M Tech Dissertations
Permanent URI for this collectionhttp://drsr.daiict.ac.in/handle/123456789/3
Browse
17 results
Search Results
Item Open Access Public toilet hygiene management system(Dhirubhai Ambani Institute of Information and Communication Technology, 2018) Chandra, Shubham; Srivastava, Sanjay; Roy, Anil K.Hygiene in public toilets require accurate monitoring of the gases produced as byproducts. Even if the complaint is made, the chances of getting it cleaned is very low so the problem remained unsolved. Although smart urinal pots are available in the market but deploying it in public toilet is not an affordable solution. When the toilet is in use for long run without cleaning, intensity of odor increases to unhygienic level. We propose a low cost sensor node which can be used to monitor and communicate the hygiene data to the servers. We also propose an application to collect user feedback along with the sensor readings. An algorithm which can combine both these inputs into a community benchmark hygiene rating. Database also handles the reporting and follow-ups by the civic authorities. To eliminate the calibration process of the sensor, we are proposing a system where the sensor node will learn from user feedbacks given over time. Based on the feedbacks received, we remove user bias by normalizing user input in overlapping time-stamp data. Once the user input is normalized, it will be fed as an input to calculate Odor Bandwidth Spectrum whereas output will be detailed classification of hygiene level based on the intensity of odor produced.Item Open Access Job recommendation system(Dhirubhai Ambani Institute of Information and Communication Technology, 2018) Thakur, Palak; Majumder, PrasenjitRecommender systems are being used in all types of web based services. As the amount of data available in each service is huge it is not possible to search all the relevant and useful information. So for this purpose we need a recommender system, as it will take into consideration the user preferences for recommending the most relevant items. Many different recommendation algorithms have been used for this purpose. Collaborative filtering, matrix factorization techniques, content based methods have been extensively researched upon. Also significant amount of work has been done by using deep learning approaches in the existing recommendation frameworks. However these methods have suffered from the problem of overspecialization, also these approaches have not been very successful in the cold start scenario. The methods also do not effectively capture the change in user preferences. Recent developments in the field of deep reinforcement learning in games have shown to outperform the human players. This technique has largely remained unexplored in the field of recommendation. Reinforcement learning can very effectively handle the cold start issue as it is supposed to work in unknown state as well. Also as it is continuously learning from its interaction with the environment, so it can be used for capturing the changing user preferences. Based on which we can handle the recommendations which is personalized as per the user preferences. In this work, we describe a reinforcement learning based job recommendation system in which we model the recommendation problem as a Markov Decision Process(MDP) which is a 5- tuple (S, A, P, R, g) including states, actions, reward, a transition probability, and a discount factor. Reinforcement Learning is concerned with how the recommender agent takes actions in an environment so as to maximize the cumulative reward. We have also implemented, the collaborative filtering approach proposed in [7], the baseline approach suggested in ACM Recommender Systems Challenge 2017 [1] and the content-based approach proposed in [13]. We have finally shown that reinforcement learning method outperforms the other three.Item Open Access Personalized gait abnormality detection system(Dhirubhai Ambani Institute of Information and Communication Technology, 2018) Dhokai, Ronak; Sasidhar, KalyanGait refers to walking manner of a person, and it is also an indication of neurologicalhealth status of a person. Gait variability can occur due to factors like aging,injuries and diseases. If not notified or diagnosed at an early stage, this variabilityof gait could lead to lifetime abnormality.In this work, we have proposed a smartphone based solution for the task ofcapturing gait and performing abnormality detection on the sensed data. Usingthe built-in accelerometer, we collected walking data from 10 different users,which consisted of both normal and minor abnormalities. Features such as stridetime and stride length were extracted and the sudden changes in the walk weredetected by calculating the extent of deviation of these features between the walkdata. Individual user based threshold value of deviation was estimated and thedetection algorithm performance was evaluated for each of the 10 users.Item Open Access Tensor Product and Acyclic Edge Colouring(Dhirubhai Ambani Institute of Information and Communication Technology, 2017) Shihora, Rutvi; Muthu, Rahul"The assignment of colours to the edges of graph G such that no two adjacent edges get the same colour and there is no 2-coloured cycle in G is known as Acyclic Edge Colouring. The minimum number of colours needed to acyclically edge colour the graph is known as acyclic chromatic index and is denoted by a0(G). The difference between a0(G) and 4(G) is known as the gap of the graph. Determining the value of a0(G) either algorithmically or theoretically has been a very difficult problem. It belongs to the class of NP-complete problems. The value of a0(G) has not yet been determined even for the highly structured class of graphs like complete graphs.Determining the exact values of a0(G) even for very special classes of graphs is still open. Tensor Product, also known as Kronecker Product is a special type of graph product. Any graph that can be represented as a tensor product has the same number of irreducible factors, even though the factor graphs may be different for different factorizations. There exists an algorithm that recognizes tensor product graphs in polynomial time and finds a factorization for the same. This thesis addresses the problem of finding a0(G) for a graph G which is a tensor product of either K2, paths or cycles. We have provided optimal and sub-optimal colouring techniques for colouring the tensor product graph, whose factors are either gap 0 or gap 1 graphs. This thesis focuses on the tensor product graphs whose factors are specific gap 0 and gap 1 graphs which are edges, paths or cycles. This can later be extended to arbitrary gap 0 and gap 1 graphs."Item Open Access Integrate active and meta-cognitive learning with extreme learning machine(Dhirubhai Ambani Institute of Information and Communication Technology, 2017) Dave, Keval Narayanbhai; Joshi, Manjunath V."The extreme learning machine has become very popular due to its fast training and good generalization performance. The two considerations for extreme learning machine are fixed network structure and redundancy in training data. In this thesis, we propose efficient semi-supervised learning algorithm to integrate active and metacognitive learning with extreme learning machine to arrive at non-redundant data and optimal network structure for better classification. Semi-supervised learning makes use of unlabeled data along with small labeled data for training. Active learning a special case of semi-supervised learning is applied to select non-redundant data and to reduce the labeling costs. In active learning, the algorithm can iteratively query the user for label of new data points. Metacognitive learning proposes the addition of neuron while training which is used to achieve optimal network structure. Also, the use of regularization has shown improvement in accuracy. The proposed approach is evaluated for classification on various benchmark datasets from UCI machine learning repository. This datasets has features extracted for various real world problems.The proposed method is compared with three state-of-the-art methods based on accuracy, training time and amount of data required for training. Performance evaluation shows that proposed approach gives similar or better accuracy than other approaches with reduced training time."Item Open Access Algorithms For Computing Prime Implicates Using Distributivity And Labelled Resolution In Modal Logic(Dhirubhai Ambani Institute of Information and Communication Technology, 2017) Agarwal, Rishabh; Raut, Manoj Kumar"Modal Logic is used for Knowledge Representation and Reasoning in many problems. Suppose we have a knowledge Base(KB) and a Query (Q), the question here is whether KB j= Q or not. Knowledge Compilation is one of the techniques that have been proposed to deal with the computation intractability of such query answering problems. Knowledge Compilation (KC) techniques have been proposed successfully in Modal logic to overcome such logical entailment problem. So KC is split into two phases such as on-line and off-line. The KB is preprocessed in offline phase into another Knowledge Base KB0,where KB0 contains the set of prime implicates and queries are answered from KB0 in polynomial time. In this thesis, we have implemented the prime implicate computation algorithm [3] in modal logic. We have also suggested an algorithm to compute prime implicates using labelled resolution in modal logic and proved its correctness."Item Open Access Study of Total Graph Using Dynamic Graph Operation(Dhirubhai Ambani Institute of Information and Communication Technology, 2017) Patel, Dhaval; Muthu, Rahul"Auxiliary graphs are graphs which are used to translate one graph problem to another graph problem. Total graph is also an auxiliary graph which is used to translate the total coloring problem into the vertex coloring problem. Dynamic graph operation on a graph is defined as sequence of update operation on the graph, where update operations are add vertex/edge and remove edge/vertex. This thesis derives some new characteristics of total graph. By using existing properties and new properties, we have developed algorithms to find maximal total sub graph of an arbitrary nontotal graph. We have devised a procedure by using which we are able tomigrate set of edges form vertex-vertex part to edgevertex part, together with some allied operations such that the resulting graph is also a total graph. For super total graph we have provided some characteristics which is required to add a vertex/edge."Item Open Access Research Article Recommendation System Using LDA Topic Modeling(Dhirubhai Ambani Institute of Information and Communication Technology, 2017) Tyagi, Saurabh; Dasgupta, Sourish"A personalised recommender system for research articles suggest research papers according to the user preferences. Many times user wants the suggestions according to his/her own interest. In this research article recommender system user gives the preferences of his/her domain and the system improves as he/she select the research articles. The system uses wikipedia documents to train basic models for twelve different domains. The topic modeling technique is used to define the topic of a given article. The wikipedia documents generate the trained model using Latent Dirichlet allocation (LDA), eighty topics for each domain. After generating these LDA models for all domains, system assigns a topic to the each significant word in every research article in the research article index, called a topic sequence. The topic sequencer generates the topic sequence for the top three most favourable domains, using domain selector. Domain selector use the probability of the specific word in LDA and select the topic of the highest probability for that word in the a specific domain."Item Open Access Characterization of Realization Graphs(Dhirubhai Ambani Institute of Information and Communication Technology, 2017) Seksaria, Khushboo; Muthu, Rahul"Given a degree sequence d, usually it generates several different realizations belonging to several distinct isomorphism classes. In general, the structure of an arbitrary realization cannot be known from its degree sequence, so it may not be possible to determine whether vertices with specified degrees will be adjacent (or non-adjacent), but in some cases it might be possible. The degree sequence is one of the simplest parameters associated with a graph, however, due to the fact that same degree sequence can belong to several pairwise nonisomorphic graphs (different realizations of a given degree sequence) the utility of a degree sequence in a graph problem is limited. Studying the structure of these different realizations of a degree sequence and then, to understand the relationships that exist among graphs with the same degree sequence will be our main focus, which can be very well studied with the help of realization graphs. Graphs constructed by applying some specific rule to any given graph are called auxiliary graphs. Such graphs are often used to translate one graph problem into another. A Realization graph, G(d) is an auxiliary graph defined as the graph (R, d) where R is the set of realizations of d and d is the set of edges where two vertices G, G0 of R are adjacent if and only if performing a single 2-switch can transform G into G0 [3] This thesis consists of several problems related to different realizations of degree sequences. In this thesis, we have looked at various graph families and studied whether they can be characterized by their degree sequences. Also, we have derived some new characteristics for realization graphs. We have also derived a method for verification of whether or not a given degree sequence is uniquely realizable."Item Open Access Abnormal Gait Detection using Smartphone(Dhirubhai Ambani Institute of Information and Communication Technology, 2017) Satyajeet, Satyam; Sasidhar, Kalyan"Gait cycle is repetitive walking pattern involving steps and strides. Difference between abnormal gait and normal gait lies between gait parameters and both are compared for prediction. We are proposing a method which is cheap and using only Smartphone embedded accelerometer to extract gait parameters. The advantages are low cost and low power supply requirements with everyone having Smartphone making it user friendly. We collected data for normal and abnormal patients having various kinds of diseases. Problems such as Rheumatoid Arthritis (RA), Osteoarthritis (OA), sciatica, calcaneal spur (or heel spur), Ankylosing spondylitis, Motor Injury, polio and Rotation of knee. The classifiers used were Naives Bayes (NB), Decision Tree (DT) and Random Forest (RF) out of which RF performed best giving 91.52% accuracy on 10-fold cross validation Set. DT and NB were giving accuracy of 86.38% and 89.69%."