Cognitive radio is a promising technology that improves the spectral utilisation by allowing unlicensed secondary users to access underutilised frequency bands in an opportunistic manner. This task can be carried out through spectrum sensing: the secondary user monitors the presence of primary users over the radio spectrum periodically to avoid harmful interference to the licensed service. Traditional energy based sensing methods assume the value of noise power as prior knowledge. They suffer from the noise uncertainty problem as even a mild noise level mismatch will lead to significant performance loss. Hence, developing an efficient robust detection method is important. In this thesis, a novel sensing technique using the F-test is proposed. By assuming a multiple antenna assisted receiver, this detector uses the F-statistic as the test statistic which offers absolute robustness against the noise variance uncertainty. In addition, since the channel state information (CSI) is required to be known, the impact of CSI uncertainty is also discussed. Results show the F-test based sensing method performs better than the energy detector and has a constant false alarm probability, independent of the accuracy of the CSI estimate. Another main topic of this thesis is to address the sensing problem for non-Gaussian noise. Most of the current sensing techniques consider Gaussian noise as implied by the central limit theorem (CLT) and it offers mathematical tractability. However, it sometimes fails to model the noise in practical wireless communication systems, which often shows a non-Gaussian heavy-tailed behaviour. In this thesis, several sensing algorithms are proposed for non-Gaussian noise. Firstly, a non-parametric eigenvalue based detector is developed by exploiting the eigenstructure of the sample covariance matrix. This detector is blind as no information about the noise, signal and channel is required. In addition, the conventional energy detector and the aforementioned F-test based detector are generalised to non-Gaussian noise, which require the noise power and CSI to be known, respectively. A major concern of these detection methods is to control the false alarm probability. Although the test statistics are easy to evaluate, the corresponding null distributions are difficult to obtain as they depend on the noise type which may be unknown and non-Gaussian. In this thesis, we apply the powerful bootstrap technique to overcome this difficulty. The key idea is to reuse the data through resampling instead of repeating the experiment a large number of times. By using the nonparametric bootstrap approach to estimate the null distribution of the test statistic, the assumptions on the data model are minimised and no large sample assumption is invoked. In addition, for the F-statistic based method, we also propose a degrees-of-freedom modification approach for null distribution approximation. This method assumes a known noise kurtosis and yields closed form solutions. Simulation results show that in non-Gaussian noise, all the three detectors maintain the desired false alarm probability by using the proposed algorithms. The F-statistic based detector performs the best, e.g., to obtain a 90% detection probability in Laplacian noise, it provides a 2.5 dB and 4 dB signal-to-noise ratio (SNR) gain compared with the eigenvalue based detector and the energy based detector, respectively.

Scientific machine learning (SciML) is a new branch of AI research at the edge of scientific computing (Sci) and machine learning (ML). It deals with efficient amalgamation of data-driven algorithms along with scientific computing to discover the dynamics of the time-evolving process. The output of such algorithms is represented in the form of a governing equation(s) (e.g., ordinary differential equation(s), ODE(s)), which one can solve then for any time point and, thus, obtain a rigorous prediction.  In this thesis, we present a methodology on how to incorporate the SciML approach in the context of clinical trials to predict IPF disease progression in the form of governing equation. Our proposed methodology also quantifies the uncertainties associated with the model by fitting 95\% high density interval (HDI) for the ODE parameters and 95\% posterior prediction interval for posterior predicted samples. We have also investigated the possibility of predicting later outcomes by using the observations collected at early phase of the study. We were successful in combining ML techniques, statistical methodologies and scientific computing tools such as bootstrap sampling, cubic spline interpolation, Bayesian inference and sparse identification of nonlinear dynamics (SINDy) to discover the dynamics behind the efficacy outcome as well as in quantifying the uncertainty of the parameters of the governing equation in the form of 95 \% HDI intervals. We compared the resulting model with the existed disease progression model described by the Weibull function. Based on the mean squared error (MSE) criterion between our ODE approximated values and population means of respective datasets, we achieved the least possible MSE of 0.133,0.089,0.213 and 0.057. After comparing these MSE values with the MSE values obtained after using Weibull function, for the third dataset and pooled dataset, our ODE model performed better in reducing error than the Weibull baseline model by 7.5\% and 8.1\%, respectively. Whereas for the first and second datasets, the Weibull model performed better in reducing errors by 1.5\% and 1.2\%, respectively. Comparing the overall performance in terms of MSE, our proposed model approximates the population means better in all the cases except for the first and second datasets, assuming the latter case's error margin is very small. Also, in terms of interpretation, our dynamical system model contains the mechanistic elements that can explain the decay/acceleration rate of the efficacy endpoint, which is missing in the Weibull model. However, our approach had a limitation in predicting final outcomes using a model derived from  24, 36, 48 weeks observations with good accuracy where as on the contrast, the Weibull model do not possess the predicting capability. However, the extrapolated trend based on 60 weeks of data was found to be close to population mean and the ODE model built on 72 weeks of data. Finally we highlight potential questions for the future work.