Segmentation of femur in QCT image is always challenging due to its complex ball and socket joint with acetabulum. Recently, deep learning based techniques have been used in image segmentation. However, the success of these methods depends mainly on the accurate annotations which is costly, needs expert opinion, or manual intervention. To overcome these challenges, we propose a semi-supervised learning based approach where we have developed a U-Net based framework for femur segmentation from sparsely annotated Quantitative Computed Tomography (QCT) slices. We annotate only the part of QCT slices at the proximal end of the femur joint. We then integrate the original metadata using instance numbers from QCT files. This semi supervised approach facilitates to work with DICOM medical image data with less annotation. Our framework is cost effective since it saves cost for manual intervention and annotation by medical experts. We have used performance metric Dice Similarity Coefficient (DSC) and found that, we have achieved DSC of 93.7% for unseen patients, and DSC of 99.2% for patients in validation stage which are promising results.

In a Named Data Network (NDN), contents are cached in-network nodes to satisfy interest requested by con-sumers quickly. Hence, the caching policy is important for the efficient use of cache queues and content delivery and so cache replacement strategy because of limited memory in the content stores (CS). In our research, we propose a cache eviction policy with the goal of making the best usage of the memory queues of NDN nodes.

Voice based interactive system has numerous ap-plications including patient care system, robotics, interactive learning tool etc. Speech Emotion Recognition (SER) is a vital part of any voice based interactive system. Providing an efficient SER framework in multi-lingual domain is highly challenging due to the difficulties in feature extraction from noisy voice signals, language barrier, issues due to gender dependency, domain generalization problem etc. Therefore, all of these challenges have made multi-domain SER interesting to the researchers. In our study, we provide a multi-domain SER framework where popular benchmark corpora have been integrated and used together for training and testing with the goal of removing language barriers and the corpus dependency. Moreover, we have utilized the role of gender on acoustic signal features to improve the performance in multi-domain. We design a hierarchical Convolutional Neural Network (CNN) based framework that finds the influence of genders while recognizing emotions in multi-domain cross-corpus system. We have used Unweighted Average Recall (UAR) for measuring performance in the multi-domain corpus to address data imbalance problem. We validate our proposed framework by conducting extensive experiments with benchmark datasets. The results show that using the proposed gender-based SER model with multi-lingual cross-corpus performs better than the conventional SER models. Our novel multi-domain cross-corpus SER will be very helpful for designing different multi-lingual voice- based interactive applications.

Human-Machine interaction through audio or speech is always challenging because of the difficulties in emotion detection from audio data. Emotion is a natural way to express the individual's mental state. When a person is in an unusual mental state, the voice changes accordingly in his or her subconsciousness. Emotion detection from continuous speech with multi-lingual platform is more difficult due to complicated feature extraction and feature modeling process. Voice features change due to physical, background and channel noise or sue to other articulatory obstacles. In spite of all these uncertainties, researchers find that among glottal wave forms (consists of glottal excitation and vocal tract filters), glottal pulses carry basic information of voice because these pulses are out of articulatory effects or inner-mouth changes. In this research, we capture real-time speech signal, separate the glottal pulses and buffer the signal into specified frames to identify the phonemes those are affected mostly in emotionally unstable conditions. In these frames, we identify the basic voice controlling features like pitch, intensity, jitter to get these phonemes. For this scheme, we use a real-life continuous dataset which is a multilingual corpus of emotional speech. After feature extraction from the dataset, different classification methodologies have been applied to predict the emotional state and we find our prediction model with around 83% accuracy.

Network Function Virtualization (NFV) is gaining popularity among network operators to provide cost effective and dynamic network services. NFV enables faster service by deploying virtual instances of network functions. While serving dynamic and varying traffic demands, network operators can get benefit from knowing the requirement for the number of Virtual Network Functions (VNFs), ahead of time. VNF requirement prediction method mostly depends on the fluctuation of network traffic load. Predicting the required number of VNFs helps the operator to manage network resources in better ways. VNF prediction method is being considered as an interesting research field for researchers. In our research, we propose VNF requirement prediction methods based on Deep Neural Networks (DNN) and Long Short Term Memory (LSTM) Networks. We provide experimental results which show promising accuracy and improvement of our method, compared to machine learning approaches used before for VNF requirement prediction. Network resource management techniques can be benefited enormously from our higher accuracy based approaches.

Twitter is one of the most popular social media platforms, and a widely used data channel for the propagation of information. Since too many open end users access and use the powerful channel for information propagation, it is becoming increasingly difficult to separate reliable information from the overwhelming pool of information. With the advent of social media generated "fake news" and with their growing influence on the society, the issue of detecting authentic information gains utmost importance. The purpose of our work is to measure the reliability or correctness of the information that is being propagated using Twitter. However, to measure the reliability it is important to preprocess the tweets and to find the similar events. For doing this, an effective clustering method is required which will measure the similarity between the tweets using both semantic and syntactic similarity. We also propose an efficient way to compute the credibility of the sources and how information propagates around the network.

In a Wireless Sensor Network (WSN), it is common to use high end nodes known as relay nodes with more battery power and communication range for forwarding data to the base station. To assist sensor nodes if the relay nodes are added, communication becomes better because relay nodes are used to bridge the communication or routing. However, this is also important to keep the minimum number of relay nodes working to save the network cost. Most of the existing research focuses on placing the relay nodes in wireless sensor networks. It may happen that the different relay nodes forward data of the different sensor nodes on the basis of the availability. When data is not forwarded, relay nodes may be in sleep condition. In this paper, we provide a load balancing strategy to find the minimum number of relay nodes to be working with the goal of having longer network lifetime. We validate our claim by providing simulation results and we find it is possible to save a significant part of duty cycle using our model.

Emotion sensing or detection is nowadays vital research area since it has many applications in mental-health recognition based technology, biometric security analysis, etc. It is a challenging research area because voice features can vary based on gender, physical or mental condition and environmental noise. In our research, we provide a novel framework for emotion detection based on jitter computing. Here, rather than using the entire voice signal, we use short time significant frames, which would be enough to identify the emotional condition of the speaker. This makes our framework less costly. We collect data set from real users and apply our method. We compare our method with other popular methods and we find that our method provides better accuracy, true acceptance rate, less error rate.

Locating a moving target for an Underwater Wireless Sensor Network (UWSN) is a challenging task since traditional radio link based tracking methods can not be directly applied to the under water network. Underwater sensors use acoustic links for communication. Besides under the water, network architecture is also different from the traditional terrestrial WSN architecture. Replacement difficulty of underwater sensors in case of power failure adds more challenges. Therefore, a target localization protocol in UWSNs should have energy saving strategy. In our paper, we incorporate local search based energy saving tracking method for an UWSN. To handle the challenges, we present a meta-heuristic based algorithm by keeping the minimum number of sensors active which ultimately increases the network lifetime. We validate our method by experimental results and find that our algorithm can detect a moving target with less energy consumption. We measure the performance of our method by comparing the target trajectory with the true trajectory. We also compare the energy consumption with that of another frequently used method for this type of network. We find that our method works better.