DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Entry of molecular water into the silica glass crack tip during fracture: origin of environmentally-enhanced mechanical fatigue
No full textDecember 2023Water's role in diminishing mechanical strength and accelerating fatigue in silicate glasses is well-known in literature, but the mechanism behind this remains a topic of some debate. A novel experimental approach is presented to investigate the influence of water on glass fracture mechanics. Water content has been directly measured in fresh fracture surfaces generated by the double cleavage drilled compression method using attenuated total reflection, a surface-sensitive application of Fourier-transform infrared spectroscopy. Speciation and stress intensity dependence of water uptake during fracture have been determined quantitatively. This study finds that environmentally-enhanced fatigue is accompanied by significant molecular water entry, on the order of 0.1-1 wt%, with no observation of additional hydroxyl after fracture. In light of the new data from this study and extant work on water-enhanced internal friction, this study suggests that stress relaxation at the crack tip may be the driving factor behind environmentally-induced fatigue.Ph
LLM Experimentation through knowledge graphs: Towards improved management, repeatability, and verification
No full textGenerative large language models (LLMs) have transformed AI by enabling rapid, human-like text generation, but they face challenges, including managing inaccurate information generation. Strategies such as prompt engineering, Retrieval-Augmented Generation (RAG), and incorporating domain-specific Knowledge Graphs (KGs) aim to address their issues. However, challenges remain in achieving the desired levels of management, repeatability, and verification of experiments, especially for developers using closed-access LLMs via web APIs, complicating integration with external tools. To tackle this, we are exploring a software architecture to enhance LLM workflows by prioritizing flexibility and traceability while promoting more accurate and explainable outputs. We describe our approach and provide a nutrition case study demonstrating its ability to integrate LLMs with RAG and KGs for more robust AI solutions
Learning linear evolution of partial differential equations through koopman-inspired implicit neural representation
No full textDecember 2023Living in the age of big data, learning models from data is of great importance; that is, representingsystems using provided or collected data as opposed to a system representation derived from the
governing equations, which may be nearly impossible to derive (e.g., soft robotics) or may not even
be known (e.g., coarse-grained systems). Despite the recent success of data-driven methods such as
dynamic mode decomposition, SINDy, operator inference, these data-driven methods suffer from
several pitfalls: 1) requiring the full measurement of the system state, which is usually unavailable
in realistic setting; 2) unable to handle spatial-temporal data on complex, dynamically changing
mesh; 3) lack inductive biases and interpretability.
To address those challenges, this work introduces the Koopman Operator Implicit Neural
Representation (KoIN) by leveraging recent advances in the Koopman operator theory and implicit
neural representations. A neural implicit representation framework is employed to reduce spatial
complexity into latent representation where the dynamics is governed by a linear system. This
framework can be viewed as learning a Koopman operator of the nonlinear systems governed
by partial differential equations in a mesh-agnostic way. Moreover, we leverage auto-encoding
to avoid explicit encoding such that full state measurement is avoided. The effectiveness of this
framework is validated on several canonical PDE data-sets with a diverse set of initial conditions
ranging from 2D wave to incompressible Navier-Stokes equations. Our framework shows better
generalizations than state-of-the-art methods on linear PDEs while slightly inferior on nonlinear
PDEs. Overall, our framework trains much faster thanks to the simple linear representation.M
Diffusion equation-based room acoustic modeling using a physically informed neural network
No full textAugust 2024School of ArchitectureIn coupled volumes sound energy oscillates between the two rooms. To understand this sound energy flow, we use a diffusion equation model. The diffusion equation with a finite difference solution has been used to model sound energy flow in rooms. However, the finite difference approach requires a fine mesh, making it less computationally efficient for complex geometries or large scales. This work implements a mesh free solution using a physically informed neural network (PINN) with automatic differentiation. We can use PINN to simulate a sound energy impulse response in a room and predict room acoustics of complex geometries.M
Figuratively Speaking: Authorship Attribution via Multi-Task Figurative Language Modeling
No full textThe identification of Figurative Language (FL) features in text is crucial for various Natural Language Processing (NLP) tasks, where understanding of the author's intended meaning and its nuances is key for successful communication. At the same time, the use of a specific blend of various FL forms most accurately reflects a writer's style, rather than the use of any single construct, such as just metaphors or irony. Thus, we postulate that FL features could play an important role in Authorship Attribution (AA) tasks. We believe that our is the first computational study of AA based on FL use. Accordingly, we propose a Multi-task Figurative Language Model (MFLM) that learns to detect multiple FL features in text at once. We demonstrate, through detailed evaluation across multiple test sets, that the our model tends to perform equally or outperform specialized binary models in FL detection. Subsequently, we evaluate the predictive capability of joint FL features towards the AA task on three datasets, observing improved AA performance through the integration of MFLM embeddings
Explanation Ontology: A general-purpose, semantic representation for supporting user-centered explanations
No full textIn the past decade, trustworthy Artificial Intelligence (AI) has emerged as a focus for the AI community to ensure better adoption of AI models, and explainable AI is a cornerstone in this area. Over the years, the focus has shifted from building transparent AI methods to making recommendations on how to make black-box or opaque machine learning models and their results more understandable by experts and non-expert users. In our previous work, to address the goal of supporting user-centered explanations that make model recommendations more explainable, we developed an Explanation Ontology (EO). The EO is a general-purpose representation that was designed to help system designers connect explanations to their underlying data and knowledge. This paper addresses the apparent need for improved interoperability to support a wider range of use cases. We expand the EO, mainly in the system attributes contributing to explanations, by introducing new classes and properties to support a broader range of state-of-the-art explainer models. We present the expanded ontology model, highlighting the classes and properties that are important to model a larger set of fifteen literature-backed explanation types that are supported within the expanded EO. We build on these explanation type descriptions to show how to utilize the EO model to represent explanations in five use cases spanning the domains of finance, food, and healthcare. We include competency questions that evaluate the EO’s capabilities to provide guidance for system designers on how to apply our ontology to their own use cases. This guidance includes allowing system designers to query the EO directly and providing them exemplar queries to explore content in the EO represented use cases. We have released this significantly expanded version of the Explanation Ontology at https://purl.org/heals/eo and updated our resource website, https://tetherless-world.github.io/explanation-ontology, with supporting documentation. Overall, through the EO model, we aim to help system designers be better informed about explanations and support these explanations that can be composed, given their systems’ outputs from various AI models, including a mix of machine learning, logical and explainer models, and different types of data and knowledge available to their systems
Towards generating coherent stories from image sequences: a computational approach using suspension of disbelief
No full textAugust 2024School of Humanities, Arts, and Social SciencesThis dissertation explores a computational model for generating the connections needed to make a story from a sequence of images. Storytelling is an important cognitive task that humans use to communicate and organize information. Being able to automate storytelling would benefit interactive media such as video games, make it easier to automatically summarize large data sets in a way that people understand, and make AI systems more capable of explaining their actions. One storytelling task that humans perform is visual storytelling, the task of making a story from a sequence of images. To do so, humans establish recurring characters and props, connect actions in different images into a plot, and fit their plot to common emotional arcs, like tragedies. Humans performing visual storytelling reinterpret what actions they see in the images and interpret visually distinct objects in different images as being the same if doing so serves the story they want to make. Existing computational story generation systems do not show these human phenomena and cannot reinterpret the information in images for the sake of the story they aim to create. The aim of this thesis research is to create a computational system that makes the connections between images needed to make a story and can reinterpret the information in those images for the sake of that story. The system accepts the objects and their possible actions in a sequence of images as its input. It then establishes what objects are the same between images to make recurring characters and props, connects actions in different images into sequences to make a plot, and fits its plot to common emotional arcs. The system forms these connections by looking at what evidence supports them and creates different sets of connections depending on how it prioritizes different kinds of evidence.
To evaluate the system, human participants were asked to perform the visual storytelling task, and the objects and actions they identified in the images were gathered as input for the system. By varying the system's parameters, diverse sets of connections were generated from the same sequences of images. These sets of connections were examined to see if the system was equating objects and sequencing actions for its plot in a way that was consistent with how the system was expected to form these connections given its different evidence priorities. The results demonstrate the system's ability to vary the extent to which it equates visually distinct objects for the sake of its story and to adapt its interpretation of actions based on its desired emotional story arc.Ph
Mechanical characterization of full-thickness burned skin in incision and cutting experiments
No full textDecember 2023Patients with full-thickness circumferential burns may need to undergo a high-risk, urgent procedure known as escharotomy. Necrotic skin tissue, known as eschar, due to burn injury is incised to the subcutaneous fat layer to preserve surrounding healthy tissues and relieve pressure in the burn region. Medical personnel must be well-trained to perform this procedure, requiring the development of high-fidelity simulators with accurate haptic feedback. Such simulators require novel synthetic tissues engineered based on accurate mechanical properties of burned skin tissues. The mechanical behavior of skin tissue is known to significantly change with thermal injury.The goal of this thesis is to characterize the incision and cutting forces in full-thickness burned skin tissue. Porcine skin tissue is known to possess similar properties to human skin and is often used as a model for human tissue, as the usage of human tissue in experiments presents ethical and other concerns.
Experiments were conducted on unburned and burned porcine skin tissues at rates relevant to the skin surgery, i.e., 2 mm/s and 8 mm/s. The rate dependent mechanical behavior of skin tissues was investigated along with the effect of thermal injury on the mechanical behavior of skin tissue at each cutting rate. The analysis was conducted using parameters obtained from the incision and cutting force profiles. The peak incision force and energy required to rupture the skin tissue were extracted from the incision force profile. The average cutting force and the energy required to cut the tissue were estimated from the cutting force profile for the initial and final 10 mm length of cut and also for the full length of cut. It is found that the cutting force of full-thickness burned porcine skin tissue is rate dependent, whereas the incision force is rate independent. The mechanical behavior of skin tissue under thermal injury is found to change significantly during the cutting experiments, consistent with observations in the literature.M
First principles characterization of the chiroptical and magnetoelectric properties of chiral crystals
No full textAugust 2024School of ScienceResearch from the past two decades into chiral crystals has uncovered a variety of novel and impressive optical and magnetoelectric transport properties, prompting a push for first principles theorization to aid in materials discovery for potential applications in spintronic logic devices. Herein we provide a theoretical framework for the prediction of circular dichroism (CD) in crystals, demonstrating how the optical response is highly directionally dependent in complex perovskites. We further decompose the CD using orbital projections to show how the inorganic versus organic pieces of a perovskite affect its chiroptical properties. We next use first principles methods to quantify the internal magnetic field in arbitrary crystals, classifying the spin-orbit coupling (SOC) by its Rashba, Dresselhaus, and Weyl character. We demonstrate how chirality can be induced in achiral materials through the application of mechanical strain, and how the resulting SOC displays Weyl-like behavior. We follow this by assessing the use of inversion symmetry-breaking as a means of cryogenic cooling. Through the application of an external electric field, we quantify the entropy increase due to induced Rashba splitting in zincblende III-V semiconductors. We will show how sub-1K cooling is feasible through the use of electrically-tunable spin-orbit materials, providing an alternative to standard cryogenic cooling methods which require rare materials. Finally, we develop a theoretical framework for the computation of bilinear magnetoresistance in Rashba-split semiconductors. We show how the angular displacement between the electric and magnetic fields leads to minimized and maximized currents in wurtzite III-V systems. Our findings provide us with further opportunities to elucidate the rich physics arising from the lack of inversion symmetry.Ph
Leveraging satellite imagery and high-frequency sensors to understand variability in lake water clarity across spatial and temporal gradients
No full textAugust 2024School of ScienceFreshwater lakes provide many ecosystem services and are highly influential in global carbon, nutrient, and water cycling. Due to their location at the low-point of a watershed, changes in climate and land cover greatly impact lakes, and lakes are often considered sentinels of environmental change. Because of their importance to both humans and the surrounding environment, many monitoring programs and scientific studies focus on long-term changes in lake water quality. These programs frequently measure water clarity, a measurement of light transmittance through water, both because of its value for human uses such as drinking water and recreational use and because of its importance in regulating ecological variables such as light availability, thermal stratification and water temperature, primary productivity and dissolved oxygen, and fish and zooplankton habitat. Accordingly, water clarity measured as Secchi depth is one of the most frequently measured variables in lakes. Despite this, however, water clarity data is still limited across space and time. Secchi depth measurements are typically taken only monthly in the summer, and a majority of publicly available Secchi depth data is focused on a relatively small portion of large, well-monitored, north-temperate lakes. It is known that water clarity can vary daily in response to discrete disturbances, seasonally in response to phenological variability in stratification, zooplankton and fish species abundance, and meteorological drivers, and over years to decades in response to changes in climate and land cover. However, the current availability of water clarity data limits the comparison of variability across space and time. There has been no investigation of daily to seasonal variation at regional to global scales, and long-term water clarity data is lacking across most regions of the globe aside from North America and Europe. This dissertation examines variability in water clarity, as well as the causes and effects of this variability in lake ecosystems, across a broad range of spatial and temporal scales. The first project relates variability in global water clarity at a scale of days to decades using high-frequency underwater light sensors and public Secchi depth data. The second project explores seasonal variation in water clarity across the contiguous United States using satellite imagery. The third project uses satellite imagery to assess changes in water quality across three decades throughout southern Africa. The fourth project uses large-scale models of water clarity, lake morphometry, and stratification depth to explore variability in the depth of primary productivity across the United States. The sum of these projects examines water clarity and its influence on lake ecosystems at spatial and temporal scales that have not been previously studied.
My first project (Chapter 2) aims to relate short-term variability (days to months) in water clarity to long-term variability (years to decades). To do this, I used data from underwater light sensors across a suite of 35 global lakes to calculate short-term changes in light attenuation. Long-term data was assessed using publicly available Secchi depth measurements spanning an average of 12 years. Using Taylor’s Law, which states that the log of the variance of a positive quantity is linearly related to the log of the mean, I showed that long-term measurements of water clarity can reliably predict short-term variance, and vice versa. Despite the complex drivers of variability in water clarity, this chapter shows that it is predictable across a wide range of time scales, climate and land cover types, and lake morphologies. This project is the first to show that Taylor’s Law applies to high-frequency ecological measurements, providing valuable insight into landscape and macrosystems ecology and linking scales of measurement.
The second project (Chapter 3) explores seasonal patterns of water clarity in relation to climate and land cover. Using satellite remote sensing and machine learning, I created a model to estimate Secchi depth across over 90,000 lakes in the United States. I used time series clustering analysis to show that lakes in the US broadly fall into two distinct seasonal patterns of water clarity. The first pattern corresponds to a spring clear water phase, which is common in eutrophic and algal-dominated lakes and is most common in lakes with agricultural or developed watersheds. The second pattern demonstrates a summertime peak in clarity, and represents lakes that are optically dominated by dissolved organic matter or oligotrophic lakes that experience spring and fall algal blooms. This pattern is more common in lakes with higher forest and wetland cover in the watershed. I also show that the timing of peak water clarity is shifting earlier across all major ecoregions regardless of the seasonal pattern shown. This is likely due to a universal driver such as the increasing duration of thermal stratification and decreasing duration of ice cover that is occurring across many lakes. This project is the first to analyze seasonal patterns in water clarity at large spatial scales, and demonstrates the importance of land cover and climate in driving seasonality in lakes.
My third project (Chapter 4) aims to assess long-term trends in water quality across southern Africa in relation to changing climate and land cover. Southern Africa is a water-scarce region that is heavily impacted by drought and land cover change such as urbanization and deforestation, but has highly limited monitoring of water quality and quantity. This project uses remote sensing to delineate 6,900 lakes and reservoirs and estimate water clarity, chlorophyll, turbidity, and total suspended solids for the first time in most of these lakes and reservoirs. I show that water clarity across the region is heavily impaired, with a median Secchi depth of only 1.3 m. Despite changes in climate and land cover, water clarity has largely remained the same across the region over the past 30 years. However, some reservoirs have experienced large shifts in water clarity and quality due to invasive species, fluctuations in water volume, and watershed remediation efforts.
The final project (Chapter 5) partitions the potential for primary productivity based on light availability between the epilimnion, hypolimnion, and benthos in lakes across the United States. To accomplish this, I developed a machine learning model to estimate thermocline depth using lake morphometry and meteorological driver data. This model correctly determines if a lake is stratified or isothermal with 87% accuracy, and estimates thermocline depth with a mean absolute error of less than 1.2 m. Pairing this model with the water clarity model from Chapter 3 and a lake depth model, I show that primary productivity is possible throughout large portions of the hypolimnetic and benthic zones of lakes. Consistent stratification supports the potential for a deep chlorophyll maximum in one-third of US lakes, and a majority of lakes support benthic productivity. The relative volume epilimnion and hypolimnetic water and the relative surface area of the benthos that supports potential productivity varies seasonally and by ecoregion. Fluctuations in water clarity are the most important drivers of potential productivity at depth for shallow lakes, and in deep lakes the depth of the thermocline is the most important driver. This project is the first to examine the potential for productivity in non-surface waters at a large scale and shows that estimates of regional or global lake productivity need to properly account for variability in water clarity and thermocline depth.Ph