Doctor of Philosophy, The Ohio State University, 2024, Linguistics

Decades of psycholinguistics research have shown that human sentence processing is highly incremental and predictive. This has provided evidence for expectation-based theories of sentence processing, which posit that the processing difficulty of linguistic material is modulated by its probability in context. However, these theories do not make strong claims about the latent probability distribution of the human comprehender, which poses key research questions about its characteristics. Computational language models that define a conditional probability distribution are helpful for answering these questions, as they can be trained to embody different predictive processes and yield concrete surprisal predictors (i.e. negative log probabilities) that can be evaluated against measures of processing difficulty collected from human subjects. This dissertation evaluates Transformer-based large language models, which are artificial neural networks trained to predict the next word on massive amounts of text, as expectation-based models of sentence processing. Experiments reveal a robust and systematic divergence between the predictive processing of large language models and that of human subjects, the degree of which increases reliably with their number of parameters and amount of training data. This strong effect indicates that human sentence processing is not driven by the predictions made by these large-scale machine learning models, and highlights a fundamental shortcoming of large language models as models of human cognition. This dissertation additionally elucidates this discrepancy between humans and large language models. A series of analyses shows that large language models generally underpredict human-like processing difficulty by making 'superhumanly accurate' predictions of upcoming words, which may be a manifestation of the extensive real-world knowledge gleaned from large sets of training examples that are not available to humans. The learning trajectorie (open full item for complete abstract)

Committee: William Schuler (Committee Chair); Michael White (Committee Member); Micha Elsner (Committee Member); Tal Linzen (Committee Member) Subjects: Computer Science; Linguistics

Doctor of Philosophy, The Ohio State University, 2023, Materials Science and Engineering

In many scientific and engineering fields, the physical properties of materials can become a vital factor that limits the development of technologies or opens a new pathway to emerging applications. The study of ultrashort laser interaction with solid materials is no different. While the material damage induced by ultra-intense laser pulses has become the principal bottleneck in advances of high power laser technology, the highly localized nature of laser ablation can also expand the toolbox for high-precision material manufacturing. Moreover, few-cycle pulses can make ultrafast optical switching possible, which can be orders of magnitude faster than state-of-the-art electronic switches. In this thesis, with the assist of extensive experiments, a comprehensive computational model is proposed to study the mechanisms of ultrashort laser induced excitation and damage in dielectric materials. First, to capture the dynamic interplay of physical processes during the laser and thin-film material interaction, I wrote a two-dimensional finite difference in time domain (FDTD) multi-physical model incorporating the electromagnetic wave propagation, strong-field Keldysh photoionization theory, impact ionization, Drude-Lorentz model, etc. The simulation results for bulk fused silica and femtosecond laser pulses at varying durations and fluences agree well with the measurements. Then I modeled the laser interactions with multi-layer dielectric (MLD) mirrors and gratings designed for broadband pulses and predicted the laser induced damage thresholds (LIDTs). Next, laser damage experiments and computational modeling were performed to study the laser induced damage in the MLD mirrors and gratings designed for femtosecond laser pulses at 2-micron wavelength. The LIDTs measured by the experiments are consistent with the modeling results. I also observed the blister formation in both gratings and mirrors at fulences below the ablation threshold. The inner structure of the blisters w (open full item for complete abstract)

Committee: Enam Chowdhury (Advisor); Wolfgang Windl (Committee Member); Gregory Lafyatis (Other); Steve Niezgoda (Committee Member); Roberto Myers (Committee Member) Subjects: Materials Science; Physics

Doctor of Philosophy, Case Western Reserve University, 2023, EECS - Computer and Information Sciences

This dissertation presents a comprehensive approach to advancing proteomics and under-studied proteins in network biology, emphasizing the development of reliable algorithms, fair evaluation practices, and accessible computational tools. A key contribution of this work is the introduction of RoKAI, a novel algorithm that integrates multiple sources of functional information to infer kinase activity. By capturing coordinated changes in signaling pathways, RoKAI significantly improves kinase activity inference, facilitating the identification of dysregulated kinases in diseases. This enables deeper insights into cellular signaling networks, supporting targeted therapy development and expanding our understanding of disease mechanisms. To ensure fairness in algorithm evaluation, this research carefully examines potential biases arising from the under-representation of under-studied proteins and proposes strategies to mitigate these biases, promoting a more comprehensive evaluation and encouraging the discovery of novel findings. Additionally, this dissertation focuses on enhancing accessibility by developing user-friendly computational tools. The RoKAI web application provides a convenient and intuitive interface to perform RoKAI analysis. Moreover, RokaiXplorer web tool simplifies proteomic and phospho-proteomic data analysis for researchers without specialized expertise. It enables tasks such as normalization, statistical testing, pathway enrichment, provides interactive visualizations, while also offering researchers the ability to deploy their own data browsers, promoting the sharing of findings and fostering collaborations. Overall, this interdisciplinary research contributes to proteomics and network biology by providing robust algorithms, fair evaluation practices, and accessible tools. It lays the foundation for further advancements in the field, bringing us closer to uncovering new biomarkers and potential therapeutic targets in diseases like cancer, Alzheimer' (open full item for complete abstract)

Committee: Mehmet Koyutürk (Committee Chair); Mark Chance (Committee Member); Vincenzo Liberatore (Committee Member); Kevin Xu (Committee Member); Michael Lewicki (Committee Member) Subjects: Bioinformatics; Biomedical Research; Computer Science

Master of Science, The Ohio State University, 2020, Aero/Astro Engineering

Surrogates for computational fluid dynamics (CFD) offer the potential to significantly reduce computational expense associated with multi-discipline interactions in highly complex flow conditions. Motivated by this, and an increase in broad use of neural networks, this thesis project seeks to systematically assess accuracy, robustness, and efficiency of neural network surrogates for CFD predictions of surface pressure, heat flux, and shear stress distributions across an axisymmetric double cone and cylinder in hypersonic flow.

Committee: Jack McNamara (Advisor); Jen-Ping Chen (Committee Member); Daniel Reasor (Committee Member) Subjects: Aerospace Engineering

Doctor of Philosophy, The Ohio State University, 2020, Chemical Engineering

Quantitative Structure-Activity Relationships (QSARs) are empirical or semi-empirical models which correlate the structure of chemical compounds with their biological activities. QSAR analysis frequently finds application in drug development and environmental and human health protection. It is here that these models are employed to predict pharmacological endpoints for candidate drug molecules or to assess the toxicological potential of chemical ingredients found in commercial products, respectively. Fields such as drug design and health regulation share the necessity of managing a plethora of chemicals in which sufficient experimental data as to their application-relevant profiles is often lacking; the time and resources required to conduct the necessary in vitro and in vivo tests to properly characterize these compounds make a pure experimental approach impossible. QSAR analysis successfully alleviates the problems posed by these data gaps through interpretation of the wealth of information already contained in existing databases. This research involves the development of a novel QSAR workflow utilizing a local modeling strategy. By far the most common QSAR models reported in the literature are “global” models; they use all available training molecules and a single set of chemical descriptors to learn the relationship between structure and the endpoint of interest. Additionally, accepted QSAR models frequently use linear transformations such as principal component analysis or partial least squares regression to reduce the dimensionality of complex chemical data sets. To contrast these conventional approaches, the proposed methodology uses a locality-defining radius to identify a subset of training compounds in proximity to a test query to learn an individual model for that query. Furthermore, descriptor selection is utilized to isolate the subset of available chemical descriptors tailored specifically to explain the activity of each test compound. Finally, this (open full item for complete abstract)

Committee: James Rathman (Advisor); Bhavik Bakshi (Committee Member); Jeffrey Chalmers (Committee Member) Subjects: Chemical Engineering

Master of Science, The Ohio State University, 2017, Mechanical Engineering

The use of a manual wheelchair (MWC) for everyday mobility is associated with some degree of biomechanical risk, particularly to the user's trunk and upper extremities (UE), due to the loads placed on the body during propulsion and transfers. An improperly fitting wheelchair can require users to exert higher force or result in awkward positions that can place unnecessary strain on the UE. The combination of repetitive motion, higher peak forces and large joint deflections may result in musculoskeletal problems or injuries. Clinical fitting methodologies are primarily categorical and qualitative and as such are based on the clinician's perception and previous experience. Therefore, they do not provide a good basis for quantitative prediction of the impact of the wheelchair system on the user's biomechanics and the associated risk for developing additional musculoskeletal problems. Recent studies have focused on the identification of MWC user UE injuries and clinical prescription adjustments to prevent those injuries. While many adjustments have been supported using experimental data, computational modeling allows for a wider range of test case scenarios and the inclusion of additional factors that cannot be easily estimated in vivo, including the impact of deviations and changes to a wheelchair prescription on the user's force generation capabilities and more accurate risk identification. A few biomechanical models exist in current literature, but they are not adaptable for widespread use, utilize private software, are subject-specific or are insufficient in analyzing the user and wheelchair system. The MWC Propulsion Model 2017, created in OpenSim software by adapting a previously validated walking biomechanical model for application to a MWC and user, seeks to overcome the limitations of existing models, including accounting for a larger number of degrees of freedom and asymmetry. At this stage, the MWC Propulsion Model 2017 serves as a clinical teaching tool, (open full item for complete abstract)

Committee: Sandra Metzler (Advisor); Robert Siston (Committee Member); Carmen DiGiovine (Committee Member) Subjects: Biomechanics; Engineering; Health Sciences; Rehabilitation

Doctor of Philosophy, The Ohio State University, 2015, Electrical and Computer Engineering

American Sign Language (ASL) is a multichannel communication system that involves manual components such as hand-shape and movement, and nonmanual components as body posture, head motion and facial expressions. While significant progress has been made to understand the features defining ASL manuals, after years of research, much still needs to be done to understand its nonmanual components. ASL nonmanual linguistic research has been typically addressed by manually annotating facial events (e.g., brow raising, mouth opening, among others), and comparing the frequency of such events to find some grammatical clues about a given event in a sentence or as linguist called them construction. This tedious process is difficult to scale, especially when the number of facial events and the number of samples grow. Additionally, another major obstacle to achieve this goal is the difficulty in finding correlations between facial features and linguistic features, especially since these correlations may be temporally defined. For example, a facial feature (e.g., head moves down) occurring at the end of the movement of another facial feature (e.g., brows moves up), may specify a Hypothetical conditional, but only if this time relationship is maintained. It is however unknown for many grammatical constructions the facial features that define these dynamical facial expressions of grammar. In this work, we introduce a computational approach to efficiently carry out analysis of nonmanuals. First, a computational linguistic model of the face is defined to characterize the basic components used in ASL facial and head nonmanuals. Our results verify several components of the standard model of ASL nonmanuals and, most importantly, identify several previously unreported features and their temporal relationship. Notably, our results uncovered a complex interaction between head position and mouth shape. These findings define some temporal structures of ASL nonmanuals not previously identified b (open full item for complete abstract)

Committee: Aleix Martinez (Advisor); Kevin Passino (Committee Member); Yuan Zheng (Committee Member) Subjects: Computer Science; Electrical Engineering; Linguistics

Doctor of Philosophy, The Ohio State University, 2014, Psychology

A powerful theoretical framework for exploring recognition memory is the global matching framework, in which a cue's memory strength reflects the similarity of the retrieval cues being matched against all of the contents of memory simultaneously. Contributions at retrieval can be categorized as matches and mismatches to the item and context cues, including the self match (match on item and context), item noise (match on context, mismatch on item), context noise (match on item, mismatch on context), and background noise (mismatch on item and context). I present a model that directly parameterizes the matches and mismatches to the item and context cues, which enables estimation of the magnitude of each interference contribution (item noise, context noise, and background noise). The model was fit within a hierarchical Bayesian framework to ten recognition memory datasets that employ manipulations of strength, list length, list strength, word frequency, study-test delay, and stimulus class in item and associative recognition. Estimates of the model parameters revealed at most a small contribution of item noise that varies by stimulus class, with virtually no item noise for single words and scenes. Despite the unpopularity of background noise in recognition memory models, background noise estimates dominated at retrieval across nearly all stimulus classes with the exception of high frequency words, which exhibited equivalent levels of context noise and background noise. These parameter estimates suggest that the majority of interference in recognition memory stems from experiences acquired prior to the learning episode.

Committee: Per Sederberg PhD (Advisor); Roger Ratcliff PhD (Committee Member); Jay Myung PhD (Committee Member) Subjects: Psychology

Master of Arts, The Ohio State University, 2012, Psychology

There is building evidence in long-term recognition memory paradigms documenting a null list length effect (LLE) in which performance is not improved for items studied in shorter lists compared with items from longer lists. The global-matching mechanism implemented in many recognition memory models predicts that the LLE is a direct consequence of item noise. In these models, item noise is also assumed to drive the word-frequency effect (WFE) -- a recognition advantage for words that occur with low frequency in English compared to words that occur with high frequency. The current experiments modified a short-term recognition memory task (the Sternberg paradigm) to include an extended and/or filled delay between study and test lists, and manipulated word frequency. Results demonstrated a null LLE when the design included both a longer study-test lag and a distracter task. Frequency effects emerged only when there was an unfilled delay and a distracter task separating the study-test cycles. These results suggest that item interference is not implicated in short-term recognition memory for words, and contextual reinstatement must be sufficiently noisy to demonstrate the low frequency word advantage. BCDMEM is able to succinctly capture the range of findings that seem to provide evidence against distinct short-term- and long-term-memory systems.

Committee: Simon Dennis PhD (Advisor); Mark Pitt PhD (Committee Member); Per Sederberg PhD (Committee Member) Subjects: Cognitive Psychology

Master of Arts, The Ohio State University, 2011, Psychology

One of the critical findings in recognition memory is the null list strength effect (LSE): Strengthening items does not hurt the performance of other studied items. Memory models were able to predict the null LSE by using differentiation, which states that repetitions of a single item accumulate into a single strong memory trace. A hypothesized boundary of differentiation is that repetitions in different contexts will create new memory traces instead of differentiating them. Three novel list strength experiments tested this hypothesis by repeating words across three different study lists, followed by a test of all studied lists. Results indicated that as list strength increased, there was both a null LSE and no change in the false alarm rate (FAR), which is contrary to the predicted strength based mirror effect. These two results in tandem provide a challenge for differentiation models. Simulations of three different REM model variants along with a modified version of BCDMEM indicated that retrieval expectations may be responsible for changes in the FAR in response to list strength.

Committee: Simon Dennis PhD (Advisor); Roger Ratcliff PhD (Committee Member); Per Sederberg PhD (Committee Member) Subjects: Cognitive Psychology

Doctor of Philosophy, Case Western Reserve University, 2009, Mathematics

Computational models provide a useful tool for experimentalists in understanding the processes occurring in a biological system that may otherwise be impossible toobserve directly. The pivotal role of the liver in metabolic regulation makes it a challenging organ to model and simulate. Computational models that can adequately describe hepatic metabolism further the understanding of the functions within the organ. This thesis designs, identifies and analyzes three computational models of hepatic metabolism which account for the complexity of liver biochemistry, hepatic heterogeneity and perfused organ states. These models are governed by systems of ordinary or partial differential equations that depend on a large number of parameters that need to be identified. The classical deterministic parameter estimation problem is recast in the form of Bayesian statistical inference, allowing the integration of a priori belief and data from several experiments. In this approach, the unknowns are modeled as random variables and their values are probability densities. Effcient Markov Chain Monte Carlo techniques are designed and adapted to draw samples effectively from the parameter densities. Setting deterministic models inside a statistical framework makes it possible to study the correlations of different pathways with the time courses of metabolites. This methodology is applied to quantify the sensitivity of various hepatic pathways related to glucose production to redox state under varying conditions, providing insight into the regulation of hepatic gluconeogenesis. The Bayesian framework that we utilize allows us to incorporate into our parameter estimation process information available prior to considering the data. We show that the choices made in the encoding of this a priori information may affect both the parameter estimation and the corresponding model predictions by introducing three priors for a particular model and scrutinizing their effects.

Committee: Dr. Daniela Calvetti PhD (Committee Chair); Dr. David Gurarie PhD (Committee Member); Dr. Richard Hanson MD (Committee Member); Dr. Erkki Somersalo PhD (Committee Member) Subjects: Biochemistry; Biomedical Research; Mathematics

Doctor of Philosophy, Case Western Reserve University, 2007, Computing and Information Science

Large scale two-hybrid screens have generated a wealth of information describing potential protein-protein intereactions (PPIs). When interacting proteins are associated with each other to generate networks, a map of the cell, picturing potential signaling pathways and interactive complexes is formed. PPI networks satisfy the small-world property and their degree distribution follow the power-law degree distribution. Recently, duplication based random graph models have been proposed to emulate the evolution of PPI networks and to satisfy these two graph theoretical properties. In this work, we show that the previously proposed model of Pastor-Satorras et al.(2003) does not generate a power-law degree distribution with exponential cutoff as claimed and the more restrictive model by Chung et al.(2003) cannot be interpreted unconditionally. It is possible to slightly modify these models to ensure that they generate a power-law degree distribution. However, even after this modification, the more general l-hop degree distribution achieved by these models, for l>1, are very different from that of the yeast proteome network. We address this problem by introducing a new network growth model taking into account the sequence similarity between pairs of proteins as well as their interactions. The new model captures the l-hop degree distribution of the yeast PPI network for all l>0, as well as the immediate degree distribution of the sequence similarity network. We further utilize the PPI networks to discover possible pathway segments. Discovering signal transduction pathways has been an arduous problem, even with the use of systematic genomic, proteomic and metabolomic technologies. The enormous amount of data and how to interpret and process this data becomes a challenging computational problem.In this work we present a new framework to identify signaling pathways in PPI networks. Our goal is to find biologically significant pathway segments in a given interaction network. Fi (open full item for complete abstract)

Committee: Jiong Yang (Advisor) Subjects: