Graduate School of Engineering and Science
Permanent URI for this collectionhttps://hdl.handle.net/11693/115678
Browse
Recent Submissions
Item Open Access What captures our attention? How communicative and non-communicative actions by human and robot agents influence visual attention under perceptual load(2025-08) Karaduman, Tuvana DilanThis study investigated how agent type (human vs. robot), communicativeness (communicative vs. non-communicative), and perceptual load (low vs. high) interactively influence visual attention and task performance. A 2 (Agent) × 2 (Communicativeness) × 2 (Perceptual Load) within-subjects design was employed with 34 participants. To examine attentional allocation under perceptual load, Areas of Interest (AOIs) were defined for the central letter array and the peripheral agent videos. Behavioral performance (reaction time, accuracy) and multiple eye-tracking metrics (total dwell time, dwell time per fixation, first fixation latency, saccadic velocity, pupil dilation) were analyzed. Behavioral results confirmed the effect of the perceptual load manipulation, with participants being significantly less accurate and slower in high-loaded conditions. In the presence of a distractor, reaction times and accuracy were further modulated by significant interactions between agent, perceptual load, and communicativeness. Eye-tracking analyses revealed that initial orienting was driven by the communicativeness and perceptual load, when communicative cues captured attention faster only under high load. Dwell time per fixation remained stable across the conditions. On the other hand, total dwell time revealed a critical three-way interaction. Under high load, communicative actions performed by humans increase total dwell time, whereas if the action is performed by a robot, dwell time decreases. Pupil data supported this, indicating that under high load, non-communicative actions required more cognitive effort than the communicative ones. Initial orienting of eye metrics was guided by task demand; overall engagement with the nature of the action is a product of a complex and sensitive cognitive process sensitive to perceptual load. Robotic social cues are processed dissimilar compared to human ones.Item Embargo A surface dipole-based framework for fast and efficient solution of electromagnetic scattering problems(2025-07) Dilek, DoğaçThis thesis introduces a surface-based fast method for solving large-scale electromagnetic scattering problems within the Method of Moments (MoM) framework. Addressing the parallelization and low-frequency challenges of the Multilevel Fast Multipole Algorithm, the proposed approach represents conventional Rao-Wilton-Glisson (RWG) basis functions with uniformly distributed Hertzian dipoles placed on subdomain surfaces. This structural simplification facilitates translation operations, improves implementation efficiency, and supports parallel computing. The method comprises three stages: mapping the original current that is modeled with RWG basis functions to dipoles, translating fields via Green's function, and reconstructing local fields through inverse mapping. Numerical results are provided at both subdomain and full problem levels. Accuracy is first tested on isolated box interactions and then validated in a complete scattering problem involving a perfectly conducting sphere across different frequencies. The proposed method is benchmarked against MoM, a volumetric dipole-based approach, and the analytical Mie series solution, showing excellent agreement in all cases. Aimed to implement the method in Python with just-in-time (JIT) parallelization via Numba, the method is expected to demonstrate practical efficiency. Future work may explore symmetry-based translation reuse, machine learning for operator modeling, and GPU acceleration, making the method a promising candidate for hardware-friendly electromagnetic simulations that is strongly scalable for parallelization.Item Embargo Open-loop scheduling for minimizing polynomial functions of age of information(2025-07) Saydam, Mehmet SemihAge of Information (AoI) is a metric that quantifies freshness of information in a status update system, making it crucial for applications where timely updates are essential, such as real-time monitoring systems, IoT networks, and mission-critical communication systems. This thesis explores age-agnostic cyclic scheduling in multi-source, single-server Generate-At-Will (GAW) status update systems, where the goal is to minimize the expected value of a non-linear polynomial function of AoI in a discrete-time setting. In the literature, approaches that aim to minimize the weighted sum of average AoI often rely solely on the first two moments of packet service times. However, the use of non-linear functions of age as the information freshness metric, requires an analytical model to obtain the distribution of AoI which also uses the distributions of packet service times as input to the model. In this work, given a cyclic transmission pattern, we use the theory of discrete-time absorbing Markov chains to obtain the distribution of the AoI of each user, which then allows us to find the expected value of any nonlinear function of individual ages, referred to as the Value of Information (VoI) in this work. Subsequently, using the proposed analytical method, we propose a metaheuristic based space-search algorithm, specifically leveraging the Simulated Annealing (SA) technique, to obtain a cyclic schedule with the goal of minimizing a polynomial cost function of age. Numerical results are presented to validate the proposed approach.Item Open Access Compact and non-compact formulations for segment routing(2025-08) Yayla, Osman KağanSegment routing allows flexible control of packet routing using a segment list. A segment list consists of node-segments that utilize equal-cost multiple paths and adjacency-segments that directly use available links between source and destination nodes. This characteristic of segment routing makes it a good candidate for simplified and scalable traffic engineering. This thesis aims to minimize maximum link utilization (congestion) and processing energy costs simultaneously. In the thesis, a segment-based compact model and a path-based column generation model are developed. Initially, the models are utilized with maximum link utilization as the sole objective; both the compact model and the column generation model outperform existing methods in the literature, with the column generation approach excelling in its ability to solve large-scale and complex instances. Then, a multi-objective framework is tested for various parameter configurations that influence the energy consumption and congestion. The experiments over realistic network instances show potential for exploiting the energy-congestion trade-off for energy savings with minimal impact on congestion.Item Open Access Diaza-nazarov cyclization: an umpolung approach to access multi-substituted pyrazoles(2025-07) Karacaoğlu, Umut MertThe Nazarov reaction is a 4 thermal conrotatory electrocyclization reaction which is used to obtain five-membered ring structures.1 Among the heterocyclic variants for this pericyclic reaction is the aza-Nazarov reaction, which was previously studied by the Türkmen Research Group2,3 which incorporates a nitrogen atom into the ring, making it possible to access heterocyclic structures. To elaborate this concept, this project focuses on the development of a variant of the Nazarov reaction, which is scarcely studied on, namely the diaza-Nazarov reaction that yields a 3-hydroxypyrazole motif with a substituent at the N1 position. The substrate for the electrocyclization is prepared in 5 steps to construct an N-acyl azo derivative, and it was discovered that the treatment of this compound with 1-1.5 equivalents of TFA at room temperature gives the corresponding pyrazole derivative in high yields. In this project, the diaza-Nazarov reaction was developed, the conditions were optimized, substrate scope studies were conducted, a reaction mechanism was proposed, a control experiment was done to inquire about the β silicon effect, and further transformations were done on the diaza-Nazarov product to demonstrate that it is possible to do functionalization at later stages.Item Embargo Ultrafast laser synthesis of nanoporous zeolite A(2025-08) Doğan, Meryem MerveZeolites are microporous aluminosilicate self-assembled nanocrystals. Zeolite A is the first commercially synthesized zeolite, which has cubic unit cell that consist of alumina and silica tetrahedra together with the extraframework cations like Na+, K+, Ca2+. Zeolite A has been widely studied zeolite in the literature owing to the properties like high porosity, high surface area (∼ 600 m2/g), and high chemical and thermal stability, and high ion exchange capacity. All of these properties make Zeolite A used in various industrial applications like CO2 adsorption, wastewater treatments, biosensor applications, etc. Although zeolites have a wide range of industrial uses, they have still been extensively investigated in academia as well. Accordingly, various zeolite synthesis methods have been developed, but all have limitations. While conventional hydrothermal synthesis offers benefits such as high-quality discrete crystals, ease of use, safety, and industrial scalability, it lacks precise control over nucleation and growth. A method that combines these advantages with the ability to produce defect-free crystals using low-energy photons within a short reaction time has not been developed yet. To address this, we introduce a new method for synthesizing zeolites (TPA-silicalite-1, zeolite Y, zeolite A, and hierarchical ZSM-5) using ultrafast laser energy deposition. In this approach, energy is deposited on a timescale comparable to the polymerization reactions that drive crystal formation. In this thesis study, we further investigate the ultrafast laser pulses on the synthesis mechanism and crystalline structure of nanoporous zeolite A. Utilizing a femtosecond laser at 1040 nm wavelength, we achieved controlled energy deposition in the precursor suspension, accelerating the reaction via multiphoton absorption and laser-induced flows for nearly uniform-sized zeolite A crystals (∼260 nm). Through the controlled deposition of energy, this method achieves rapid crystallization of Zeolite A with high crystallinity (90-100%) and a narrower particle size distribution compared to the crystals synthesized via the conventional hydrothermal method. Comprehensive characterizations, including Scanning Electron Microscopy (SEM), Energy Dispersive X-Ray Spectroscopy (EDXS), X-Ray Diffraction (XRD), High Resolution Transmission Electron Microscopy (HRTEM), Selected Area Electron Diffraction (SAED), Thermo-gravimetric Analysis (TGA), Brauener-Emmet-Teller (BET), and Fourier Transformed Infrared (FTIR) Spectroscopy, revealed that the laser-synthesized zeolites exhibit structural integrity and quality comparable to conventionally synthesized counterparts. Moreover, CO2 adsorption capacity analysis was carried out to evaluate the gas capture performance and practical applicability of Zeolite A synthesized via the ultrafast laser synthesis method. The ultrafast laser synthesis method was successfully repeated over 80 times to enable various characterizations. This novel technique offers a rapid and alternative approach to synthesizing zeolites with precise control over structural and functional properties.Item Open Access Cell-free syhthetic biology enabled rna toehold switch system for influenza viruses(2025-08) Atilla, AbdurahmanInfluenza A H1N1 continues to pose a major public health threat due to its rapid transmission and capacity to cause seasonal epidemics and pandemic outbreaks. Rapid, accurate, and cost-effective diagnostic approaches are essential for timely intervention and control of viral spread. In this study, we present a synthetic biology based diagnostic strategy that employs programmable RNA-based regulatory elements, known as toehold switches, to selectively detect Influenza A H1N1 viral RNA sequences. These switches were designed to remain translationally inactive in the absence of the viral RNA and to activate protein expression upon specific sequence recognition. The initial switch designs were computationally generated and analyzed using thermodynamic modeling tools such as NUPACK and RNAfold, allowing for assessment of structural stability and identification of high-entropy regions that could affect translation. Particular attention was given to the accessibility of the ribosome binding site and start codon regions, as local structural hindrances in these areas were found to correlate with poor performance. Based on the entropy profiles and free energy distributions, selected constructs were subjected to rational sequence redesign to enhance conformational accessibility and minimize undesired leakiness. These optimized switches were then cloned into T7 promoter-driven plasmids and tested through in vitro transcription–translation reactions. The resulting GFP-based fluorescence measurements allowed us to quantitatively compare expression levels in the presence and absence of the target RNA. Experimental data showed that optimized switch designs provided significantly higher signal-to-noise ratios, reduced background expression, and more consistent fold-change values across replicates compared to their unoptimized counterparts. Overall, this study demonstrates the potential of rationally engineered RNA-based switches as a modular, programmable, and low-cost diagnostic platform for Influenza A H1N1. Moreover, the design framework established here can be generalized to support the development of similar RNA-sensing tools for other viral pathogens.Item Open Access Development of anti-diabetic living drugs via synthetic biology approach(2025-08) Tunç, NazlıcanType 2 Diabetes Mellitus (T2DM) is a prevalent metabolic disorder characterized by insulin resistance and impaired glucose regulation. Peptide-based drugs such as GLP-1 and its analog Exendin-4 are widely used in clinical treatment due to their ability to enhance insulin secretion and improve glycemic control. However, frequent injections, enzymatic degradation in the gastrointestinal tract, and short half-life limit their therapeutic efficiency and patient compliance. To address these challenges, this study aims to develop a living therapeutic system that enables the dynamic, gut-responsive production of anti-diabetic peptides using engineered Escherichia coli Nissle 1917. In the first part of the study, whole-cell biosensors responsive to physiologically relevant stimuli such as fatty acids, bile salts, and aspirin were constructed using synthetic regulatory elements. These biosensors were initially characterized through the expression of a fluorescent reporter gene (sfGFP) to determine their dose-response behavior and functionality. Following successful characterization, the reporter gene was replaced with either GLP-1 or Exendin-4 coding sequences, fused to various signal peptides (PhoA, MalE, TorA, DsbA, PelB) to promote extracellular secretion. Gibson Assembly and classical cloning techniques were used throughout the construct designs. The functional activity of secreted peptides was evaluated through ELISA-based quantification and in vitro bioassays using MIN6 insulin-secreting cells. MTT assays were performed to assess cell viability, and glucose-stimulated insulin secretion assays were conducted to determine the biological activity of the secreted peptides. Among the tested constructs, signal peptide–fused versions of GLP-1 and Exendin-4 showed significant effects on cell viability and insulin secretion, indicating successful expression and functionality of the therapeutic peptides. This study demonstrates the feasibility of combining probiotic bacteria with metabolite responsive gene circuits for targeted peptide delivery. The developed platform presents a promising strategy for designing next-generation living drugs capable of responding to the host environment and offering a self-regulated treatment for metabolic diseases like T2DM.Item Embargo The design and photophysics of a trinuclear iron triad(2025-07) Şekercileroğlu, Emir UtkuEarth‑abundant alternatives to noble‑metal photosensitizers are urgently sought for solar‑driven catalysis and photoredox chemistry. This thesis describes the design, synthesis, and comprehensive characterization of a fully iron‑based triad that melds a pyrazine‑functionalized Fe(II) N‑heterocyclic carbene (Fe‑NHC) chromophore with two {Fe(CN)5} acceptor units. Guided by ligand‑field considerations, the strongly σ‑donating NHC scaffold raises the iron eg orbitals, while the pyrazine contributes low‑lying π* levels, collectively stabilizing the metal‑to‑ligand charge‑transfer (MLCT) manifold. Stepwise coordination of the pentacyanoiron fragments broadens the absorption cross‑section into the visible‑to‑near‑IR region and establishes an electronically coupled Fe-Fe-Fe architecture that facilitates energy transfer between metal centers. Femtosecond transient‑absorption spectroscopy reveals sub‑picosecond intersystem crossing, inter-ligand charge transfer that bridges multiple MLCT states, energy transfer between metal centers with 380nm and 500nm excitation and exhibits ~70 ps excited state lifetime with 650nm excitation, accompanied by a non‑decaying excited‑state absorption persisting beyond the 5 ns experimental window. These findings establish a modular strategy for constructing high‑performance, all‑iron photosensitizers and provide fundamental insight into inter‑metal electronic communication within polynuclear architectures.Item Embargo Surrogate modeling of flows in hemodynamics and aerodynamics using machine learning(2025-08) Dülger, KeremThis thesis investigates the integration of traditional physics-based simulations and modern machine learning (ML) techniques for modeling fluid dynamics in both biomedical and engineering domains. The primary motivation stems from the computational limitations of high-fidelity simulations, particularly when exploring wide parametric spaces such as stenosis geometries in arteries. The study begins with a detailed Computational Fluid Dynamics (CFD) analysis of blood flow through idealized stenosed vessels. Stenosis geometry was parametrized using length (L), height (N), and shape exponent (n), and simulations were performed using both ellipsoid and superellipsoid profiles. The results demonstrated that stenosis height (N) has the most significant impact on hemodynamic parameters such as wall shear stress and pressure drop, with changes exceeding 300% in some configurations. However, the high computational cost associated with these simulations highlighted the need for surrogate models capable of rapid prediction. To bridge this gap, an intermediate ML study was conducted using NACA 4- and 5-digit airfoil data. This stage served as a controlled environment to understand the ML modeling pipeline, from dataset design and input encoding to hyperparameter tuning and generalization analysis. A compact, rule-based input encoding was developed from NACA designations and flow conditions, allowing neural networks to predict aerodynamic coefficients efficiently. The insights gained here, in areas such as dataset sensitivity, activation function choice, and model complexity, provided a solid foundation for applying ML to biomedical flows. Building on this groundwork, the final phase of the thesis applied ML models to predict hemodynamic quantities in stenosed vessels directly from geometric parameters. Various algorithms, including neural networks, support vector regression, and ensemble models, were trained on CFD-generated data. Neural networks showed the highest accuracy and strongest sensitivity to dataset size, making them particularly promising for biomedical prediction tasks. The resulting surrogate models significantly reduced computational requirements while maintaining high predictive fidelity, enabling scalable evaluations suitable for real-time clinical decision-making or large parametric studies. Overall, this thesis demonstrates a three-phase strategy: using CFD for physical insight, NACA-based ML for model development, and stenosis-based ML for biomedical application. The result is a flexible and robust modeling framework that leverages the strengths of both simulation and data-driven methods to accelerate research in fluid dynamics.Item Open Access Investigation of ion dynamics of devices with ionic liquid mixture electrolytes using X-ray photoelectron spectroscopy with electrical biasing(2025-07) Kutbay, EzgiX-ray Photoelectron Spectroscopy (XPS) has long been used to investigate surface composition, chemical states, and electronic environments in materials. In this study, XPS is employed not only for these traditional roles but also to extract local electrical potential profiles by monitoring shifts in core-level binding energies under applied DC and/or AC biases. Two complementary device architectures are utilized to achieve this objective: (i) platinum–platinum (Pt–Pt) coplanar capacitors, which are well-suited for investigating frequency-dependent charge screening, potential drop across the ionic liquid (IL) layer, and circuit modeling under square-wave excitation; and (ii) multilayer graphene–multilayer graphene (MLG–MLG) devices, which are employed to explore electrosorption dynamics, open-circuit potential (OCP) effects, and residual charge behavior after shorting. A Square Wave (SQW) AC signal with varying frequencies was employed to resolve dynamic charging and discharging processes in these systems. The co-planar capacitor configuration used was featuring a polyethylene membrane (PEM) coated with either the ionic liquid N,N-Diethyl-N-methyl-N-(2-methoxyethyl)ammonium bis(trifluoromethanesulfonyl)imide (DEME-TFSI) alone, or a ~1:1 volume mixture of DEME-TFSI and N,N-Diethyl-2-methoxy-N-methylethanaminium tetrafluoroborate (DEME-BF4). The measurements were conducted under operando conditions, allowing simultaneous acquisition of XPS spectra and current data. Although ionic liquids offer advantages such as wide electrochemical windows and thermal stability, their high viscosity and cost can hinder ionic mobility and conductivity. A promising approach to address these limitations involves mixing different ILs to tailor their properties; for example, adjusting the ratio of DEME-TFSI and DEME-BF4 alters electrolyte characteristics. The pronounced size difference between the TFSI and BF4 anions provides an opportunity for detailed investigation via XPS. Additionally, a DEME-TFSI system containing ~10% rubidium bis(trifluoromethanesulfonyl)imide (Rb-TFSI) was explored to assess the influence of a small, mobile alkali cation (Rb⁺) on surface composition and charge storage. XPS revealed polarity-sensitive surface accumulation of Rb⁺, correlating with a sharp increase in currents. Overall, this non-invasive methodology, leveraging both Pt–Pt and MLG–MLG architectures, demonstrates that XPS is a powerful tool for probing local electrochemical processes. The technique offers valuable insights that can contribute to the development of next-generation energy harvesting and storage systems.Item Embargo CRISPR-based synthetic translational regulation using non-conventional yeast(2025-07) Albayrak, DamlaEfficient and programmable gene expression systems are essential for improving recombinant protein production in non-conventional yeast hosts such as Pichia pastoris. In this thesis, a synthetic gene regulation platform was established in P. pastoris by integrating rationally engineered GAP promoters with CRISPR/dCas9-based transcriptional activation modules. The aim was to convert the native constitutive GAP promoter into a tunable element capable of both activation and repression through gRNA-guided recruitment of effector domains, thus paving the way for orthogonal and context-specific control of gene expression. Two synthetic promoter variants (version 1 and version 2) were designed by introducing targeted mutations to create novel gRNA binding sites without disrupting core promoter function. These promoters were cloned upstream of an eGFP reporter and integrated into the genome of P. pastoris. Colony screening under various carbon sources (glucose, glycerol, ethanol, and methanol) ii revealed that most mutant promoters retained expression levels comparable to the wild-type PGAP, while certain clones displayed elevated eGFP production due to multiple gene integrations. Quantitative PCR analysis was employed to identify single-copy integrants for further use. Subsequently, a CRISPRa system comprising dCas9, MS2-binding scaffold RNAs, and the VP64 activation domain was introduced into selected single-copy clones. Ten custom-designed gRNAs (five for each promoter version) were tested under four carbon conditions to assess their activation potential. Notably, version 1 demonstrated robust transcriptional activation with specific gRNAs, especially v1-g2-c1, which significantly enhanced eGFP expression across all tested conditions. In contrast, version 2 failed to elicit notable activation, possibly due to unfavorable gRNA positioning or inhibitory mutations within the promoter sequence. This work introduces a modular and orthogonal transcriptional regulation system in P. pastoris, offering dynamic control over synthetic promoters using CRISPRa components. The approach establishes a foundation for future metabolic engineering strategies and recombinant protein expression systems that are independent of traditional inducible promoters and adaptable to various industrial contexts.Item Open Access Novel models and methods for accelerating parallel full-batch gnn training on distributed-memory systems(2025-07) Bağırgan, Ahmet CanGraph Neural Networks (GNNs) have emerged as effective tools for learning from graph-structured data across diverse application domains. Despite their suc cess, the scalability of GNNs remains a critical challenge, particularly in full-batch training on large-scale, irregularly sparse, and scale-free graphs. Traditional one dimensional (1D) vertex-parallel training strategies, while widely adopted, often suffer from severe load imbalance and excessive communication overhead, limit ing their performance on distributed-memory systems. This thesis addresses the scalability limitations of 1D approaches by investigating alternative partitioning strategies for parallelization that better exploit the structure of modern graph workloads. A systematic evaluation framework is developed to assess parallel GNN training performance across a range of datasets with varying sparsity and degree distributions. The framework captures key performance indicators such as computational load balance, inter-process communication volume, and paral lel runtime. Extensive experiments are conducted on two Tier-0 supercomputers, LUMI and MareNostrum5, using hundreds of real-world graph instances. On average of 22 well-known GNN datasets, the results show up to 61% decrease in total communication volume and up to 39% decrease in parallel runtime com pared to 1D partitioning strategies on 1024 processes. These improvements are consistent across graphs with high variance in degree and sparsity, confirming the robustness of the proposed approaches. The findings demonstrate the potential of moving beyond traditional 1D paradigms and provide practical insights into scalable and communication-efficient GNN training on distributed platforms.Item Open Access Noise-tolerant wavefront shaping for focusing light through multimode fibers(2025-07) Ammar, AmnaMultimode optical fibers (MMFs) offer unique advantages for high-resolution imaging, optical communication, and power delivery. However, their complex modal structure poses significant challenges for the precise prediction of light propagation. This thesis explores the upper bounds of intensity enhancement achievable in light focusing through multimode fibers (MMFs) using phase-only wavefront shaping techniques designed to be robust against noise. We begin with a theoretical analysis of modal propagation and introduce the transmission matrix (TM) formalism as a foundation for describing input-output f ield relationships in MMFs. We then explore digital optical phase conjugation (DOPC) and feedback-based wavefront shaping strategies, emphasizing their per formance limitations under realistic experimental constraints. Acentral contribution of this thesis is the introduction of a generalized expres sion for the enhancement factor, incorporating both the input participation ratio and the phase error coefficient. We demonstrate that enhancement is strongly influenced by the choice of input basis and the presence of experimental noise. Using common-path interferometric transmission matrix (TM) measurements, we demonstrate that the Dual Reference Algorithm (DRA) implemented in the Hadamard basis outperforms the widely used Stepwise Sequential Algorithm (SSA) operating in the canonical (SLM pixel) basis. Our experimental results confirm that Hadamard-based wavefront shaping offers superior noise resilience, yielding intensity enhancement factors approaching the theoretical upper bound. We further conduct a detailed analysis of experimentally measured transmis sion matrices (TMs), revealing that the segment size on the SLM significantly influences modal coupling and focusing performance. Finally, we introduce an operator-based framework that encodes the radial memory effect for a focused beam, extending beyond the conventional rotational memory effect in multimode fibers (MMFs). This approach enables beam scan ning via controlled shifts of the input SLM pattern, paving the way for advanced applications in fiber-optic imaging and beam steering. Overall, this thesis presents a unified framework that bridges theory and ex periment to optimize wavefront shaping in multimode fibers (MMFs), with direct implications for endoscopic imaging, clean-beam fiber amplification, and pro grammable fiber-based optical systemsItem Open Access Creating competing charge-transfer pathways for water oxidation photocatalysis(2025-07) Tunçer, Hüseyin OrhunReplacing fossil fuels with renewable energy requires scalable and economic catalysts. However, solar conversion to chemical energy is heavily dependent on rare-earth metal-based photosensitizers and catalysts. Thus, it is necessary to build photocatalytic dyads that consist of first-row transition metals. However, Fe(II) polypyridyls, unlike their Ru(II) and Ir(III) analogues, show weak octahedral field splitting, which changes the photophysics to be unfavorable for catalysis. In this thesis, we focused on coupling metal-to-ligand charge transfer (MLCT) states with metal-to-metal charge transfer (MMCT) states to improvethe water oxidation catalytic activityof Prussian blue analogue(PBA) based photocatalysts. We first synthesised two of the most commonly studied Fe(II) polypyridyls, Fe(bpy)2(CN)2 and K2Fe(bpy)(CN)4. Then we coupled these chromophores with [Co(bpy)(H2O)n]2+ and [Zn(bpy)(H2O)n]2+ groups to prepare FeCo and FeZn compounds. FeZn compounds are used as a reference to investigate Fe cyanopolypyridyls in a network system since Zn ions are not expected to exhibit MMCT. With FeCo systems, it isaimed to investigate the role of the MMCT on the decay dynamics of the MLCT process. Characterization studies show that the local environment of Fe sites is affected by the coordination of Co and Zn ions to the cyanide bridge, resulting in a stabilization of the t2g and eg levels. We performed a series of photocatalytic experiments revealing that the highest catalytic activity is achieved for systems that possess MMCT processes directly coupled with the MLCT process through cyanide bridging group. Ultrafast transient absorption experiments indicate that the photophysics of the MLCT process is significantly different in Fe and FeCo compounds since 5MC is stabilized in energy compared to the 3MC state, and MMCT serves as an alternate relaxation pathway for the excited MLCT state. Finally, we combined our experimental findings to establish Jablonski diagrams for Fe, FeZn, and FeCo compounds and further proposed a mechanism for the photocatalytic water oxidation process. This thesis indicates that the cyanide bridge could provide an ideal platform to couple different charge transfer processes due to its rigid and short nature. The results also suggest that combining charge transfer processes could be a viable pathway to favor the decay dynamics of MLCT states for photocatalytic applications.Item Open Access An optimization approach to white matter brain tumor resection(2025-07) Bozkurt, Sena AslıPreserving brain functionality while performing maximal tumor resection remains a significant challenge in current clinical practice, as existing approaches do not offer a systematic framework to balance tumor removal with functional outcomes. To address this, we developed a mathematical optimization model for brain tumor resection problem that aims to preserve network efficiency as much as possible by resecting a predefined volume of tumor. Network efficiency is quantified using the Global Efficiency (GE) metric. The model includes spatial contiguity constraints to ensure that the resected area forms a clinically realistic connected region. To improve computational efficiency and scalability, we proposed enhanced formulations and introduced an aggregation heuristic that groups nodes that are physically closer to each other, significantly reducing solution time while maintaining high-quality results. For the experimental analysis, we used real brain network data representing healthy white and gray matter regions and generated synthetic tumor instances and modeled their disruptive impact on the brain network. We first validated the performance of the aggregation heuristic by comparing it against the original model. Results show that the heuristic drastically reduces computational time while achieving comparable GE values. We then investigated the impact of tumor size and found that larger tumors disrupt the network more severely even before resection, and lead to lower GE values post-resection at the same threshold levels. Then, we examined the impact of tumor location and observed that centrally located tumors lead to greater decreases in GE compared to peripheral tumors when the same tumor volume is removed. Finally, using a real tumor instance, we compared the ”best” resection guided by our model with the ”worst” resection, and demonstrated that our approach leads to significantly better preservation of brain network efficiency.Item Open Access Contextual object detection via image inpainting(2025-07) Çavdar, SinanObject detection in aerial imagery is a critical task in computer vision with ap plications in urban monitoring, disaster management, and military surveillance. Current approaches often encounter challenges, such as a lack of representative features of objects with cluttered backgrounds. To address these challenges, we introduce Contextual Object Detection via Image Inpainting (CODI), a novel approach that extract the representative features of the image’s semantic context derived from the transformer units of the inpainting model, and our proposed fusion module learns how to inject these contextual features with the representa tive features of the objects obtained from the related Feature Pyramid Network (FPN) of the object detection model. The outputs, which we called Contextual Pyramid Network (CPN), of the fusion model are used to replace the original FPN layers for better localization and labeling accuracy over the objects. Our experiments were conducted on DOTA and HRSC2016 datasets. Our model achieved promising results on mAP by prevailing over Oriented RCNN, which is one of the best-performing models on the DOTA dataset. Specifically, by apply ing the same pre-processing and post-processing, CODI achieved an improvement of 0.67% in mean average precision (mAP) at single-scale and 0.6% at multi-scale over the Oriented RCNN on the test set. On the HRSC2016 dataset, CODI achieved an impressive mAP of 90.57%, outperforming Oriented RCNN by 0.27 percentage points. This result places CODI among the top-performing models using a ResNet-50 backbone. In that sense, our approach provides a foundation for future applications, paving the way for more precise object localization and identification in remote sensing.Item Open Access Anticipation in discrete-time networked systems(2025-07) Murat, NaziraWe investigate networks of coupled dynamical systems in discrete time where the units anticipate the states of their neighbors and try to align their states accord ingly. Anticipation is done using past state information and hence introduces a memory e↵ect in the form of a time delay. We show that, under specific conditions depending on the network structure, the system converges to consensus faster un der anticipation. On the other hand, the delays introduced by anticipation can also break consensus altogether under certain other conditions. One of the main results of this study is the derivation of necessary and su cient conditions on the parameters of the anticipation algorithm for faster consensus. We apply the results to coupled nonlinear systems and analyze the stability of synchronized states under anticipation. We show that anticipation can induce synchronization in networks of chaotic maps, which may otherwise be di cult to synchronize.Item Open Access Toward ecological validity in biological motion perception : dissociating real-time and video stimuli(2025-07) Çakmakci, Elif AhsenTraditional laboratory settings often fail to capture the dynamics of real-world social cognition. To address this gap, a novel experimental setup was developed that minimizes the confounding effects between real and video-based stimulus presentation by using a transparent OLED screen to display live human actions in the visual periphery. This allowed for direct comparison of “real” (live) and video actions under controlled conditions. Attentional load (high/low), stimulus type (real/video), and action type (actions with response-eliciting capacity, actions without response eliciting capacity, no peripheral action stimuli) were systematically varied. Behavioral results showed greater distraction from real actions than video counterparts, especially under low-load conditions. EEG analyses (ERP and cortical oscillations) revealed distinct neural patterns for real versus video actions. Notably, neural differences emerged even in the absence of visible stimuli when the actor’s physical presence was known in the live session, indicating that presence alone is enough to modulate brain responses. Contrary to expectations, actions with higher social engagement potential did not capture more attention, possibly due to interaction constraints. The findings highlight the role of physical presence and contextual realism in shaping cognitive processing and underscore the value of ecologically valid paradigms in social neuroscience.Item Open Access Experience replay strategies for improving performance of deep off-policy actor-critic reinforcement learning algorithms(2025-07) Lorasdağı, Mehmet EfeWe investigate an important conflict in deep deterministic policy gradient algorithms where experience replay strategies designed to accelerate critic learning can destabilize the actor. Conventional methods, including Prioritized Experience Replay, sample a single batch of transitions to update both networks. This shared data approach ignores the fact that transitions with high temporal difference error, while beneficial for the critic’s value function estimation, may correspond to off-policy actions that can introduce misleading gradients and degrade the actor’s policy. To resolve this, we introduce Decoupled Prioritized Experience Replay, a novel framework that explicitly separates the transition sampling for the actor and critic to serve their distinct learning objectives. For the critic, it employs a conventional prioritization scheme, sampling transitions with high temporal difference error to promote efficient learning of the value function. For the actor, however, Decoupled Prioritized Experience Replay introduces a new sampling strategy. It selects batches that are more on-policy by minimizing the KullbackLeibler divergence between the actions stored in the buffer and those proposed by the current policy. We integrate Decoupled Prioritized Experience Replay with the state-of-the-art Twin Delayed Deep Deterministic policy gradient algorithm and conduct an evaluation on six standard continuous control benchmarks from OpenAI Gym and MuJoCo. The results show that Decoupled Prioritized Experience Replay consistently accelerates learning and achieves superior final performance compared to both vanilla and prioritized replay. More critically, Decoupled Prioritized Experience Replay maintains learning stability and converges to strong policies in tasks where standard prioritized replay failed to learn. Further ablation studies indicate that the decoupling mechanism is an important factor in this robustness and that the benefits of Decoupled Prioritized Experience Replay are achievable with a computationally inexpensive search, making it a practically effective solution for improving off-policy learning.