Item EmbargoCollective pulse amplification in burst mode fiber laser amplifiers in gain-managed nonlinearity regime(Bilkent University, 2023-08) Maghsoudi, Amirhossein; İlday, Fatih ÖmerUltrafast lasers have diverse applications, ranging from optical metrology and spectroscopy to microscopy. Outside of the research laboratory, the most important application continues to be material processing, particularly precision micromachining. In traditional ultrafast processing, high-energy pulses at low repetition rates are used. The recently introduced ablation-cooled regime achieves greater speeds and material removal efficiencies through the use of moderate-energy pulses at much higher repetition rates. However, higher repetition rates also demand short pulses, and higher speeds require higher average powers, such as sub-50 fs pulses at kilowatt average powers. Such parameters are far beyond the current state of the art, but there appears to be no fundamental reason such performance could not be achieved. One common approach to exploit the ablation-cooled regime is to utilized bursts, or groups of pulses. However, such burst-mode amplifiers have resulted in relatively pulse durations to date. This thesis aims to overcome this limitation by implementing burst-mode operation in another recently discovered regime, namely, that of gain-managed nonlinear amplification. In ultrafast laser material processing, sub-50 femtosecond pulse widths and kW average powers are desirable. However, such lasers have yet to be reported in the literature due to the complexity of such systems. For obtaining sub-50 femtosecond pulses, the pulse evolution needs to be well-engineered. Additionally, kW average powers require active cooling systems. A practical method to decrease the average powers in the laser systems yet achieve the same pulse energies is to amplify bursts of pulses instead of a continuous stream of them. These bursts allow the energy to be more confined in time and the average power to be lower. In the no-burst case, various nonlinear pulse amplification regimes aid the design processes and produce high-quality pulses. One such regime is the gain-managed nonlinearity, observed after shifting the gain spectrum towards longer wavelengths accompanied by nonlinear spectral broadening. This process results in very broad optical spectra and short pulses for the amplified pulses. Obtaining sub-50 fs pulses in the continuous case is often complicated, and the results are hard to reproduce. However, with the gain-managed nonlinearity regime, sub-50 femtosecond pulses can be routinely obtained by correctly choosing the amplifier parameters. Despite that, with the inclusion of bursts, pulse parameters tend to be much worse, mainly due to having non-uniform pulses within a burst. Hence, obtaining sub-50 femtosecond pulses in burst-mode lasers is very challenging, and so far, no such lasers have been reported. Combining this regime with continuously pumped signal burst lasers, we have obtained pulses with 900 nJ energies and sub-50 femtosecond pulse widths with pulse repetition rates as high as 240 MHz and burst repetition rate of 1 MHz. The design process is aided by gaining knowledge regarding the nonlinear amplification process and the gain medium’s response time through mathematical modeling and numerical simulations. Burst-mode amplification in the gain-managed nonlinearity regime leads to new physical effects, whereby the amplification of one pulse depends on the other pulses constituting a burst. Item EmbargoAutomatic method for generation of sememe knowledge bases from machine readable dictionaries(Bilkent University, 2023-09) Battal, Ömer Musa; Koç, AykutThe minimal semantic units of natural languages are defined as sememes. Sememe Knowledge Bases (SKBs) are organized word collections annotated with appro-priate sememes. As external knowledge bases, SKBs have successful applications in multiple high-level language processing tasks. However, the construction of mainstream SKBs is performed by linguistic experts over extended periods, which restricts their prevalent usage. We present MRD4SKB as an automatic SKB generation method from readily available Machine Readable Dictionaries (MRDs). Construction of MRDs is more straightforward than SKBs, and many prominent MRDs are present in various forms. Consequently, the presented MRD4SKB is viable as a fast, flexible, and extendable method for SKB construction. Several variants of MRD4SKB, based on matrix factorization and topic modeling, are proposed to generate SKBs automatically. The performance of the automatically generated SKBs is evaluated and compared with that of other SKBs, which are constructed manually or semi-manually. Item EmbargoEnhancing reliability in semantic communication: a stochastic approach to semantic-graph modeling(Bilkent University, 2023-09) Yetim, Sadık Yağız; Arıkan, OrhanSemantic communication is expected to play a critical role in reducing traffic load in future intelligent large-scale sensor networks. With advances in Machine Learning (ML) and Deep Learning (DL) techniques, design of semantically-aware systems has become feasible in recent years. This thesis focuses on improving the reliability of the semantic information represented in a graph-based language that was previously developed. Inaccuracies in the representation of the semantic information can arise due to multiple factors, such as algorithmic shortcomings or sensory errors, deteriorating the performance of the semantic extractor. This thesis aims to model the temporal evolution of semantic information, represented using the graph language, to enhance its reliability. Each unique graph configuration is treated as a distinct state, leading to a Hidden Semi-Markov Model (HSMM) defined over the state space of the graph configurations. The HSMM formulation enables the integration of prior knowledge on the semantic signal into the graph sequences, enhancing the accuracy in identifying semantic innovations. Within the HSMM framework, algorithms designed for graph smoothing, semantic information fusion, and model learning are introduced. The efficacy of these algorithms in improving the reliability of the extracted semantic-graphs is demonstrated through simulations and video streams generated in the CARLA simulation environment. Item EmbargoA preclinical arbitrary waveform magnetic particle imaging scanner for multi-frequency imaging(Bilkent University, 2023-08) Yılmaz, Beril Alyüz; Çukur, Emine Ülkü SarıtaşMagnetic Particle Imaging (MPI) is a tracer based tomographic imaging modal-ity that images the spatial distribution of the magnetic nanoparticles (MNPs) using their nonlinear magnetization response. MPI is a rapidly growing and safe imaging modality with high temporal and spatial resolution, together with high sensitivity. In MPI, different types of MNPs and the properties of their local environment such as viscosity and temperature can be identified via the relax-ation behavior of the MNPs. The optimal drive field (DF) frequency depends on the application of interest. In addition, the sensitivity of quantitative mapping can benefit from imaging at multiple DF frequencies. However conventional MPI systems utilize an impedance matching circuitry tuned to a specific DF frequency to mitigate the reactive power, which in turn restricts the operation of the MPI systems to that frequency. In this thesis, a preclinical arbitrary waveform (AW) MPI scanner is proposed to enable flexible functionality in a wide range of oper-ating frequencies. The AW MPI scanner features three specialized components:(1) an AW drive coil with a reduced inductance achieved by utilizing Rutherford cable windings to enable wideband imaging in a preclinical-size MPI scanner, (2) a gradiometric receive coil designed to have zero mutual inductance with the AW drive coil to alleviate the effect of the direct feedthrough signal while sensitively receiving the MNP signal, and (3) additional capacitor banks to block DC cur-rent while avoiding distortions in the DF waveform. This thesis also proposes a technique for multi-frequency imaging in a single scan using the developed AW MPI scanner. Experimental imaging results demonstrate that MPI images and relaxation maps can be successfully achieved at multiple DF frequencies using the developed AW MPI scanner and the proposed multi-frequency imaging technique. Item EmbargoMEMS based ultrasonic gas sensor with universal sensing capability(Bilkent University, 2023-09) Erkan, Derin; Tatar, ErdinçGas sensors are a critical technology for life safety, process control, and most recently air quality measurements. Currently utilized gas sensing technologies need to be tailored to each specific gas, using either a chemically reactive substrate or an optical detector sensitive to certain gas types, providing very good selectivity at the expense of flexibility. In contrast, acoustic sensors promise a potentially universal method of gas sensing with lower selectivity, by measuring the speed of sound in a resonant cavity and inferring the gas content. In this work, a proof of concept for a MEMS based acoustic gas sensor is proposed. A horizontal cavity allows for a compact design, compared to vertical designs shown in the literature. Fabrication is simplified compared to existing CMUT/PMUT designs by using electrically tunable in-plane resonators as transducers. Fabrication of the designed sensor is carried out using an in-house developed SOI-MEMS process, while acoustic cavities are fabricated from silicon. During operation, one resonator excites the cavity while the other resonator measures the response. Frequency sweeps of the resonators while varying the tuning allows full characterization of device response. Overlaying sweeps at different tuning parameters reveals the cavity response, while testing with no cavity rules out parasitic effects. Both speed of sound and quality factor are observed, which can be used to improve selectivity in gas mixtures. The proof of concept device is tested in ambient air, measuring the speed of sound in air as 342 m/s, consistent with the literature and with external measurements. Item Open AccessEnd-to-end hybrid architectures for effective sequential data prediction(Bilkent University, 2023-08) Aydın, Mustafa Enes; Kozat, Süleyman SerdarWe investigate nonlinear prediction in an online setting and introduce two hybrid models that effectively mitigate, via end-to-end architectures, the need for hand-designed features and manual model selection issues of conventional nonlinear prediction/regression methods. Particularly, we first use an enhanced recurrent neural network (LSTM) to extract features from sequential signals, while pre-serving the state information, i.e., the history, and soft gradient boosted decision trees (sGBDT) to produce the final output. The connection is in an end-to-end fashion and we jointly optimize the whole architecture using stochastic gradient descent. Secondly, we again use recursive structures (LSTM) for automatic fea-ture extraction out of raw data but accompany it with a traditional linear time series model (SARIMAX) to deal with the intricacies of the sequential data, e.g., seasonality. The unification of the models is again in a joint manner; it is through a single state space and we optimize the entire architecture using particle filter-ing. The proposed frameworks are generic so that one can use other recurrent architectures, e.g., GRUs, and differentiable machine learning algorithms as well as time series models that have state space representations in lieu of the specific models presented. We demonstrate the learning behavior of the models on syn-thetic data and the significant performance improvements over the conventional methods and the disjoint counterparts over various real life datasets, with which we also show the generic nature of the frameworks. Furthermore, we openly share the source code of the proposed methods to facilitate further research. Item Open AccessAge of information-oriented comparative evaluation of channel access mechanisms in multi-rate wireless lans(Bilkent University, 2023-08) Erdem, Umut Utku; Karaşan, EzhanDelay-sensitive applications have recently garnered significant attention because of the increasing demand for real-time data and time-critical information. In delay-sensitive systems, the timeliness of the delivered information is crucial to guarantee a reliable operation. A performance metric called Age of Information (AoI) is introduced in the literature to measure the freshness of information. In this study, various channel access methods are comparatively evaluated for stations transmitting age-sensitive status update packets over a multi-rate IEEE 802.11 WLAN. For wireless networks carrying conventional data traffic, the legacy channel access mechanism imposed by the Distributed Coordination Function (DCF) allows sources to access the channel equally. This mechanism results in a throughput-fair bandwidth allocation which is also known as a performance anomaly in the literature. Airtime-fair channel access methods have been proposed in the literature for multirate wireless LANs to mitigate this anomaly. Recently, there has been a surge of interest in status update systems with the emergence of performance criteria called age of information. Age-based performance metrics (AoI, peak AoI) are more effective to satisfy the requirements of the carried age-sensitive traffic as opposed to using conventional performance metrics (throughput, delay, or loss). In this study, we propose a novel channel access mechanism for age-sensitive traffic which is devised to lessen the mean Peak AoI (PAoI) averaged over all the sources in the network, which is termed as the system PAoI. The proposed channel access mechanism effectively reduces the system PAoI compared to LCA and PFCA. Although system PAoI performance improvement depends on the system configuration, i.e. packet size, the multi-rate mixture of the network etc., system PAoI can be reduced up to 12.04% and 27.44% compared to the legacy channel access and airtime-fair channel access, respectively, for the considered system configurations in this study. Although the proposed channel access mechanism outperforms legacy and airtime-fair channel access mechanisms in terms of system PAoI, it may lead to a reduction in the overall throughput of the system compared to airtime-fair channel access. Item EmbargoA comprehensive analysis of GaN HEMTs: electro-mechanical behavior, defect generation, and drain LAG reduction with HfO2 layers(Bilkent University, 2023-07) Güneş, Burak; Özbay, EkmelGallium Nitride High Electron Mobility Transistors (GaN HEMTs) have rapidly emerged as a transformative technology, owing to the unique properties of the substrate material. They are poised to become a revolutionary advancement in RF amplifier applications, primarily due to their capability to operate at high frequencies and power levels with superior efficiency compared to conventional devices. Despite the rapid progressions, a noticeable gap persists in the literature regarding the relation-ship between mechanical stresses, defect generation, and their subsequent impact on the electrical characteristics of AlGaN/GaN HEMTs. Moreover, current dispersion effects, which are trapping induced reductions in output power, continues to remain a pressing issue. To address these limitations, this study first adopts a multifaceted approach and integrates mechanical simulations and Raman spectroscopy, in order to resolve fine details of stress distributions that a diffraction-limited Raman probe cannot resolve. This enables an extensive modeling of stresses in a typical HEMT structure and helps elucidate the underlying dynamics of defect generation, with the ultimate goal of informing and guiding the development of advanced fabrication techniques. In a second study, an ultrathin blanket dielectric deposition approach was devised to alleviate surface trapping, and consequently, mitigate current dispersion. The proposed streamlined fabrication process yielded a substantial improvement in device performance without compromising the transistor transfer characteristics. Item EmbargoSegmentation informed deep learning algorithms for cardiac MRI reconstruction(Bilkent University, 2023-08) Acar, Mert; Çukur, TolgaDeep learning methods have produced impressive results in accelerated magnetic resonance imaging (MRI) reconstruction from under-sampled k-space acquisitions. However, existing MRI reconstruction models are commonly trained with loss functions that uniformly weigh contributions from separate voxels across the field-of-view (FOV), without attributing focus on relatively important regions within the FOV. Furthermore common frameworks for model training rely on availability of large sets of fully-sampled MRI data to construct a ground-truth for the network output. This heavy reliance is undesirable as it is challenging to collect such large datasets in many applications, and even impossible for high spatiotemporal-resolution protocols. In this thesis, we first introduce a self-supervised learning methodology for dynamic cardiac MRI that trains the network to reconstruct acquisitions in the absence of fully-sampled data. We then introduce a segmentation-aware reconstruction framework which implicitly guides the reconstruction process around an ROI with the segmentation error signal. Lastly, we introduce RATNet, a reconstruction framework augmented with attention capabilities which explicitly carries spatial information into the reconstruction process to focus around regions of interest. Self-supervision reduces the excessive demand on fully-sampled data whereas the segmentation-aware re-construction framework backpropagates the spatial information signal in to the reconstruction network. Lastly, RATNet incorporates the attention layers into reconstruction which are sensitive to focusing information supplied by the spatial information network. We demonstrate recovering fully-sampled images from under-sampled acquisitions in cardiac MRI and show their state-of-the-art performance in medical image reconstruction. Item EmbargoWireless meta-structured RF probes for vibration sensing(Bilkent University, 2023-07) Kılıç, Tuğba; Demir, Hilmi VolkanVibration signals are widely used for different monitoring purposes in numerous areas of applications. Sensing vibration and examining its properties play a critically important role essential to damage monitoring especially in the fields of construction and machinery. Detection of possible damages to these structures/machines requires cost-effective and easy-to-use solutions both to protect human health and/or reduce the cost of potential damage to the structures/machines. In this thesis, to offer an efficient and reliable solution for monitoring the health and integrity of various structures and machinery, we proposed and developed a new class of meta-structure based vibration probes that offer high-resolution and real-time wireless monitoring capabilities in vibration sensing. Operating in the radio frequency (RF) domain, this sensor concept relies on the near-field coupling of two nested split ring resonators (NSRRs), each of which is free to move toward each other. In response to the mechanical vibration occurring on a surface to which one of the NSRRs is attached, the amplitude of the electromagnetic wave read out only in vertical direction with respect to the NSRR probe from the coupled-NSRR pair by a transceiver antenna monotonously changes, making the sensing system capable of detecting mechanical vibrations over a wide RF range. The most important advantage of the proposed sensing architecture is that the resonant frequency read-out is very strongly dependent on the spacing between the coupled-NSRR probes, which makes wireless vibration detection at low amplitudes possible. The experimental findings show that this system can wirelessly measure vibration amplitudes as low as 50 µm. Equally important, this opportunely enables a high level of vibration resolution of (differentiation of two close vibration amplitudes separated by) 38.4 µm with an average error rate of only 1.2%. The sensing system exhibits a sensitivity level of 866 kHz/mm. The wireless and passive nature of the proposed system, together with the cost-effectiveness of our NSRR probes, make it highly promising for real-life applications including remote structural health monitoring, deformation detection, and vibration wave monitoring. Item Open AccessNovel deep learning algorithms for multi-modal medical image synthesis(Bilkent University, 2023-08) Dalmaz, Onat; Çukur, TolgaMulti-modal medical imaging is a powerful tool for diagnosis and treatment of various diseases, as it provides complementary information about tissue morphology and function. However, acquiring multiple images from different modalities or contrasts is often impractical or impossible due to various factors such as scan time, cost, and patient comfort. Medical image translation has emerged as a promising solution to synthesize target-modality images given source-modality images. Ability to synthesize unavailable images enhance the ubiquity and utility of multi-modal protocols while decreasing examination costs and toxicity exposure such as ionizing radiation and contrast agents. Existing medical image translation methods prominently rely on generative adversarial networks (GANs) with convolutional neural networks (CNNs) backbones. CNNs are designed to perform local processing with compact filters, and this inductive bias is prone to limited contextual sensitivity. Meanwhile, GANs suffer from limited sample fidelity and diversity due to one-shot sampling and implicit characterization of the image distribution. To overcome the challenges with CNN based GAN models, in this thesis, first ResViT was introduced that leverages novel aggregated residual transformer (ART) blocks that synergistically fuse representations from convolutional and transformer modules. Then SynDiff is introduced, a conditional diffusion model that progressively maps noise and source images onto the target image via large diffusion steps and adversarial projections, capturing a direct correlate of the image distribution and improving sample quality and speed. ResViT provides a unified implementation to avoid the need to rebuild separate synthesis models for varying source-target modality configurations, whereas SynDiff enables unsupervised training on unpaired datasets via a cycle-consistent architecture. ResViT and SynDiff was demonstrated on synthesizing missing sequences in multi-contrast MRI, and CT images from MRI, and their state-of-the-art performance in medical image translation was shown. Item EmbargoSignal prediction for magnetic particle imaging using a model-based dictionary approach(Bilkent University, 2023-07) Alpman, Aslı; Çukur, Emine Ülkü SarıtaşMagnetic particle imaging (MPI) is a tracer-based medical imaging technique that enables quantification and spatial mapping of magnetic nanoparticle (MNP) distribution. The magnetization response of MNPs depends on both experimental conditions such as drive field (DF) settings and viscosity of the medium, and the magnetic parameters such as magnetic core diameter, hydrodynamic diameter, and magnetic anisotropy constant. A comprehensive understanding of the magnetization response of MNPs can facilitate the optimization of DF and MNP type for a given MPI application. This thesis proposes a calibration-free algorithm using model-based dictionaries for MNP signal prediction at untested experimental conditions. The proposed algorithm also incorporates non-model-based dynamics by modeling them as a linear time-invariant system. These dynamics include the system response of the measurement setup as well as the magnetization dynamics not accounted for by the employed coupled Brown-N´eel rotation model, such as dipolar interactions and non-uniaxial magnetic anisotropy. The proposed iterative calibration-free algorithm simultaneously estimates the dictionary weights and the transfer functions due to non-model based dynamics. Experiments on in-house magnetic particle spectrometer (MPS) setup demonstrate that the pro-posed algorithm successfully predicts the MNP signals at untested viscosities within the biologically relevant range, as well as at untested DF settings. Item Open AccessAnalysis of gender bias in legal texts using natural language processing methods(Bilkent University, 2023-07) Sevim, Nurullah; Koç, AykutWord embeddings have become important building blocks that are used profoundly in natural language processing (NLP). Despite their several advantages, word embed-dings can unintentionally accommodate some gender- and ethnicity-based biases that are present within the corpora they are trained on. Therefore, ethical concerns have been raised since word embeddings are extensively used in several high level algorithms. Furthermore, transformer-based contextualized language models constitute the state-of-the-art in several natural language processing (NLP) tasks and applications. Despite their utility, contextualized models can contain human-like social biases as their training corpora generally consist of human-generated text. Evaluating and re-moving social biases in NLP models have been an ongoing and prominent research endeavor. In parallel, the NLP approaches in the legal area, namely legal NLP or computational law, have also been increasing recently. Eliminating unwanted bias in the legal domain is doubly crucial since the law has the utmost importance and effect on people. We approach the gender bias problem from the scope of legal text processing domain. In the first stage of our study, we focus on the gender bias in traditional word embeddings, like Word2Vec and GloVe. Word embedding models which are trained on corpora composed by legal documents and legislation from different countries have been utilized to measure and eliminate gender bias in legal documents. Several methods have been employed to reveal the degree of gender bias and observe its variations over countries. Moreover, a debiasing method has been used to neutralize unwanted bias. The preservation of semantic coherence of the debiased vector space has also been demonstrated by using high level tasks. In the second stage, we study the gender bias encoded in BERT-based models. We propose a new template-based bias measurement method with a bias evaluation corpus using crime words from the FBI database. This method quantifies the gender bias present in BERT-based models for legal applications. Furthermore, we propose a fine-tuning-based debiasing method using the European Court of Human Rights (ECtHR) corpus to debias legal pre-trained models. We test the debiased models on the LexGLUE benchmark to confirm that the under-lying semantic vector space is not perturbed during the debiasing process. Finally, overall results and their implications have been discussed in the scope of NLP in legal domain. Item Open AccessFederated MRI reconstruction with deep generative models(2023-07) Elmas, Gökberk; Çukur, TolgaMulti-institutional efforts can facilitate training of deep MRI reconstruction models, albeit privacy risks arise during cross-site sharing of imaging data. Federated learning (FL) has recently been introduced to address privacy concerns by enabling distributed training without transfer of imaging data. Existing FL methods employ conditional reconstruction models to map from undersampled to fully-sampled acquisitions via explicit knowledge of the accelerated imaging operator. Since conditional models generalize poorly across different acceleration rates or sampling densities, imaging operators must be fixed between training and testing, and they are typically matched across sites. To improve patient privacy, performance and flexibility in multi-site collaborations, here we introduce Federated learning of Generative IMage Priors (FedGIMP) for MRI reconstruction. FedG-IMP leverages a two-stage approach: cross-site learning of a generative MRI prior, and prior adaptation following injection of the imaging operator. The global MRI prior is learned via an unconditional adversarial model that synthesizes high-quality MR images based on latent variables. A novel mapper subnetwork produces site-specific latents to maintain specificity in the prior. During inference, the prior is first combined with subject-specific imaging operators to enable reconstruction, and it is then adapted to individual cross-sections by minimizing a data-consistency loss. Comprehensive experiments on multi-institutional datasets clearly demonstrate enhanced performance of FedGIMP against both centralized and FL methods based on conditional models. Item EmbargoColloidal quantum well light-emitting waveguides(Bilkent University, 2023-07) Işık, Ahmet Tarık; Demir, Hilmi VolkanMicro/nanoscale semiconductor light-emitting devices of colloidal nanocrystals offer low-cost solutions while delivering high performance in ambient lighting systems, displays, and photonic circuits. Colloidal quantum wells (CQWs) are excellent candidates as active materials for these optoelectronic devices owing to their superior properties including suppressed Auger recombination, large absorption cross-section, and narrow emission linewidth. In this thesis, as our first study, we proposed and demonstrated dual-color lasing using heterostructures of CQWs as the gain media in an all-solution-processed dual-color optical cavity for the first time. Here, a multilayered waveguide architecture consisting of green- and red-emitting CQWs, separated with a transparent low refractive index colloidal spacing layer of silica nanoparticles (NPs) suppressing otherwise detrimental nonradiative energy transfer between them, enabled amplified spontaneous emission (ASE) simultaneously in two colors at the threshold level of ~17 µJ/cm2 . We further adapted this multilayer waveguide configuration to a whispering-gallery-mode (WGM) cavity by fabricating a microdisk structure directly out of these layered CQWs-NPs-CQWs colloids. The resulting device showed dual-color multimode lasing both at 569 and 648 nm at the same time with the threshold of ~106 µJ/cm2 . Then, as the second study of this thesis, we developed a colloidal waveguide light-emitting diode (LED) structure of CQWs that changes the direction of light from the surface to the edge of the device by combining the active CQW region with a slit-shaped waveguide architecture that confines the light within the emissive layer and guides it through the lateral axis. Driving this LED waveguide of 900 µm in length by 150 µm in width at a current density level of 5.6 A/cm2 , we observed the output emission reached a luminance level of ~20,400 cd/m2 . These unique waveguiding architectures integrated into the light emitting devices of CQWs hold great promise for on-chip photonic applications including CQW dual-color excitation for biological imaging and CQW LED-based photonic integrated circuits. Item Open AccessRetrieving Turkish prior legal cases with deep learning(Bilkent University, 2023-06) Öztürk, Ceyhun Emre; Koç, AykutThis study utilizes deep learning models to retrieve prior legal cases in the Court of Cassation in Turkey. Given the vast legal databases that legal professionals need to navigate and the ability of computers to handle large amounts of text quickly, information retrieval algorithms prove beneficial for legal practitioners. In this thesis, we introduce our legal recurrent neural network (RNN) models and the BERTurk-Legal model. We also introduce dense word embeddings for the Turkish legal domain. Moreover, we employ RNN autoencoders, Legal RNN autoencoders, combinations of RNN autoencoders with BM25 algorithms, and BERTurk-Legal to retrieve prior legal cases. We obtain the best results with the BERTurk-Legal model. Item EmbargoOutage capacity and throughput maximization using theoretical and learning-based approaches(Bilkent University, 2023-07) Masrur, Saad; Gezici, SinanThis thesis explores two research problems in wireless communications: the optimal channel switching and randomization problem in flat-fading Gaussian noise channels, and channel selection and switching approaches based on the upper confidence bound (UCB) bandit algorithm. In the first part of the thesis, the optimal channel switching and randomization problem is formulated and its solution is characterized for flat-fading Gaussian noise channels with the aim of outage capacity maximization under average power and outage probability constraints. For the single user scenario, it is proved that the optimal solution can always be realized by performing one of the following strategies: (1) Transmission over a single channel with no randomization. (2) Channel switching between two channels with no randomization. (3) Randomization between two parameter sets over a single channel. Hence, the solution can easily be obtained by considering only these three strategies. However, for the multiuser scenario, obtaining the optimal solution can have very high computational complexity. Therefore, an algorithm is proposed to calculate an approximately optimal channel switching and randomization solution (with adjustable approximation accuracy) based on the solution of a linearly constrained linear optimization problem. In the second part of the thesis, we consider the case of unknown channel statistics at the transmitter, and propose channel selection and channel switching approaches based on the UCB bandit algorithm for communications between a transmitter and a receiver over a block fading channel. In the absence of channel switching in a block, we propose a UCB bandit algorithm for selecting the best channel among the possible set of channels for maximizing the number of correctly received symbols per unit of time. In the presence of channel switching, we first define a set of virtual channels by considering all possible channel pairs with various power levels and timesharing factors. Then, a UCB bandit algorithm is utilized to determine the best virtual channel; hence, to find the optimal channel switching strategy. Also, a low complexity version of this algorithm is proposed for efficient convergence to the optimal solution when a high number of virtual channels exists. In addition, for comparison purposes, theoretical limits are presented when the channel statistics are available at the transmitter. Simulation results indicate that the proposed UCB bandit algorithms can achieve very close performance to theoretical limits over a sufficiently large number of blocks, and make benefits of channel switching be realized. Item EmbargoMicro-3D sculptured metastructures(Bilkent University, 2023-06) Atak, Anıl Çağrı; Demir, Hilmi VolkanToday three-dimensional (3D) printers are highly popular and find use in a vast range of applications thanks to their capability to construct complex 3D structures. However, 3D-printing vertical structures with a high aspect ratio of height to width remains a pending challenge especially when a high lateral resolution is required in large footprints. In this thesis, we propose and demonstrate micro-3D sculptured metastructures using the idea of constructing deep trenches to erect their high aspect ratio metal lines along long strips. To generate such deep-trenched 3D-patterns, our construction relies on nonlinear absorption process, enabling the two-photon polymerization (2PP). In our fabrication, the 2PP process requires optical trajectory optimization, followed by electroplating thick metal film and dry etching seed layer. To test the developed process flow of 2PP, we built three-dimensional RF metastructures showcasing the depth effect as a third dimension. Based on our systematic numerical and experimental studies, our designed metastructure resonators are found to fall within a targeted specific operating resonance frequency range, with their resonance frequency being con-trolled and shifted and their quality factor (Q-factor) tuned as a function of their cross-sectional aspect ratio. In the thesis, with these proof-of-concept demonstrations, we show that such 2PP-defined high aspect ratio RF resonators highly benefit in terms of tunability of their resonance frequencies, along with increased Q-factor and reduced footprint. The findings of this thesis indicate that the pro-posed fabrication method of producing deep trenches via 3D-printing provides rich opportunities to implement high aspect ratio, complex structures that are highly miniaturized. Item EmbargoOptically transparent metamaterial RF absorbers(Bilkent University, 2023-05) Şahin, Furkan; Ertürk, Vakur BehçetRecent advances in metamaterials have allowed to impart unique properties to flat RF absorbers including broadband absorption, low thickness (in terms of the longest operating wavelength) and polarization insensitiveness, all essential to high-performance absorbers. For these RF absorbers, introducing additional properties of high optical transparency (in the visible range) and mechanical ro-bustness opens up also stealth window applications. However, achieving all of these critical characteristics in a single design is a challenging task. In this thesis, to address this challenge, we propose and demonstrate an optically transpar-ent, broadband, and polarization-insensitive RF absorbing metamaterial that is extremely thin (thickness = 0.079λL; λL: longest operating wavelength). Our design consists of a single dielectric layer of polymethyl methacrylate (PMMA) sandwiched between the top and bottom indium tin oxide (ITO) films, altogether providing high optical transmission. The bottom ITO film acts as a ground plane, which reduces the RF transmission significantly. On the other hand, the top ITO film adorns a unique pattern that minimizes the RF reflection across a particular frequency range. Here we obtained these customized ITO patterns using a novel design methodology. We developed the fabrication process specific to the pro-posed RF structure and fabricated their prototypes. To validate numerical simu-lation results, we measured experimentally the RF absorption of these fabricated prototypes. The experimental results show that the proof-of-concept absorbers achieve over 90% absorption between 4.4-11.2 GHz and over 95% absorption between 4.8-10.6 GHz. Furthermore, we found the fabricated absorbers to be in-sensitive to polarization angles and preserve 90% absorption for oblique incidence angles of 60° for TM and 40° for TE polarizations in agreement with the numer-ical predictions. Also, besides RF characterizations, we optically recorded the transmittance in the visible range to be 65% on average for the tested absorbers. These findings indicate that the proposed single-dielectric-layered architecture of optically transparent, broadband, polarization-insensitive RF absorbers, featur-ing a record relative thickness of 0.079λL, holds great promise for use in stealth window applications. Item Open AccessAge aware power allocation for energy-efficient wireless networks using RSMA(Bilkent University, 2023-06) Akyürek, Selin; Karaşan, EzhanWith the commercial deployments of 5G, research in Beyond 5G (B5G) and 6G networks has started. Within the context of meeting all needs and demands of future generation networks, the predicted usage is envisaged in three cases: massive Machine-Type Communications (MTC), ultra-reliable low-latency com-munications, and enhanced mobile broadband. This thesis focuses on massive Machine-Type Communications (mMTC). Energy efficiency, under the banner of green communications and network-ing is one of the branches complementary to the research conducted on MTC. mMTC, industrial and medical Internet of Things (IoT) type technologies will demand not only networking capabilities for massive access, enhanced commu-nications, but also sustainability and power efficiency. Rate Splitting Multiple Access (RSMA) presents a candidate massive access scheme with spectral ef-ficiency, energy efficiency, reliability, Degree-of-Freedom (DoF) and Quality of Service (QoS) enhancements in most of user deployments and network loads over traditional access schemes used in 5G. Within the scope of the thesis, we propose an age-aware power allocation policy for minimizing the network’s Weighted-Sum Average AoI (WSAoI). To our knowledge, this is the first work in the literature which combines the Age of Information (AoI) concept and RSMA framework. For downlink communication, we formulate the network’s WSAoI minimiza-tion as a Markov Decision Process (MDP) and investigate an optimal as well as suboptimal policies for the Base Station (BS) to select a scheme among RSMA, Orthogonal Multiple Access (OMA), and Nonorthogonal Multiple Ac-cess (NOMA). We prove existence of an optimal policy. Complexity of com-putation is reduced by using an action elimination technique, and by using a sub-optimal policy with performance close to the optimal. We also investigate the tradeoff between energy and the WSAoI of the network. The adaptive RSMA only scheme outperforms adaptive RSMA/NOMA/OMA and OMA/NOMA on the basis of network’s WSAoI. For example, when RSMA is selected, the per-formance metric, WSAoI, at 14, 15, and 16 dB SNR values, is on average, re-spectively 35.8%, 15.7%, and 12.7% less than the NOMA/OMA cases. Overall, it is seen that, the optimum policy becomes more likely to operate in the RSMA mode with an increase in Signal to Noise Ratio (SNR). By using RSMA scheme instead of NOMA/OMA scheme, power consumption can be saved in average 65.8%, 62.3%, and 59.6% for the selected WSAoI values of 4, 3, and 2, respec-tively.