Doctor of Philosophy (Ph.D.)

Permanent URI for this collectionhttp://192.248.9.226/handle/123/747

Browse

Now showing 1 - 4 of 4

item: Thesis-Full-text
Data-dependent resource allocation and routing in multi-path neural networks for vision-based learning
(2023) Tissera MHGD; Rodrigo R
As the depth of a neural network increases, the non-linearity and more parameters allow it to learn more complex functions. While network deepening has been proven e_ective, there is still an opportunity for e_cient feature extraction within a layer that will improve the overall performance for the complexity of the network. Widening networks by adding more _lters to each layer is the naive approach towards strengthening layer- wise feature extraction. It is an ine_cient scaling option, considering the number of parameters being quadratic with the number of _lters employed per layer. In contrast, parallel extractors in each layer provide an e_cient scaling option. However, without context-dependent input allocation among these processes, such parallel computations tend to learn similar features, collapsing to a single computation. Thus, it is vital to study the parallel stacking of computations layer-wise and design a routing method that allocates incoming feature maps to these computations. The expected outcome is to group homogeneous feature maps in parallel layers and employ exclusive _lter sets to each of the groups (paths) so that the _lter sets of each path can specialize in extracting features exclusive to each group. To allow the network input to be routed end-to-end over such parallel paths, we propose data-dependent parallel resource allocation methods layer-wise. Given a layer of parallel tensors, we _rst employ sub-networks that produce gating coe_cients to weigh cross- connections to the next layer of parallel tensors. Then, the next layer's parallel tensors are constructed by getting summations of the current layer's tensors, each weighted by the corresponding gating coe_cient. We demonstrate that our multi-path networks outperform previous widening and adaptive feature extraction, ensembles, and deeper networks with comparable complexity using image recognition challenges. To further regularize gating sub-networks, we think of a gating network's path allocation as a soft clustering of its input feature maps. Thus, we propose a neural mixture model-based clustering objective to use as a regularization loss for the gating networks, which We _rst study as a standalone neural network-based clustering approach. The proposed clustering framework uses a neural network to learn cluster distributions in mixture modeling instead of tuning human-de_ned distributions. We adopt the Expectation-Maximization (EM) algorithm to train the network and perform batch- wise EM iterations where the forward pass acts as the E-step and the backward pass as the M-step. For image clustering, we use the mixture-based EM objective as the clustering objective, along with consistency optimization. Our networks outperform traditional and single-stage deep clustering methods that still depend on k-means. Finally, we propose using this clustering objective to regulate gating networks to get distributed gating activation patterns. We show that the skewed gating patterns can be improved with such regularization loss as a local regularization. We further present the need for a global regularization method that takes the end task performance into account. We also suggest extending research towards sparse resource allocation, along with gating networks to handle more diversity.
item: Thesis-Full-text
Modeling the medium access control layer performance of cellular vehicle-to-everything mode 4 and IEEE 802.11p
(2021) Priyankara WNBAG; Samarasinghe T
The capability of vehicle-to-everything (V2X) communication to wirelessly exchange information on speed and location of vehicles over an ad hoc network envisions promise substantially reducing vehicle collisions, congestion, fuel usage and pollution. V2X communication plays a pivotal role in intelligent transport systems (ITS), with IEEE 802.11p and cellular V2X (C-V2X) being the two competing enabling technologies. This thesis focuses on discrete-time Markov chain (DTMC) based modeling of the medium access control (MAC)-layer performance of the two enabling technologies for evalua- tion, comparison and enhancement. Firstly, DTMC-based models for the MAC layer operations of IEEE 802.11p and C-V2X Mode 4 are developed, considering periodic and event-driven messages. The results show that IEEE 802.11p is superior in average delay, whereas C-V2X Mode 4 excels in collision resolution, which leads to its higher throughput. Then, the models are extended to support the parallel operation of four multi-priority data streams, which are crucial for quality of service (QoS). Results show that IEEE 802.11p is superior in maintaining fairness among multi-priority data streams. It is also shown that the higher delay values in C-V2X lead to unfavorable packet delays in the low priority streams. The thesis studies the allocation of multiple candidate single-subframe resources (CSRs) per vehicle as a solution. It proposes a methodology to determine the number of CSRs for each vehicle based on the number of total vehicles, and to assign the multiple data streams for simultaneous transmission. The numerical results highlight the achievable delay gains of the proposed approach and its negligible impact on packet collisions.
item: Thesis-Abstract
Joint channel-physical layer network coding in multi-way wireless relay systems
(2021) Balasuriya DN; Wawegedara KCB; Dias SAD
During the last two decades or so, physical layer network coding (PNC) has received a considerable attention as it provides superior spectral e ciency over conventional relaying, in wireless relay systems. However, error performance of the network coded relay systems is inferior to that of conventional relaying under poor quality channel conditions. On the other hand, channel coding provides improved error performance over noisy and fading channels. In channel and PNC coded wireless relay systems, a better performance can be achieved by performing channel decoding and network coding at the relay jointly compared to separately. However, the existing joint channel decoding and network coding algorithms cannot achieve a good trade-o between error performance and spectral e ciency when applied in a multi-way wireless relay system. This is mainly due to the fact that the existing algorithms operate the constituent sub-decoders independently. With the advancement of new trends such as Internet of Things (IoT), multi-way wireless relay system has been a popular network topology, hence joint channel decoding and network coding algorithms having very good spectral e ciency-error performance trade-o s are highly desired. This thesis presents, as the key technical contribution to the existing body of knowl- edge, a joint channel-physical layer network coding (JCPNC) algorithm for multi-way wireless relay systems, which achieves an improved trade-o between error performance and spectral e ciency. This improved performance is a result of harnessing additionaldiversity combining schemes are proposed and they are compared with each other. The thesis also presents an improved symbol value selection algorithm for the conventional non-binary symbol- ipping low density parity check decoder which is adopted to pro- duce a low-complexity JCPNC algorithm. Moreover, a novel JCPNC algorithm which can be employed in asymmetric multi-way wireless relay systems, is developed. Finally, the convergence behaviour of the proposed JCPNC algorithm is analyzed using extrinsic information transfer (EXIT) characteristics of the constituent sub-decoders. The error performance of the proposed algorithms is extensively investigated using computer simulations. The simulation results demonstrate that the proposed JCPNC algorithm and its variants achieve superior spectral e ciency-error performance trade- o s than the existing counterpart JCPNC algorithms. time diversity by exchanging information between constituent sub-decoders. Several
item: Thesis-Abstract
An Energy efficient distributed cluster based self organising algorithm for ad-hoc deployed wireless sensor networks
(2016-02-06) Gamwarige, PS; Kulasekere, C
Wireless sensor networks (WSNs) consist of a large number of inexpensive, low-power, sensors that can be placed in an ad hoc fashion to form a data gathering network. Subse- quent to the sensor node deployment, the nodes will self-organize themselves to periodically collect reliable information from the environment to a central location called base station (BS). Once the nodes are deployed, upgrading and maintaining them is not practical. In such a scenario, the main concern would be the optimal utilization of the sensor energy, so that the entire sensor bed lasts as long as possible gathering useful information. Inter node communication for network organization and information gathering requires the most energy. Therefore, it is necessary to manage these activities in an energy e cient manner to optimize the lifetime of the sensor network. This research focuses on nding energy e cient methods of operating the sensor bed such that the lifetime is maximally extended. Distributed clustering provides an e ective way for self-organizing the wireless sensor networks for periodic data gathering applications. The research identi es the most positive and negative aspects of the currently available distributed clustering algorithms. Based on these ndings, the research proposes a new energy e cient distributed clustering algorithm where the cluster heads (CHs) are selected based on relative residual energy level of sensors. Further, the cluster boundary determination and cluster head role rotation is governed by the cluster heads residual energy level. The algorithm favors more powerful nodes over the weaker ones thus makes local energy balancing to prolong the lifetime of the entire sensor network at a very low energy overhead. The proposed algorithm has realized near ideal local energy balancing. The proposed algorithm is also extended to achieve global energy balancing by introducing a mix strategy of communication (multi-hop and direct) from cluster head to base station. The research shows that the algorithm performance is in line with the desired objectives using analytical proofs to back the simulation test results. Further, the research proposes an analytical framework in determining the cluster distribution of the presented algorithm. Subsequently, the framework was extended to other similar types of distributed clustering algorithms. Finally, the research proposes an analytical technique in nding optimum al- gorithm parameters such as the cluster head message broadcasting range and cluster head role rotation.

Browse

Recent Submissions