DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Decision support models and policy innovations to support automating store fulfillment
No full textDecember 2024School of EngineeringOmnichannel services, such as buy-online-pickup-in-store, curbside pickup, and ship-from-store, have shifted the order-picking tasks that used to be completed by in-store customers doing their own shopping to the responsibility of retailers. To support research on omnichannel serivces,relevant connected in-store and online customer data sets for omnichannel retail research are generated via a mapped categorization of products into product families. Using this mapping to connect previously separate in-store and online customer data sets, these data sets focus on a grocery retail environment, and collect additional data from publicly available websites. These connected data sets contain information about product family data on in-store and online customer demand values, impulse purchases, product dimensions, weight, and price. Additional data is provided on in-store and online customer arrival data. These data sets aid this work in generating numerical insights and can support future grocery retail logistical research. To support omnichannel services, many retailers have deployed a store fulfillment strategy, where online orders are picked from inventory in brick-and-mortar stores. As store fulfillment is currently a labor-intensive operation, this dissertation explores a new policy that relies on the assistance of in-store customers for item extraction from the store shelves and a fleet of Autonomous Mobile Robots (AMRs) to collect and transport them to a designated station. While a set of dedicated pickers and AMRs are manageable by the store, the arrival of in-store customers who are willing to assist an AMR at a given location in the store is out of the store's control, and therefore, uncertain. We model the stochastic order-picking problem with uncertain synchronization times of in-store customers and AMRs, first as a static approach via generating a consensus from multiple scenarios and decisions to visit picking locations are made at the beginning of the picking journey. Then we consider managing resources in a dynamic way, where the store makes new decisions as new information becomes available. We model the dynamic problem as a Markov Decision Process to determine how a retailer should dynamically assign tasks to a set of AMRs and dedicated pickers. We develop a heuristic solution framework that generates a set of initial assignments and routes for picking resources and dynamically updates them as the actual synchronization times between AMRs and in-store customers unfold. We analyze multiple strategies to generate the initial set of task assignments and routes as well as update such decisions based on the system state. We test our proposed approaches using actual online grocery data. Computational results illustrate the potential for AMRs and in-store customers augmenting the dedicated pickers to achieve equivalent pick rates compared to systems with only dedicated pickers. We further demonstrate that it is more effective for achieving higher picking performance to have in-store customers help the AMRs compared to a warehouse like environment where dedicated pickers are synchronized with AMRs. Moreover, our proposed policy improves the operating margin of the store compared to utilizing only dedicated pickers. Lastly, our solution approach is capable of generating high-quality solutions at a pace suitable for practical settings. In addition to fulfilling online customer requests, omnichannel retailers also must support in-store customers, who want to interact with products and often drive sales through impulse purchases and customer loyalty. Yet, how best to support both online and in-store customer channels efficiently and seamlessly is a current challenge for retailers. Thus, the second focus of this work is to explore whether new material handling equipment has the potential to be deployed in a retail store environment to support omnichannel services. To do so, we utilize pick performance data from a newly designed and built picker-to-stock robotic platform suitable for piece-level pick, sort, and place tasks in retail environments. Then an agent-based simulation model is created to mimic a store's logistical operations that integrates data from the robotic platform's lab demonstrations and data from online and in-store customer demand. An iterative process determines the minimum amount of manual and robotic resources needed to operate the store that satisfies a given service level for online order fulfillment and replenishment tasks. Then, to assess the economic viability of deploying such a robotic platform with currently achieved values and improved performance, these resource levels are combined with operational metrics obtained from the simulation and various cost aspects via an economic analysis model. Computational experiments show that deploying the robotic platform for picking and restocking goods in a store environment is operationally and economically viable for retail grocery stores providing omnichannel services using a store fulfillment strategy.Ph
Leveraging in-silico mediated approaches for the implementation of multimodal chromatography in mab purification
No full textDecember 2024School of EngineeringThe ever-expanding mAb therapeutic market, the concomitant diversification of new modalities derived from the mAb structure as well as the emergence of novel chromatographic materials continue to challenge the adaptability of the mAb platform processes. Additionally, while multimodal chromatography has been demonstrated to offer unique selectivity to address key purification challenges, they are seldom employed for mAb processes due to their complex behavior. Advancements in high-throughput techniques have greatly intensified the exploration of a large experimental design space, but the expansive combinations of protein diversity and available chromatographic material can ultimately complicate identification of versatile purification processes. This thesis work addresses this challenge through the development of an in-silico mediated workflow to efficiently evaluate and compare the extent of synergistic impurity separation across all possible resin sequences, with a focus on the implementation of multimodal chromatography. This approach was characterized by extensive selectivity screens to generate product and impurity retention databases, which then became inputs to an in-silico process ranking tool to rapidly and quantitatively identify highly separable processes based on novel separability indexes. The first application of this approach was for the expedited development of non-protein A 3-step processes for the effective removal of CHO HCPs for a mAb therapeutic. The resulting processes were determined to be highly separable as well as orthogonal, resulting in HCPs levels to below 100 ppm while maintaining good mAb recovery. To further demonstrate the utility of this approach, the in-silico mediated workflow was adapted for the rapid development of a 2-step polishing sequence for the removal of challenging product-related impurities for a mAb and Fc-containing therapeutic. The resulting databases revealed an unexpected ability of multimodal resins to resolve differently glycosylated variants (glycoforms), enabling the development axviii
straight flowthrough and integrated polishing sequence using only multimodal resins for the orthogonal removal of aggregates and glycoforms. Both applications of this workflow demonstrated its versatility to efficiently address industrially relevant purification challenges for mAb-based therapeutics while also providing new insights into purification opportunities afforded by multimodal chromatography. Building upon these insights, the application of multimodal cation exchange (MMCEX) resins was extended towards their development as potential alternatives to affinity solutions for the removal of LC impurity from a Fab therapeutic. The results revealed that selectivity was determined by a complex interplay of global protein biophysical properties and the relative electrostatic and hydrophobic properties of the multimodal ligands. Though complex, the additional levers to modulate selectivity revealed several MMCEX bind-elute opportunities for the purification of a Fab therapeutic containing a very challenging LC impurity load. Finally, the insights on multimodal chromatography learned throughout this work inspired the high-throughput evaluation of a library of novel MMCEX and MMAEX membrane adsorbers for the removal of CHO HCPs. The results showcased the potential of these novel polymer grafted membrane adsorbers to replace resin-based chromatography and enabled a systematic investigation into the impact of ligand chemistry on selectivity, thereby informing the rational design of next-generation multimodal ligands. As a whole, this thesis work demonstrates a platform workflow that is highly applicable across various industrial purification challenges, establishes the broad preparative utility of multimodal chromatography, and provides valuable implications towards development of next-generation chromatographic materials.Ph
Homophily and influence in online interactions
No full textMay2025School of ScienceWith the volume of social interactions that occur online every day and the increasing relevance online social spaces have in information dissemination, it is important to understand how interactions between people and other actors online affect how people adopt opinions and consume information. Some of these behaviors, like homophily, echo chamber formation, and misinformation can result in different populations of social media users consuming entirely non-overlapping sets of information about events occurring around them. These behaviors can drive the polarization of people online and increase misunderstanding between groups of ideologically divided people. This thesis aims to examine these behaviors in social networks, particularly how people are influenced online, how this influence is being used online, how to detect these groups of individuals, and how to test / validate these methods.
To identify and quantify these behaviors, this thesis performs two main analyses on social networks and human behavior. We first analyze social behaviors directly, by both developing social experiments, tested on human participants, to test human opinion dynamics directly and by analyzing human opinion dynamics in large Twitter datasets. In this Twitter dataset, we analyze millions of tweets, retweets and replies to examine information diffusion before the 2016 and 2020 U.S. presidential elections, identify the influencers that propagate information, and examine how these influencers and their interactions changed between the two periods. In this work we find that people online are susceptible to both the influence of unknown / anonymous users online as well as opinions / messages being propagated by bots or LLM agents. We also find that the information diffusion between influencers and users of Twitter have polarized further between the 2016 and 2020 elections, and that the set of influential accounts has begun to shy away from established media type accounts and shift to strong political personalities and unaffiliated lesser known people online.
In second set of works, we both develop methods for detecting clusters of users in social networks and for generating networks that are adequate benchmarks for testing such methods. We build on existing Modularity community detection methods and extend them to handle issues that occur in large / heterogeneous networks. We then identify and explore the creation of synthetic benchmark networks, defining properties necessary to generate networks with heterogeneous inter-community connectivity, creating hierarchical community structure and better representing behaviors present in many real world social networks.Ph
Applications of gel electrophoresis to study kinetically stable proteins in mushrooms and sesame seeds
No full textMay2025School of ScienceKinetically stable proteins (KSPs) are a special class of proteins that possess a high activation energy barrier for unfolding and thus exhibit remarkable resistance to denaturation. This stability allows these proteins to persist for longer periods of time in harsh in vivo conditions, protecting them from irreversible inactivation by aggregation or proteolysis. The correlation between protein kinetic stability (KS) and resistance to denaturation by the surfactant sodium dodecyl sulfate (SDS) has been well documented. This has led to the use of SDS-resistance as a probe for KS. Several polyacrylamide gel electrophoresis (PAGE) based assays have been developed to identify KS in protein isolates and complex lysates. This study involved the development of an aqueous extraction method for mushrooms, an organism of much interest for potential applications in mycoremediation, biofuel production, and the medicinal field. Mushrooms have a variety of enzymes that are excreted to break down nutrients found in the surrounding environment into compounds that can be readily absorbed by the mycelia. Due to the excretory nature of these proteins, it was hypothesized that mushrooms would possess many proteins with high KS to maintain their function in a variety of pHs, temperatures, and humidity levels. Here, the KS of proteins from several edible mushrooms (white button, cremini, portabella, enoki, shiitake, lion mane, king oyster, and oyster) was probed by diagonal two-dimensional (D2D) SDS-PAGE, which exploits the SDS-resistance of KSPs. This study revealed significant variations in KSP abundance, with white button and oyster mushrooms exhibiting particularly high amounts. Conversely, lion's mane displayed a lower abundance, demonstrating the species-specific nature of KSPs. The gels also highlighted the potential influences from disulfide bonds, phenolic compounds, and protease activity. Oyster and white button mushroom were furthered studied to characterize the subproteome of their KSPs.
In total, sixty-three KSPs and fifty-six KSPs were identified in oyster and white button mushroom, respectively. Approximately half of the KSPs were hydrolytic enzymes, including glycoside hydrolases and proteases. Additional KSPs that were identified were those that functioned in a bioprotective or regulatory role. Structural analysis, aided by AlphaFold, the CATH database, and homology studies, highlighted the prevalence of complex secondary structures, particularly α/β folds, and quaternary structures.
Active site alignments of kinetically stable polyphenol oxidases and hydrolases with current enzymes suggest a potential application of higher-order fungi for mycoremediation, medicinal applications, and biofuel production. The isolation and identification of these KSPs is one of the first studies focusing on the potential applications for individual proteins in higher-order fungi, rather than the application of the whole fruiting body and provided a promising foundation for enzyme activity studies as well as methods needed to probe other mushrooms and fungi.
Furthermore, the study explored the impact of post-translational modifications (PTMs), specifically glycosylation and coordinated metals, on protein stability in mushrooms. It demonstrated that glycosylation and metalation play key roles in deglycosylation using PNGase F resulted in a reduction of KSP abundance, particularly in white button mushrooms, highlighting the crucial role of glycans in maintaining structural integrity. Metal removal using EDTA also resulted in a decrease in KSP abundance, confirming the importance of metal ions in protein stability. This research underscored the untapped potential of edible mushrooms as a source of KSPs for various applications. Future studies should focus on elucidating structure-function relationships, investigating PTMs, and conducting activity assays of individual KSPs.
An additional study focused on probing the KS of allergenic proteins in the newest major allergen, sesame seeds, as highly stable proteins have often been linked to food allergies. In order to assess proteins that may possess a range of KS, a mild surfactant, sodium lauroyl sarcosinate (SAR), was used as a proxy for an intermediate level of KS. Two major allergens were identified to be kinetically stable: the 2S albumin (Ses i 1) and the 11S globulin (Ses i 6 and Ses i 7). A semi-quantitative comparison of their KS was also assessed using various SDS:SAR ratios, and their KS was then compared to their allergenic prevalence to assess their relative importance in eliciting allergy. A correlation was observed between the allergenic potential of the protein and the KS. This study demonstrated that KS could be a useful tool to probe the potential allergenicity of food proteins. The observed correlation between high stability and allergenic potency highlights the need for further investigations into the structural and functional properties of these proteins, with implications for mitigating sesame allergies and improving food safety.Ph
Designing and Evaluating an Ensemble Reasoning-based Clinical Decision Support System
No full textA clinical decision support system (CDSS) can help physicians make clinical decisions, such as differential diagnosis, therapy planning, or plan critiquing. To make such informed decisions, a physician may need to keep track of a large amount of medical data and literature, such as new research articles, pharmacological therapies, and updates in Clinical Practice Guidelines. Therefore, a CDSS can be designed to assist physicians by providing relevant evidence-based clinical recommendations, subsequently reducing the cognitive overhead required to stay up-to-date with an evolving body of literature. We designed a CDSS by leveraging Semantic Web technologies to create an AI system that reasons in a way similar to physicians. We base our abstraction of human reasoning on the Select and Test Model (ST-Model), which combines multiple forms of reasoning, such as abstraction, deduction, abduction, and induction, to arrive at and test hypotheses. Based on this framework, we perform ensemble reasoning, the integration and interaction of multiple types of reasoning. We apply our CDSS to the treatment of type 2 diabetes mellitus by designing a domain ontology, the Diabetes Pharmacology Ontology (DPO), that supports both deductive and abductive reasoning. DPO is also used to provide a schema for our knowledge representation of hypothetical patients, where each patient is encoded in RDF as a Personalized Health Knowledge Graph (PHKG). We use the Whyis knowledge graph framework to implement our CDSS. This is achieved by writing software agents to perform custom deductive reasoning and integrating abduction using an existing reasoning engine, the AAA Abduction Solver. We apply our approach to perform therapy planning on hypothetical patients
The psychedelic self social constructions of brain, personhood, and subjectivity through psychedelic science
No full textAugust2025School of Humanities, Arts, and Social SciencesSacraments, contraband, epistemic tools; psychedelic drugs have long had contested meanings in law, religion, and science. This dissertation seeks to better understand what psychedelics mean in modern society. Through a historical tracing of multiple discourses around psychedelics, this dissertation responds to critical questions about the ways in which psychedelic knowledge has been produced, and the role of psychedelic science in the coproduction of ways of knowing brain, identity, and subjectivity.In this dissertation, I consider molecular nomenclature as a technoscience and structural models as representations through which models of the neurochemical brain have historically been produced. Through archival research and interviews, I examine ways in which psychedelic science was both produced by and productive in the milieu of concurrently developing epistemic cultures of mind-brain sciences. I examine legislation and court decisions that distinguish identity based on psychedelic drug use and analyze ways in which the institutionalization of drug control rendered novel kinds of scientists as uniquely qualified to perform psychedelic research in human subjects. Finally, I investigate ways in which psychedelic researchers have produced quantitative results from qualitative accounts of subjective experience.
My findings indicate that molecular meanings of psychedelics were the product of the social relations immanent to the theoretical and practical use of molecular structural representations. Psychedelic research made serotonin visible to neuroscience through structure-activity relationship, as part of a broader pharmacologization of the brain. This way of knowing informed the legislation of drug control as well as its interpretation of historically significant state and federal supreme court decisions. During the same period, the regulatory bureaucracy around drug control created the conditions from which the current iteration of psychedelic science emerged.
This dissertation contributes to the growing attention in Science and Technology Studies on psychedelics. It calls attention to the historical significance of psychedelics in multiple forms of knowledge production and lays groundwork for future exploration of psychedelics as epistemologically significant scientific and sociopolitical actants.Ph
Characterizing and predicting challenging behavior in autism spectrum disorder using medical and environmental data
No full textMay2025School of EngineeringAutism Spectrum Disorder (ASD) continues to increase in prevalence in the population. However, about a quarter of this population meets the criteria for profound ASD, a new categorization of individuals with ASD and an IQ <50, that cannot live independently. Those with profound ASD are underserved in the literature, and have a higher rate of challenging behaviors than the overall population with ASD. Challenging behaviors may occur in response to physiological or psychological stimuli, and prevent individuals from being able to function safely in society by causing harm to themselves, peers, or caregivers. Many of those with profound ASD must live at residential facilities because of the severity of these behaviors and medical needs. Individuals' safety and daily outcomes may be better if these behaviors can be predicted in advance, and even prevented. First, this thesis explores how physiological variables may be leveraged to predict challenging behaviors of individuals with ASD living in a residential facility. Due to the many co-occurring conditions in individuals with ASD, there are many potential physiological causes of pain or discomfort. Residential facilities enable consistent recording of both physiological and behavioral data. In addition to physiological discomfort, changes in the environment may cause discomfort to these individuals. Environmental data are also examined as a predictive tool in this dissertation. The physiological data and environmental data are used in individual classification models to predict day-to-day behavioral episodes.
Next, this dissertation validates and extends these models for a subgroup of individuals. Further validation of these models is explored in several ways. Several model types are examined and compared for accuracy. The cohort is then reduced to a subgroup that can be reliably predicted by these models. The subgroup is used to examine whether the models can be improved by adding an autocorrelation component: using prior behavioral data to predict future behavior. The subgroup is also used to evaluate how many data points are needed to train the models, and how long the model predictions continue to be valid. These questions are important for implementing these models in practice.
Finally, the impact of the COVID-19 pandemic on challenging behaviors is examined. Previous work has not examined the impact of the COVID-19 pandemic on pediatric patients with ASD in a residential facility. A mixed generalized linear model was used to characterize the impact of the COVID-19 pandemic on behavior, and also included other contextual variables, including school days, weekends, and unexpected closures.
By further understanding both physiological and environmental associations with challenging behaviors, it is possible to reduce harmful impacts of the challenging behavior if it can be reliably predicted, and may be possible to prevent the behavior altogether. This dissertation examines several aspects of challenging behavior prediction in ASD, and characterizes the relationship between the COVID-19 pandemic and challenging behaviors in a residential setting.Ph
Semiconductor-rich group iv quantum alloys: the role of atomic ordering
No full textMay2025School of ScienceRecent years have witnessed rapid progress in topological quantum physics, as it introduces a new dimension in solid state physics with high potential for technological breakthroughs. Despite the explosion of interest in topological materials, their applications remain limited due to challenges in growth and incorporation with today’s microelectronics. As a potential bridge to close this gap, we propose two different strategies to achieve topological properties in Ge based alloys with ~90% Ge concentration and confirm that the atomic ordering in these alloy systems can serve as a new freedom to tune the topological properties. In the first case of GePb alloy, the random alloy with ~9% Pb can exhibit topological insulator phase, whereas atomic short-range order tends to disrupt this topology, resulting in a trivial insulator phase. This reveals that even with fixed Pb concentration, the topological properties can be tuned by manipulating the degree of SRO. However, Studies on GePb epitaxy remain limited, posing challenges for experimental realization. To propose a more experimentally accessible group-IV quantum alloy, we went in a completely different direction in the second case of GeSn alloys. Instead of using the randomness of the alloys, we proposed using ordered digital alloy structures where along one direction the Sn are ordered to form a superlattice in Ge. This made it possible for the Ge/Sn system to exhibit a rich variety of topological properties, such as triple-point semimetals, Weyl semimetals and weak topological insulators, at low Sn concentration of only 10%, which is much lower than the >35% required in GeSn random alloy. Additionally, we provide theoretical insight into the dynamic process of SRO formation during epitaxy, showing that the degree of SRO can be tuned by applying different growth techniques (MBE or CVD).Ph
Ethical Web Science Workshop Overview
No full textThe Ethical Web Science Workshop at WebSci’25 addresses the ethical challenges that arise from the rapidly evolving relationship between the Web and AI. With increasing reliance on Web-sourced data, issues of fairness, transparency, and consent have become central to technological innovation. To this end, we aim at bringing together researchers, practitioners, ethicists, and policymakers in order to create tools and guidelines for ethical compliance in Web-based research. By doing so, standards for ethical data sourcing and usage might be discussed and lead to guidelines for best practices
Cutting through the noise: do vibrato and formant shaping affect vocal detection?
No full textAugust2025School of ArchitectureOpera singers must project a single, unamplified voice above a full orchestra in halls whose reverberation commonly exceeds two seconds. Classical pedagogy credits two techniques with making this possible: vibrato—a 5–7 Hz modulation of fundamental frequency and amplitude—and formant shaping, which concentrates energy into a narrow “singer’s formant” near 3 kHz where orchestral spectra are sparse. The present thesis measures how strongly each technique contributes to audibility. A laboratory detection experiment asked six trained listeners to adjust the level of sustained vowels embedded in a steady orchestral chord. Among the four vocal conditions (±vibrato × ±formant shaping), vibrato alone produced the largest benefit, lowering thresholds to –36 dB for /A/ and –38 dB for /E/ relative to the mask. Formant shaping alone, by contrast, left /A/ unchanged and raised the /E/ threshold by roughly 6 dB. To examine whether these perceptual trends translate to production practice, four professional audio engineers mixed operatic phrases against the same orchestral texture in a constrained ProTools session. The clip gain applied to the vocal track served as a real-world proxy for the level boost required to achieve clarity. Engineers needed the smallest boost—only 4.8 dB—when formant shaping was present without vibrato, suggesting that the 3 kHz energy peak provides a spectral “handle” even though it yielded no advantage at the detection stage. Taken together, the findings show that vibrato enhances initial detectability, whereas formant shaping improves mix efficiency once the voice has been perceptually segregated. Although the study draws on a small sample (six listeners, four engineers, one singer, one orchestral chord), the convergence between laboratory and studio settings indicates that these techniques address complementary phases of the auditory-masking challenge.M