposed where input is a video clip, to achieve the high quality of content based video retrieval by discovering the temporal patterns in the video contents. On the basis of the discovered temporal patterns efficient indexing and sequence matching technique are integrated together to reduce the computation cost and to increase the retrieval accuracy, respectively. The method enriches content based video retrieval in terms of efficiency and effectiveness by using FPI and AFPI search methods in terms of precision and recall.

A combinational approach for activity recognition using context

Human action recognition is a challenging task not only because of the factors like changes in intensity, background, etc but also because of the variability in the behavioural patterns among the objects in the image which in turn affects the recognition accuracy. Analyzing all those factors and identifying the action is termed as activity recognition. In this paper, we present an approach of activity recognition with the help of context. Context can be termed as the relationship between the objects performing the activity. Activity recognition is performed based on motion identification and context information. We use Principle Component Analysis along with the low level features to perform feature extraction and then Support vector machine as a classifier which classifies the action into a class with label. Thus by performing a high level of feature extraction using context and by supervised training, we perform activity recognition and try to improve the recognition accuracy of the system.