DSpace@RPI (Rensselaer Polytechnic Institute)
Not a member yet
6809 research outputs found
Sort by
Leveraging Large Language Models for Mental Health Prediction via Online Text Data
No full textThe recent technology boost of large language models (LLMs) has empowered a variety of applications. However, there is very little research on understanding and improving LLMs' capability for the mental health domain. In this work, we present the first comprehensive evaluation of multiple LLMs, including Alpaca, Alpaca-LoRA, and GPT-3.5, on various mental health prediction tasks via online text data. We conduct a wide range of experiments, covering zero-shot prompting, few-shot prompting, and instruction finetuning. The results indicate the promising yet limited performance of LLMs with zero-shot and few-shot prompt designs for mental health tasks. More importantly, our experiments show that instruction finetuning can significantly boost the performance of LLMs for all tasks simultaneously. Our best-finetuned model, Mental-Alpaca, outperforms GPT-3.5 (25 times bigger) by 16.7\% on balanced accuracy and performs on par with the state-of-the-art task-specific model. We summarize our findings into a set of action guidelines for future researchers, engineers, and practitioners on how to empower LLMs with better mental health domain knowledge and become an expert in mental health prediction tasks
Synthesis of ionomers for electrochemical energy conversion
Get PDFMay 2022School of ScienceDue to the ever‐rising atmospheric carbon dioxide levels, there has recently been significant interest in carbon neutral technologies that can convert, store and release energy. Many of these promising technologies, such as fuel cells, electrolyzers, and flow batteries, utilize a polymeric material (ionomer) that allows the transport of ions between electrodes while simultaneously acting as a physical barrier to fuel and current. Perfluorosulfonic acid (PFSA) polymers are the most widely used ionomers for these electrochemical devices due to their excellent proton conductivity and chemical stability. However, PFSA materials lack synthetic flexibility to tune the properties of the ionomer to meet the unique operating conditions of each electrochemical device. To address this issue, hydrocarbon ionomers are explored as a viable alternative to meet the specific operating conditions required of electrochemical devices due to their broad range of polymer architectures and functionalities available to enhance the ionomer performance in the intended device. In this work, hydrocarbon ionomers are synthesized and developed to address the shortcomings of PFSA materials and elucidate the relationships of polymer structures and their conductivity, mechanical properties, and oxidative stability in environments of electrochemical devices.Ph
Modeling and improving Single-Event burnout performance from heavy ion bombardment in high-voltage 4H-SiC power devices
No full textDecember 2021School of EngineeringSilicon carbide (SiC) is a wide bandgap semiconductor recently used for developing high-voltage devices for power electronics applications on Earth and in outer space. Radiation is a serious concern for applications in the upper atmosphere and space since it can lead to Single-Event Effects (SEEs) such as permanent performance degradation, or the catastrophic device failure known as Single-Event Burnout (SEB). Despite that, SiC power semiconductor devices were expected to have improved radiation tolerance over silicon (Si) counterparts owing to its three times larger bandgap and higher operation temperature limits. Experimental studies demonstrate that SiC power devices fail at less than 50 percent of their rated blocking voltage when exposed to a single heavy ion strike and the results are similar to Si MOSFETs but lower than Si diodes. This thesis explores the physics behind failure of SiC power devices subjected to heavy ion bombardment and structural improvements to enhance SiC power device SEB performance. Physically realistic simulations are needed to understand the physics more fully and accurately capture the electro-thermal response inside a power device during and after a heavy ion strike. A high-fidelity Monte-Carlo radiation transport code was deployed to model the heavy ion strike and the corresponding generation of electron-hole pairs. Additionally, the electro-thermal effects from the ion strike were simulated using a full 3-D time-dependent electro-thermal device simulator. A toolkit was developed, enabling the radiation data generated from the Monte-Carlo simulation to be utilized by the electro-thermal device simulator, thus linking the two simulation components. The power devices investigated include the JBS diode and the MOSFET. These devices were selected for both their ubiquitousness and criticality in many power circuits. We have calibrated and validated our models against experimental studies. The failure observed in our commercial JBS and MOSFET device simulations from a heavy ion strike resulted from the generated electron-hole pairs enhancing the electric field at the epitaxial and substrate layer interface. The electric field at this interface exceeds 3 MV/cm and generates additional carriers through impact ionization. These carriers cause highly localized heating of the crystalline lattice through Joule heating. Eventually, a mesoplasma forms and the lattice temperature exceeds 3000 K, the sublimation temperature of SiC. The simulated SEB threshold voltage for these devices is 525 and 650 V for the MOSFET and JBS diode, respectively. Traditionally the ratio of the SEB threshold voltage and breakdown voltage (SEB/BV) is the figure of merit (FoM) used to evaluate how robust a design is against a heavy ion strike. A ratio of 1 means that the design is immune to SEB. The simulated MOSFET and JBS diode have a SEB/BV ratio of 0.30 and 0.38, respectively. We developed new FoMs which consider both the on-state and SEB performance for the MOSFET and JBS diode. The values obtained using these FoMs are 0.12 and 0.29 for the simulated MOSFET and JBS diode, respectively. Several designs to improve SEB performance by suppressing the electric field enhancement at the epitaxial and substrate interface were proposed and confirmed with simulations. Using these designs, the SEB threshold voltage was increased between 46 and 81 percent and the on-performance tradeoff was between 1 to 53 percent over the baseline commercial designs. These designs also increases the SEB/BV ratio by at least a factor of 1.26 over the baseline designs and our new FoM by at least a factor of 1.27. We propose several device structures using these designs to create 1200 V SEB rated power devices for both the JBS diode and MOSFET. We also evaluated the SEB performance of Superjunction (SJ) planar MOSFETs, which have a superior specific on-resistance over conventional designs but experience an SEB/BV ratio that is 46 percent lower compared to conventional non-SJ MOSFETs of similar drift layer thickness. However, employing our new FoM, the SJ MOSFET outperforms the commercial design by a factor of 7. Further SEB performance improvement can be realized by employing a Semi-SJ design, which results in a SEB/BV ratio that is 89 percent higher, the specific on-resistance increases less than 5 percent, and the FoM is 74 percent higher compared to the SJ MOSFET.Ph
Biobased colloids for food preservation and fragrances microencapsulation
No full textAugust 2021School of ScienceColloids exhibit properties intermediate between suspensions and solutions therefore enabling their unique applications in foods, pharmaceuticals, cosmetics, coatings, water treatment and more. Emulsions are one of the most important type of colloids. As meta stable dispersions, emulsions must be stabilized with emulsifying agents such as surfactants to introduce kinetic stability to combat the thermodynamic instability. As the use of synthetic surfactants has drawn increasing concerns for their environmental and health implications, biosurfactants, especially sophorolipids and their derivatives, stand out due to their unique properties such as the ability to be produced at high titers (>300 g/L), antimicrobial activities and low toxicity. The first portion of the thesis is based on the development of biobased colloidal systems using sophorolipid derivatives for food preservation. Sophorolipid butyl ester (SLBE), which is naturally derived and has the potential to both stabilize o/w emulsions and provide antimicrobial activity, is used as the surfactant for preparing hydrocolloid-based coatings/films for potential applications of food packaging. The interactions of SLBE with different oil (oregano/olive oil) mixtures and biopolymers [γ-poly(glutamic acid), γ-PGA, and chitosan, CH] was investigated. During this study, it was revealed that SLBE stabilizes o/w emulsions over 30 days with oil concentrations at least 10 times that of the surfactant. Furthermore, CH is a highly effective emulsion stabilizer with apparent favorable interactions with SLBE. These promising results of SLBE-CH interactions can be used as a platform for developing antimicrobial films/coatings for food preservation. The second portion of the thesis is based on the development of oil encapsulation systems from biobased and safe precursor molecules. Microencapsulation is widely used to encapsulate solid, liquid, and gaseous substances to prevent evaporation of volatile compounds and to achieve controlled release; however, there is increased pressure on microcapsule manufacturers to develop encapsulation systems that follow green chemistry principles and use safe biobased materials. Here, we describe interfacial polymerization encapsulation of a fragrance oil. Diphenolic acid (DPA), prepared from cellulose-derived levulinic acid and phenol, was used as a replacement for bisphenol-A. The resulting diepoxy resin DGEDP-esters were incorporated in the fragrance oil phase and, upon reaction with the aliphatic diamine hexamethylenediamine (HMDA) or chitosan oligosaccharides (COS), formed capsule walls. The emulsifiers, gum arabic (GA), polyvinyl alcohol, sophorolipid butyl ester and polyvinyl pyrrolidone/polyquaternium-11 cosurfactant system were investigated to determine their ability to support emulsion formation and interfacial polymerizations. GA at 2.5 wt% stands out among the four emulsifiers as it provides both good emulsion stability and capsule formation. The effect of DGEDP-methyl ester/HMDA ratios on various capsule properties was studied. The preferred HMDA concentration is 1 wt% since there is a sufficient concentration of HMDA to react with DGEDP-methyl ester at interfaces to create capsule walls with a relatively high crosslinking density without bridging of oil capsules to form aggregates. Furthermore, DGEDP-esters with various structures were synthesized and used as oilsoluble monomers in interfacial polymerizations to study their structure-property relationships. Use of DGEDP-ME and DGEDP-MG led to the formation of compactly crosslinked capsules walls with high microencapsulation efficiency (EE > 95%). Besides, COS cured capsules exhibit good acid stability compared to HMDA cured capsules. These observations demonstrate that DGEDPester/HMDA and DGEDP-ester/COS are promising biobased alternatives for isocyanate and formaldehyde approaches for oil encapsulation.Ph
Auto-Transfer: Learning to Route Transferrable Representations
No full textKnowledge transfer between heterogeneous source and target networks and tasks has received a lot of attention in recent times as large amounts of quality labeled data can be difficult to obtain in many applications. Existing approaches typically constrain the target deep neural network (DNN) feature representations to be close to the source DNNs feature representations, which can be limiting. We, in this paper, propose a novel adversarial multi-armed bandit approach that automatically learns to route source representations to appropriate target representations following which they are combined in meaningful ways to produce accurate target models. We see upwards of 5% accuracy improvements compared with the state-of-the-art knowledge transfer methods on four benchmark (target) image datasets CUB200, Stanford Dogs, MIT67, and Stanford40 where the source dataset is ImageNet. We qualitatively analyze the goodness of our transfer scheme by showing individual examples of the important features focused on by our target network at different layers compared with the (closest) competitors. We also observe that our improvement over other methods is higher for smaller target datasets making it an effective tool for small data applications that may benefit from transfer learning
Communication through multi-layered acoustic electric channels
No full textAugust 2015School of EngineeringSingle layered acoustic-electric channels have been used in the recent past to send
power and data through metallic barriers wirelessly. These penetration-free solutions
are valuable for maintaining the structural integrity while providing needed connec-
tivity. Most of the previous work has considered single-layered channels, though
in many applications one will encounter multi-layered channels that include one or
more metallic barriers and fluid layers like water etc. This work presents communi-
cation schemes that can be used to send data through such multi-layered acoustic-
electric channels. First, the measured characteristics of several multi-layered acous-
tic electric channels are presented and modeled, including those consisting water
sandwiched between steel plates and concentric pipes as well as those that extend
axially along a steel pipe. The channels are found to be very frequency selective.
Second, low data rate, low complexity communication techniques are developed for
these channels. Chirp-FSK and Chirp-OOK with non-coherent detection are studied
through theoretical analysis and Monte-Carlo simulations using measured channel
data. Chirp-OOK with energy detection is found to provide a good compromise be-
tween performance and implementation simplicity. Hardware implementations are
designed, constructed and tested on the actual channels. A standalone embedded
design of the communication link is used to send at a rate of 100 bps using 5 mW of
transmit power. Lastly, communication schemes to send data at higher rates (tens of
kbps) through such multi-layered channels are considered. One such scheme, using
a Decision Feedback Equalizer with 16-QAM modulation is found to be effective for
one of the channels.Ph
Using Milkyway@home to measure the mass of the orphan-chenab stream progenitor dwarf galaxy
No full textAugust 2022School of ScienceWe fit the mass and radial profile of the Orphan-Chenab Stream's (OCS) dwarf galaxy progenitor by using turnoff stars in the Sloan Digital Sky Survey (SDSS) and the Dark Energy Camera (DEC) to constrain N-body simulations of the OCS progenitor falling into the Milky Way on the 1.5 PetaFLOPS MilkyWay@home distributed supercomputer. We infer the internal structure of the OCS's progenitor under the assumption that it was a spherically symmetric dwarf galaxy comprised of a stellar system embedded in an extended dark matter halo. We optimize the evolution time, the baryonic and dark matter scale radii, and the baryonic and dark matter masses of the progenitor using a differential evolution algorithm. The likelihood score for each set of parameters is determined by comparing the simulated tidal stream to the angular distribution of OCS stars observed in the sky. We fit the total mass of the OCS's progenitor to () with a mass-to-light ratio of and () within 300 pc of its center. Within the progenitor's half-light radius, we estimate a total mass of (). We also fit the current sky position of the progenitor's remnant to be and show that it is gravitationally unbound at the present time. The measured progenitor mass is on the low end of previous measurements, and if confirmed lowers the mass range of ultrafaint dwarf galaxies. Our optimization assumes a fixed Milky Way potential, OCS orbit, and radial profile for the progenitor, ignoring the impact of the Large Magellanic Cloud (LMC). Using second-order forward automatic differentiation, we also attempt to computationally determine the systematic errors introduced from the fixed orbit, gravitational potential, and lack of an LMC. This paper describes the methods employed to implement automatic differentiation in areas of our code where derivative information is not propagated, such as through random number generation and discrete histogram binning. We find that due to the turbulent and chaotic behavior of our searchable likelihood surface, the systematic errors derived through automatic differentiation are severely overestimated. Recommendations for future work to estimate the systematic errors are provided.Ph
Privacy and quality of service aware edge network design
No full textMay 2022School of EngineeringAs the network infrastructure grows increasingly more capable of supporting large throughput, other factors such as privacy and latency remain increasingly important in network design. In this thesis, we target two trade-offs: The Privacy-Rate-Memory trade-off in the edge caching systems and the delivery-latency trade-off in transport layer protocol design. In scenarios where one server uses a multicast channel to communicate with multiple cache-enabled users, coded caching schemes achieve a low server transmission rate by jointly optimizing both the cache content placement of each user cache and the server transmission. We observe that the coded caching scheme achieves a low rate but sacrifices user privacy. An attacker can gain knowledge about user file requests by analyzing server transmissions. This observation is then formally formulated in this thesis as feasible attack algorithms that predict user file requests by either an eavesdropper or colluding users. To address the privacy leakage in coded caching that is noted above, we present two new coded caching schemes that protect user privacy from potential attackers who may be the server or colluding users. Firstly, we propose a scheme that splits the user cache into a private portion to conduct uncoded caching and a public portion to conduct the coded caching scheme. Since there is no privacy issue for uncoded caching, the user protects its privacy by losing some gain in rate. Secondly, we propose another scheme that turns the one-to-one mapping between user file requests and the server transmission into a one-to-many mapping. Therefore, the attacker cannot reversely predict the user file request from the known server transmission. The new mapping is achieved by grouping multiple files into a big file and conducting a coded caching scheme with different file sizes. For both schemes, we study the associated Privacy-Rate-Memory trade-off (PRM). We next consider the design of a quality-of-service aware transport layer for delay and packet loss sensitive applications. We first present the design and performance evaluation of a new Double Q-learning Network based transport layer protocol called RCP-DQN. Unlike prior transport layer protocols that aim at either providing guaranteed end-to-end delivery, e.g., TCP, or minimizing the end-to-end delay, e.g., UDP, RCP-DQN aims to optimize any achievable combination of these objectives, as specified by an application layer utility function. The considered utility function can be quite general and can capture a combination of delay and packet delivery metrics. RCP-DQN can be thought of as an intelligent middle-ground between UDP and TCP that maps application layer objectives subject to what is learned about the network state. It is window-based like TCP, but retransmissions are decided based on a reinforcement learning algorithm to maximize the application layer utility function. RCP-DQN employs a reinforcement learning method, a double Q-Learning network, to learn the best strategy in real-time from a history of packet (re)transmission experiences. No assumption on the shape of the utility function is needed. RCP-DQN is evaluated under a wide range of network settings and is found to outperform UDP, TCP, and ARQ for almost all settings. The performance is also evaluated with respect to TCP-friendliness and network stability. The superior performance of RCP-DQN motivates a systematic analysis of the packet Retransmission Control Problem (RCP). We formulate the RCP problem as a semi-Markov Decision Problem (SMDP) by requiring the utility function to be a time-discounted utility function. Under the SMDP setup, we study the optimal policy (ORCP) and the achieved reward. By making use of the structure of the RCP problem, we show that the optimal policy can be simplified to a control-limit policy (CRCP). We prove that the control-limit policy is still optimal and derive bounds on the performance. For applications, we present a Control-Limit policy with Estimation algorithm (CERCP) by adding network status and system performance estimations. After that, we focus on the relaxation of the requirement on the shape of the utility function. Motivated by the optimal policy, we presented a Q-Learning based policy (QRCP), which has much fewer constraints on the utility function. QRCP follows a customized Q table updating strategy and does not explicitly require the estimation of the network state and system performance. Both policies are evaluated under multiple network settings, and both of them outperform UDP, TCP and ARQ for almost all network settings.Ph
Elucidating the causal mechanisms of Alzheimer's disease using cell-based models
No full textAugust 2022School of EngineeringThe lack of a preventive therapy for neurodegenerative diseases, such as Alzheimer’s Disease (AD) has increased the urgency for gaining an understanding of the mechanisms that lead to cognitive decline and eventually death. The identification of several potential targets including the amyloid plaque, hyperphosphorylated tau tangles, microglial activation, and circadian rhythm disruption has led to the generation of various in vitro and in vivo models. In particular, the development of in vitro models of AD has been crucial in gaining a more complete understanding of AD pathology. Such in vitro models have been able to partly recapitulate key pathological hallmarks of the disease using various neuronal cell lines including induced pluripotent stem cells, immortalized neural progenitor cells and neuroblastoma cell lines in 2D and 3D cultures. It is important to elucidate further the interplay among the various disease mechanisms using cell-based models that can aid in improving the drug discovery and development process. In this thesis research, various potential mechanisms of Alzheimer’s Disease are addressed using in vitro 2D and 3D cell-based models. First, a stable AD cell line capable of overexpressing mutations in the Amyloid Precursor Protein (APP) associated with Familial AD was established, which included the Swedish double mutation (KM670/671NL) and the Indiana mutation (V717F). This cell line was then used in a cell-based assay platform to determine the efficacy of three test compounds capable of inhibiting the generation of A by targeting the transmembrane domain of APP. Moreover, it was shown that these compounds targeted the APP specifically while possessing minimal off-target activity against the function of the -secretase. This sequential assay platform can be adapted for high throughput screening of various compounds capable of targeting A inhibition. Second, an immortalized microglial cell line was used to elucidate the neuroinflammatory facet of AD. The pro-inflammatory response of BV2 microglia was characterized upon exposure to soluble A oligomers and compared with the response of the microglial cells upon exposure to A monomers. Exposure of specific chemokines and cytokines detected in an AD environment to BV2 cells altered the pro-inflammatory response into an anti-inflammatory response. These findings may be used to target modulation of microglial phenotypes in early AD. Finally, to study the effects of circadian regulation in vitro, a 3D spheroid model was established using HepG2 cells to understand further the influence of circadian regulation on genes involved in processes associated with drug metabolism, a highly integrated and complex biological mechanism. Bulk RNA sequencing was used to compare the effects of circadian synchronization in a 3D spheroid cell culture system with that of a more conventional 2D cell culture system. Using transcriptomic analysis, significant differences in circadian and drug metabolism genes were observed between the 3D and 2D culture formats. Moving forward, this 3D spheroid model may be adapted to developing neurospheroids to study the influence of circadian regulation in AD.Ph
Semi automated process for generating knowledge graphs for marginalized community doctoral-recipients.
No full textPurpose – This paper aims to describe the “InDO: Institute Demographic Ontology” and demonstrates the
InDO-based semiautomated process for both generating and extending a knowledge graph to provide a
comprehensive resource for marginalized US graduate students. The knowledge graph currently consists of
instances related to the semistructured National Science Foundation Survey of Earned Doctorates (NSF SED)
2019 analysis report data tables. These tables contain summary statistics of an institute’s doctoral recipients
based on a variety of demographics. Incorporating institute Wikidata links ultimately produces a table of
unique, clearly readable data.
Design/methodology/approach – The authors use a customized semantic extract transform and loader
(SETLr) script to ingest data from 2019 US doctoral-granting institute tables and preprocessed NSF SED
Tables 1, 3, 4 and 9. The generated InDO knowledge graph is evaluated using two methods. First, the authors
compare competency questions’ sparql results from both the semiautomatically and manually generated
graphs. Second, the authors expand the questions to provide a better picture of an institute’s doctoral-recipient
demographics within study fields.
Findings – With some preprocessing and restructuring of the NSF SED highly interlinked tables
into a more parsable format, one can build the required knowledge graph using a semiautomated
process.
Originality/value – The InDO knowledge graph allows the integration of US doctoral-granting institutes
demographic data based on NSF SED data tables and presentation in machine-readable form using a new
semiautomated methodology.Funded in part by RPI-IBM AI Research Collaboration, a member of the IBM AI Horizons network