No description available.

Weight-sharing property in convolutional neural network (CNN) is useful in reducing number of parameters in the network and also introduces regularization effect which helps to gain high performance. Non-weight-shared convolutional neural networks also known as Locally connected networks (LCNs) has potential to learn more<br>in each layer due to large number of parameters without increasing number of inference computations as compared to CNNs. This work explores the idea of where Locally connected layers can be used to gain performance benefits in terms of accuracy and computations, what are the challenges in training the locally connected networks and what are the techniques that should be introduced in order to train this network and achieve high performance. Partially-local connected network (P-LCN) VGG-16 which is hybrid of convolutional layers and Locally connected layers achieves on average 2.0% accuracy gain over VGG-16 full convolutional network on CIFAR100 and 0.32% on CIFAR10. Modified implementation of batch normalization for Full LCNs (all layers in network are locally connected layers) gives improvement of 50% in training accuracy as compared to using CNN batch normalization layer in full LCN. Since L1, L2 and Dropout regularization does not help improve accuracy of LCNs, regularization methods which focuses on kernels rather than individual weight for regularizing the network were explored. Ladder networks with semi supervised learning achieves this goal. Training methodology of ladder networks was modified to achieve ∼2% accuracy improvement on Pavia-University hyper-spectral image dataset with 5 labels per class.

Wireless sensor and actor networks are used in situations where interaction is required between a network and the environment in which the network is deployed. This research studies the functioning of a single mobile actor deployed in a sparsely connected network. When deployed in a sparsely connected network, an actor has to do more than acting. It has to perform the additional duties of an event collector - collecting events from the naturally occurring clusters - so that it can fulfill its primary obligation as an actor. The path taken by a mobile actor node is generated by a mobility model. The existing random mobility models are non-intelligent mobility models. While they may bring about a chance meeting between an actor and an event, there is no guarantee that these meetings will actually happen. This motivates the development of intelligent mobility models for the actor node, which will generate paths that are reflective of the network in which the actor is deployed. In this thesis, intelligent mobility models for the actor node were developed using the inherent clustering information of a sparsely connected network. These models were applied to an actor node in networks of varying sparseness and the following conclusions were reached: (i) Existing random mobility models are unsuitable for an actor in a sparsely connected network. (ii) High probability of events can be sensed when a sparsely connected network is used. (iii) 100% event detection by the actor node is possible at higher speeds. (iv) When the single actor functioned both as an event collector and as an actor, the number of events acted upon by the actor was very close to the number of events acted upon by an actor in a fully connected network. (v) The Correlation Theory developed in this research suggests using a combination of the intelligent mobility models to obtain the best performance results under all circumstances. (vi) Early detection of events can be supported where it is required. All of the above conclusions justify the deployment of a single actor and a sparsely connected network, either individually or as a combination.