DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Resistivity size effect in tantalum layers
No full textMay 2024School of EngineeringIn situ and ex situ transport measurements at 293 and 77 K on epitaxial Ta(001)/MgO(001) and Ta(110)/Al2O3(112 ̅0) layers with thickness d = 7-350 nm are employed to quantify the anisotropic resistivity size effect in tantalum. X-ray diffraction θ-2θ scans, ω rocking curves, and φ scans indicate a 45° rotated epitaxy for Ta(001)/MgO(001), confirming single crystal layers. In contrast, Ta(110)/Al2O3(112 ̅0) exhibits two domains with different in-plane orientations. The measured resistivity ρ vs d is larger than the Fuchs-Sondheimer prediction at small d ≤ 20 nm, but is well described by a power law with ρ ∝ d¬-2 at both 293 and 77 K. Electron scattering at domain boundaries causes a higher resistivity for Ta(110) than Ta(001) layers, with the resistivity contribution from boundary scattering being proportional to d-1. Comparison of in situ and ex situ measurements after air exposure indicates that surface oxidation results in more pronounced diffuse surface scattering at both the Ta(001) and Ta(110) surfaces. The product of bulk resistivity ρo times electron mean free path λ for Ta is ρoλ = 26.2 × 10-16 Ωm2 for Ta(001)/MgO(001) and ρoλ = 36.7 × 10-16 Ωm2 for Ta(110)/Al2O3(112 ̅0). These values are 4-5× higher than for Cu, indicating that Ta is not a promising conductor for narrow interconnect lines.M
Influence of recording techniques and ensemble size on apparent source width
No full textDecember 2023The impression of listeners to aurally “see” the size of a performing entity is crucial to the success of both a concert hall and a reproduced sound field. Previous studies have looked at how different concert halls with different lateral reflections affect apparent source width. Yet, the perceptual effects of different source distributions with different recording tech- niques on apparent source width are not well understood. This study explores how listeners perceive the width of a symphony orchestra by using four stereo and one binaural recording techniques and three wave field synthesis ensemble settings. Beethoven’s Symphony No. 8, performed by wave field synthesis in three ensemble settings, was recorded using these five recording techniques and at two distances in EMPAC concert hall. Subjective experiments were conducted using stereo loudspeakers and headphone to play back the recording clips asking the listeners to rate the perceived wideness of the sound source. Results show that recording techniques greatly influence how wide an orchestra is perceived. The primary mechanism used in judging auditory spatial impression differs between stereo loudspeaker and headphone listening. When well-written symphonic music is recorded by stereo recording techniques, the changes in instrument positions (with the same number of instruments) in terms of increasing or reducing the physical source width do not lead to an obvious increase or reduction on the spatial impression of the performing entity.M
More Samples or More Prompts? Exploring Effective Few-Shot In-Context Learning for LLMs with In-Context Sampling
No full textWhile most existing works on LLM prompting techniques focus only on how to select a better set of data samples inside one single prompt input (In-Context Learning or ICL), why can not we design and leverage multiple prompts together to further improve the LLM’s performance? In this work, we propose In-Context Sampling (ICS), a low-resource LLM prompting technique to produce confident predictions by optimizing the construction of multiple ICL prompt inputs. Extensive experiments with three open-source LLMs (FlanT5-XL, Mistral-7B, and Mixtral-8x7B) on four NLI datasets (e-SNLI, Multi-NLI, ANLI, and Contract-NLI) and one QA dataset (CommonsenseQA) illustrate that ICS can consistently enhance LLMs’ performance. An in-depth evaluation with three data similarity-based ICS strategies suggests that these strategies can further elevate LLM’s performance, which sheds light on a new yet promising future research direction
SciKG: Tutorial on Building Scientific Knowledge Graphs from Data, Data Dictionaries, and Codebooks
No full textData from scientific studies are published in datasets, typically accompanied by data dictionaries and
codebooks to support data understanding. To conduct rigorous analysis, data users need to leverage this
documentation to correctly interpret the data. While this process can be burdensome for new data users,
it is also prone to errors even for seasoned users. A computational formal model of the knowledge that
was used to create the study can facilitate better understanding and thus improved usage of the study data.
Knowledge graphs can be used effectively to capture this study knowledge. The SciKG tutorial aimed to
introduce participants to the basics of knowledge graph construction using data, data dictionaries, and
codebooks from scientific studies. It used the Center for Disease Control and Prevention’s (CDC) National
Health and Nutrition Examination Surveys (NHANES) data as a testbed and introduce standardized
terminology, novel and established techniques, and resources such as scientific/biomedical ontologies,
semantic data dictionaries, and knowledge graph frameworks in both lecture and practical sessions
Natural language understanding with semantic parsing
No full textDecember 2022School of ScienceAs a branch of natural language processing (NLP), natural language understanding (NLU) involves transforming natural language into a structured, machine-readable format. Typical target formats vary depending on the domain problems to be solved and such transformations are usually challenging due to the semantic gap between natural language text inputs and target structured outputs. The work within this dissertation is focused on how semantic parsing, by providing an intermediate representation between text and desired outputs, can aid in some typical and representative NLU problems that demonstrate some key computational aspects of understanding. Compared to single-relation question answering, answering complex questions involving multiple KG relations by using a KG imposes additional difficulties of generating a corresponding SPARQL query due to the need of predicting implicit but critical graph structure that represents the intention of a natural language query (NLQ). We thus start with a domain-specific semantic parsing approach that extracts a semantic query graph as an intermediate representation between NLQ and SPARQL, with which the overall KGQA performance measured by accuracy and F1 score on the answer set is improved. Next, we explore the use of a more general-purpose semantic parse, abstract meaning representation (AMR), in a representative task that requires quantitative and logical reasoning capabilities -- math word problem solving -- which aims to derive math expressions for solving problems described in natural language. We present our novel Graph-to-Sequence/Graph-to-Tree learning approach that leverages AMR for graph construction to model the semantic relationship among quantities appearing in problem description. An edge-aware graph neural network encoder is reused for aggregating features from the AMR-based graph to enrich text representations that are used in a sequence/tree decoder for generating math expressions. We show that using our approach with AMR as an intermediate representation between text and expressions achieves higher logical form and solution accuracy on derived expressions, compared to the baselines without using semantic parse. Finally, we identify some limitations of AMR when using it for modeling world state in text-adventure games, which requires inducing a knowledge graph from observations written in natural language. First we present our path pattern based approach to leveraging AMR for this knowledge graph prediction task. We further enrich AMR with VerbNet semantics to collect more informative path patterns for relation prediction. We then conduct a detailed comparison study on our AMR and AMR+VerbNet driven approaches and representative baselines to assess the value of semantic parse for this task. As an extension, to introduce commonsense knowledge as another source of AMR enrichment, we further present a preliminary study on mining commonsense knowledge from dictionary term definitions by using part-of-speech tag patterns and existing triple scoring models. Through in-depth evaluation on the value of intermediate semantic representations for deriving target outputs from text inputs, we believe that the techniques proposed and discussed in this work can provide guidance to other similar NLU problems where the semantic gap between text inputs and target outputs is the main challenge.Ph
Molecular dynamics simulations towards understanding thermal stiffening in polymer nanocomposites with dynamically heterogenous interfaces
No full textMay 2023School of EngineeringThere is extensive research on polymer nanocomposite systems showing that fillers can improve the mechanical properties of polymer melts. This large body of work has given rise to several postulated mechanisms to explain the mechanical reinforcement effect observed in polymer melts. These mechanisms nanoparticle and polymer jamming, free volume effects, shear field interactions, nanoparticle bridging, thermo-reversible cross-linking, and dynamically asymmetric bound layers.In the first two studies we examine the viscoelastic and dynamic properties of heterogeneously grafted nanocomposites (PGNs) as a function of grafting density and then brush height via Molecular Dynamics simulations. The nanocomposite is comprised of a nanoparticle with High-Tg polymer graft chains within a Low-Tg polymer matrix. The viscoelastic and dynamic properties were studied at a temperature below the Tg of the graft chains as a function of graft chain density. PGNs with the highest graft densities showed the greatest increase in shear storage modulus over bare nanoparticle system. In addition, shear storage moduli increased with increasing average brush height. Analysis of the simulation results revealed that the reinforcement was observed when matrix chain mobility decreased as a result of graft chains acting as immobile obstacles.
In the next part of this work, we studied dynamically asymmetric composite blends and PGNs with Mw above the entanglement length for both the Low and High–Tg chains via Molecular Dynamics simulations. The PGNs were made up of two chains having a large glass transition temperature (Tg) difference, where the grafted chains have the higher Tg. In this study both the graft and matrix chains have high-Mw’s, above the entanglement length of the chains. Simulation results showed vastly different temperature responses between the neat, blended, and PGN systems. The entangled PGNs showed enhanced reinforcement, when compared to the neat matrix,at High–Tg concentrations far lower than that of the blended systems. Additionally, a thermomechanical analysis showed unique thermal stiffening behavior, for the high-Mw matrix PGN systems, a stiffening mechanism not observed in the other systems. Analysis of the simulation results revealed that the reinforcement mechanism for blended systems was proportional to the weight fraction of the High–Tg fillers and the slow-down of the dynamics of matrix polymer chains, which was reduced at elevated temperatures. However, the high-Mw matrix PGNs showed increased reinforcement at elevated temperatures as well as increased matrix and chain mobility. The following mechanisms were identified that were responsible for this PGN reinforcement: (i) At elevated temperature Low–Tg matrix chains dynamically coupled with High–Tg graft chains transmitting the applied stress through bulk of the polymer to the nanoparticle (ii) With increasing interfacial thickness, there were an increased number of graft/matrix entanglements the which drives the level of thermal reinforcement observed.Ph
Design of integrated time of flight system in cmos technologies
No full textAugust 2020School of EngineeringLight detection and ranging (LiDAR) or Time-of-flight (TOF) distance measurement technique is a well-known apparatus used in various areas such as astronomy, geology, soil science, agriculture, law enforcement, surveying, transport, automotive vehicles, etc. Emerging application spaces include human occupancy sensing, pose estimation, and gesture recognition in elderly care facilities, day-cares, schools, hospitals, and anywhere people movement monitoring adds value. Moreover, in the current COVID-19 pandemic situation, when small businesses and commercial enterprises are suffering from an unwanted and extended period of closures, this technology could be utilized in social distancing and workplace safety monitoring. For the widespread deployment of such sensors, the primary requirements are low latency, low cost, reduced energy consumption, and a small form factor without sacrificing the measurement accuracy. While the commercial sensors are available, they are either intended for long-range automotive applications, where a high dynamic range, low noise, and high precision are needed. However, power consumption, monolithic integration, and production cost are not priorities. On the other hand, the low power, low-cost proximity sensors do not have enough distance range to fulfill the requirement of the indoor occupancy sensing. This thesis presents the design considerations, analysis, and measurement results for CMOS avalanche photodetector (APD) based TOF sensor receiver system. The underlying system architecture consists of a large area CMOS avalanche photodetector, an analog front end (AFE), and a Time-to-Digital converter (TDC). A fully integrated TOF AFE is designed in 0.35 µm CMOS technology with on-chip large area APD. Measurement results show the 300 µm× 300 µm APD achieves a responsivity of 0.29 A/W at 850nm and 0.12 A/W at 940 nm wavelength with an avalanche gain of 10 before a breakdown voltage of 10.1 V. The depletion capacitance of the APD is 30pF. The AFE achieves a transimpedance gain of 40 KΩ and a bandwidth of 50MHz. The total chip, including the APD, occupies an area of 1.48 mm2 and consumes 132mW power during operation. Direct illumination free space measurement is performed up to 1 m distance with no lens at the transmitter side and a 50mm focal length lens in front of the receiver. Additionally, the same 0.35 µm CMOS technology is used for the implementation of a novel multi-tier TDC chip for converting the time difference between the transmitted and reflected received signal into digital values. The maximum measurement range of the TDC is 2.56 µS, corresponding to a distance of 384m and measurement accuracy of ≈ 400 pS, corresponding to a distance value of 6.23cm. Minimizing the numbers of delay elements with multi-stage interpolation strategy through use of cyclic coarse and fine interpolators, the TDC could achieve an integral non-linearity as low as 0.33 LSB rms while maintaining a dynamic range of 2.56 µS. The area occupied by the TDC with pads is 3.74 mm2, and the power consumption is 165mW. The designed blocks in CMOS technology paves the way for low-cost and high accuracy TOF sensor solutions for personal and commercial indoor privacy-preserving sensing applications.Ph
Accelerator-enabled non-uniform mesh procedures for pic simulations targeting plasma surface interactions
No full textDecember 2022School of EngineeringIn fusion reactors, when a high-velocity plasma species strikes a material wall it sputters the wall impurities into the domain which can adversely affect the sustainability of the fusion reactions. High-fidelity plasma-surface interaction models are needed to understand the behavior of impurities and their impact on fusion reactions. The overall impurity transport problem requires a multi-scale model. In particular, it requires a small-scale model to perform high-fidelity simulations of the surface yield as well as the sheath region (a tiny portion closest to the wall, with a thickness of few millimeters) and a device-scale model for high-fidelity impurity transport simulation in the fusion domain (which involves dimensions in the order of meters or tens of meters). In this work, the small-scale model involving the surface and sheath region is based on kinetic descriptions of particles to simulate the sputtering of impurities from the material wall into the plasma domain as well as the sheath electric field. Subsequently, the impurities are tracked in the device-scale transport model to simulate the migration and interaction kinetics of the impurities and any further surface events/interactions. Problems of interest include conditions that often produce high gradients in the specific portions of the domain which calls for highly anisotropic and non-uniform/graded meshes. The primary focus of this work is on the design and development of accelerator-enabled (e.g., GPU-enabled) capabilities (including data structures and algorithms) for particle-in-cell (PIC) simulations on highly non-uniform/graded meshes. Specifically, we target two aspects based on novel procedures designed for: (i) block-structured, non-uniform meshes to simulate sheath physics, and (ii) 3D complex geometries and unstructured anisotropic meshes to simulate device-scale impurity transport. In particular, this research has enabled two new types of PIC simulations on non-uniform meshes targeting: (i) sheath electric field, and (ii) integrated impurity transport. In both cases, highly performant and scalable procedures as well as software on GPUs are developed. The utility of the current computational tools are demonstrated on multiple problems of interest including tokamak devices such as ITER and DIII-D. Specifically, a one-of-its-kind integrated impurity simulation, including high-fidelity 2D sheath simulation, is performed.Ph
Informing clinical assessment by contextualizing post-hoc explanations of risk prediction models in type-2 diabetes
No full textMedical experts may use Artificial Intelligence (AI) systems with greater trust if these are supported by ‘contextual explanations’ that let the practitioner connect system inferences to their context of use. However, their importance in improving model usage and understanding has not been extensively studied. Hence, we consider a comorbidity risk prediction scenario and focus on contexts regarding the patients’ clinical state, AI predictions about their risk of complications, and algorithmic explanations supporting the predictions. We explore how relevant information for such dimensions can be extracted from Medical guidelines to answer typical questions from clinical practitioners. We identify this as a question answering (QA) task and employ several state-of-the-art Large Language Models (LLM) to present contexts around risk prediction model inferences and evaluate their acceptability. Finally, we study the benefits of contextual explanations by building an end-to-end AI pipeline including data cohorting, AI risk modeling, post-hoc model explanations, and prototyped a visual dashboard to present the combined insights from different context dimensions and data sources, while predicting and identifying the drivers of risk of Chronic Kidney Disease (CKD) - a common type-2 diabetes (T2DM) comorbidity. All of these steps were performed in deep engagement with medical experts, including a final evaluation of the dashboard results by an expert medical panel. We show that LLMs, in particular BERT and SciBERT, can be readily deployed to extract some relevant explanations to support clinical usage. To understand the value-add of the contextual explanations, the expert panel evaluated these regarding actionable insights in the relevant clinical setting. Overall, our paper is one of the first end-to-end analyses identifying the feasibility and benefits of contextual explanations in a real-world clinical use case. Our findings can help improve clinicians’ usage of AI models.This work is supported by IBM Research AI, USA through the AI Horizons Network. We also thank the clinicians on our expert panel discussions. We thank Rebecca Cowan from RPI and Ching-Hua Chen from IBM Research for their helpful feedback on the work
Indoor occupancy detection for commercial office spaces using sparse arrays of time-of-flight sensors
No full textMay 2023School of EngineeringUnderstanding the locations of occupants in a commercial built environment is critical for realizing energysavings by delivering lighting, heating, and cooling only where it is needed. The key to achieving this goal is being able
to recognize zone occupancy in real time, without impeding occupants’ activities or compromising privacy. In this thesis, we investigate inexpensive, low-resolution time-of-flight (ToF) sensors, which can provide real-time estimates of the distances to objects in a room while preserving occupants' privacy. To estimate the number and locations of occupants with our sensors, we created both real-world and simulated testing environments and designed two generations of occupant-counting ToF systems in this environment. In the first-generation system, we designed sensor pods with VL53L1 ToF sensors and developed two algorithms: a multiple line based algorithm and a neural network method to count occupants in several specific zones. However, due to the low spatial resolution and low frame rate of our L1 pods, the first-generation system performed poorly when multiple people go through doorways at the same time. Hence, we designed a second-generation system using VL53L5 sensors. We developed and validated an algorithm for zonal occupancy counting that can deal with multiple people walking underneath the sensors in arbitrary directions. We also evaluated the system in realistic simulations of office spaces and in a real-world installation. Our system's performance was independently tested, demonstrating that it can handle moderately crowded commercial offices. While the systems have demonstrated good performance in zone counting, their performance depends on careful sensor placement. To address this issue, we propose an automatic sensor placement method that determines optimal sensor layouts for a given number of sensors and can predict the counting accuracy of such a layout. In particular, given the geometric constraints of an office environment, we simulate a large number of occupant trajectories. We then formulate the sensor placement problem as an integer linear programming (ILP) problem and solve it with the branch and bound method. To further estimate the locations of occupants, we built a human trajectory prediction model using inverse reinforcement learning. The motion of occupants was modeled as a Markov decision process (MDP) and the policy only depends on the current state and the current observation of the environment. With this model, we can estimate the distribution of occupant locations in the blind zones of our sensor system and predict their destinations.Ph