80264 research outputs found
Sort by
Equity-Centered and Vision-Aligned Research: Strategies of a District Research Office to Build Better Research Collaborations
No full textThe Controlled Deformation and Assembly of Anisotropic Nanostructures
No full textThe desirable properties of inorganic nanostructures are, in large part, determined by their morphologies, whether it be their plasmon resonance, surface chemistry, or assembly behavior. Therefore, great effort has been expended over the last two decades into growing the library of available inorganic nanostructures, primarily through the development of more sophisticated synthetic techniques utilizing directing ligands and differing facet reactivity. Particularly sought after are methods to fabricate low symmetry nanostructures, as these frequently exhibit more exotic properties, such as optical chirality, or the ability to organize into highly anisotropic superlattices with unique crystal habits and lattice structures. The difficulty in fabricating low symmetry inorganic nanostructures originates in the inherent isotropy of the chemical conditions used to grow them, forcing these methods to rely on the inherent asymmetry of a given material’s crystal structure or that imposed by carefully chosen directing ligands. As such, these protocols are designed ad hoc for each material, leading to a gradual expansion of available morphologies over time.
The mechanical deformation of inorganic nanostructures is a promising route toward accessing novel low symmetry morphologies and a rapid expansion of the existing library. Mechanical deformation involves the imposition of an anisotropic force onto a nanostructure, leading to bending strain and a lower symmetry product. We have recently developed a method of inorganic nanostructure deformation wherein a template particle is used to direct the shape change. In this method, noble metal nanoplates are deposited onto spherical template particles and deform to their shapes due to cumulatively strong Van der Waals interactions across the large surface area of the nanoplate. In this work we demonstrate templated deformation of micron sized silver nanoplates over 15 nm template particles, characterizing their morphology through transmission electron microscopy and producing a Kirchhoff-Love theory derived analytical model to elucidate the relationship between nanoplate thickness and final deformation morphology and demonstrate that Van der Waals forces are sufficient to induce plastic deformation in the nanoplate. We extend this model to show the effect binding ligands have on the overall mechanical properties of thin silver nanoplates and demonstrate this method could be applied to other materials. We also develop a synthesis for high aspect ratio gold nanoplates and show that their deformation around template particles on the same order of size yields high morphological diversity in the products, as well as curvature control which can be exerted by altering the size of the template particle. Using a combination of cross-section TEM imaging, chemical nanoplate overgrowth experiments, and tomographic reconstruction, we show the nanoplates are deformed elastically and accurately measure their curvature landscapes. Finally, we develop a method for the functionalization of 2 nm wide and ~1 micron long ultra-thin gold nanowires, desirable for their simultaneous flexibility and electrical conductivity, with a stimulus responsive thiol ligand based on carboxyl group chemistry. The ultra-thin gold nanowires assemble and disassemble in response to specific chemical stimuli and can be assembled into macroscopic fibers, demonstrating preliminary work that could be continued to develop fibers of these functionalized ultra-thin gold nanowires as functional materials. Overall, this work breaks new ground on the use of deformable and flexible nanostructures in generating low symmetry morphological diversity and functional anisotropic superstructures
Use of topical methylene blue to image nuclear morphometry with a low-cost scanning darkfield microendoscope
No full textSignificanceFiber-optic microendoscopy is a promising approach to noninvasively visualize epithelial nuclear morphometry for early cancer and precancer detection. However, the broader clinical application of this approach is limited by a lack of topical contrast agents available for in vivo use.AimThe aim of this study was to evaluate the ability to image nuclear morphometry in vivo with a novel fiber-optic microendoscope used together with topical application of methylene blue (MB), a dye with FDA approval for use in chromoendoscopy in the gastrointestinal tract.ApproachThe low-cost, high-resolution microendoscope implements scanning darkfield imaging without complex optomechanical components by leveraging programmable illumination and the rolling shutter of the image sensor. We validate the integration of our system and MB staining for visualizing epithelial cell nuclei by performing ex vivo imaging on fresh animal specimens and in vivo imaging on healthy volunteers.ResultsThe results indicate that scanning darkfield imaging significantly reduces specular reflection and resolves epithelial nuclei with enhanced image contrast and spatial resolution compared to non-scanning widefield imaging. The image quality of darkfield images with MB staining is comparable to that of fluorescence images with proflavine staining.ConclusionsOur approach enables real-time microscopic evaluation of nuclear patterns and has the potential to be a powerful noninvasive tool for early cancer detection
Navigating the challenges of initiating pediatric device trials – a case study
No full textIntroduction:Pediatric medical devices lag behind adult devices due to economic barriers, smaller patient populations, changing anatomy and physiology of patients, regulatory hurdles, and especially difficulties in executing clinical trials. We investigated the requirements, challenges, associated timeline, and costs of conducting a multi-site pivotal clinical trial for a Class II pediatric physiologic monitoring device.Methods:This case study focused on the negotiation of clinical trial agreements (CTAs), budgets, and Institutional Review Board (IRB) processing times for a pediatric device trial. We identified key factors contributing to delays in clinical trial execution and potential best practices to expedite the process while maintaining safety, ethics, and efficacy.Results:The total time from site contact to first patient enrollment averaged 14 months. CTA and budget negotiations were the most time-consuming processes, averaging nearly 10 and 9 months, respectively. Reliance and local IRB processing also contributed significantly to the timeline, overall adding an average of 6.5 months across institutions. Nearly half of all costs were devoted to regulatory oversight. The COVID-19 pandemic caused significant slowdowns and delays at multiple institutions during study enrollment. Despite these pandemic-induced delays, it is important to note that the issues and themes highlighted remain relevant and have post-pandemic applicability.Conclusions:Our case study results underscore the importance of establishing efficient and standardized processing of CTAs, budget negotiations, and use of reliance IRBs to expedite clinical trial execution for pediatric devices. The findings also highlight the need for a national clinical trials network to streamline the clinical trial process
Lossy Computation For Large-Scale Machine Learning
No full textIn recent years, machine learning (ML), particularly deep learning, has made significant strides in areas like image recognition and language processing. It's been shown that more parameters and data can greatly boost ML model performance. However, the growth in model and data size is outpacing hardware capabilities, leading to a gap between ML needs and hardware development. My research is aimed at creating scalable ML algorithms and systems to meet current and future ML demands, exploring methods like randomized and low-precision computations to handle larger data and model sizes without changing hardware.
First, for dealing with large datasets, such as in analyzing molecular structures or social networks where data is interconnected, Graph neural networks (GNNs) have recently emerged as one of the de-facto standard tools to analyze the graph data. Leveraging the message passing mechanism, GNNs learn the representation of each node by iteratively aggregating information from its neighbors to capture of graph structures and relationships. However, the key challenges in graph representation learning is the scalability issue as the real-word graphs may contain more than billions of nodes, resulting in significant memory and speed inefficiency when training GNNs on huge graphs. To address the challenges of memory and time inefficiency in large-scale graph learning, we introduce two lossy computation paradigms. First, we propose a memory-efficient framework for training GNNs with significantly compressed activations. Second, we present a time-efficient GNN training method with degree-based graph sparsification.
Second, regarding the challenge of handling large models, as the model size grows, large language models (LLMs) have exhibited human-like conversation ability. This advancement opens the door to a wave of new applications, such as custom AI agents. To achieve this, two essential steps are involved: fine-tuning and serving. Fine-tuning is the process of adapting the LLM to a specific task, such as understanding and responding to domain-specific inquiries. The second step, serving, is about generating outputs to the questions in real-time. However, both of these two steps are hard and expensive due to the large model scale, limiting their accessibility to most of the users. Similarly, to improve efficiency in fine-tuning and serving LLMs, we also employ lossy computation approaches. Our first method enhances memory efficiency in LLM fine-tuning through the use of randomized matrix multiplication. Our second approach introduces a prompt tuning framework that optimizes the accuracy-efficiency trade-off for compressed LLMs.
Lastly, we implement an extreme low-bit quantization technique for the KV Cache to further enhance performance
Impact and characterization of serial structural variations across humans and great apes
No full textModern sequencing technology enables the systematic detection of complex structural variation (SV) across genomes. However, extensive DNA rearrangements arising through a series of mutations, a phenomenon we refer to as serial SV (sSV), remain underexplored, posing a challenge for SV discovery. Here, we present NAHRwhals (https://github.com/WHops/NAHRwhals), a method to infer repeat-mediated series of SVs in long-read genomic assemblies. Applying NAHRwhals to haplotype-resolved human genomes from 28 individuals reveals 37 sSV loci of various length and complexity. These sSVs explain otherwise cryptic variation in medically relevant regions such as the TPSAB1 gene, 8p23.1, 22q11 and Sotos syndrome regions. Comparisons with great ape assemblies indicate that most human sSVs formed recently, after the human-ape split, and involved non-repeat-mediated processes in addition to non-allelic homologous recombination. NAHRwhals reliably discovers and characterizes sSVs at scale and independent of species, uncovering their genomic abundance and suggesting broader implications for disease
Objective Speech-Based Sociability Measure for Mental Health Assessment
No full textEMBARGO NOTE: This item is embargoed until 2026-05-01Sociability measures play a pivotal role in mental health assessment due to their associations with mood and symptoms of mental disorders. The conventional approach for assessing sociability relies on self-reports, often biased and error-prone. To address these limitations, I propose Ambiance-aware Social Interaction Measure (ASIM), an objective, fine-grained and comprehensive sociability measure. ASIM captures both individual social interactions and social ambiance – the environmental elements impacting social engagements – through the analysis of unconstrained audio data. Specifically, ASIM is structured as an 8-dimensional vector, with each element originating from proxies of social ambiance and social interactions. The number of concurrent speakers is proposed as a proxy for social ambiance, while a dedicated target speaker detection algorithm is devised to capture individual social interactions. To suit diverse usage scenarios, I present both offline processing and on-device processing solutions. In the offline processing scenario, concurrent speaker count benefits from the knowledge of large-scale self-supervised representations through model fine-tuning, and target speaker detection incorporates unsupervised source separation as a pre-processing step. For on-device processing, the concurrent speaker count is compressed to 5% of its original size to facilitate device-side implementation. Additionally, an Android application has been developed to support data collection, processing, and real-time feedback. Furthermore, target speaker detection leverages the enroll-aware attention statistic pooling method to effectively eliminate interfering speakers and enhance the model's robustness. ASIM's performance is evaluated using benchmark datasets and Rice University campus data, followed by an in-depth error analysis to understand its limitations and utilization. ASIM's utility is further substantiated through application to a clinical dataset. Results show a significant correlation between sociability patterns extracted by ASIM and self-reported metrics. Additionally, ASIM proves effective in predicting self-reported mood and sociability, showcasing its potential as an instrumental tool in both mental health research and clinical settings
A new diagonalization based method for parallel-in-time solution of linear-quadratic optimal control problems
No full textA new diagonalization technique for the parallel-in-time solution of linear-quadratic optimal control problems with time-invariant system matrices is introduced. The target problems are often derived from a semi-discretization of a Partial Differential Equation (PDE)-constrained optimization problem. The solution of large-scale time dependent optimal control problems is computationally challenging as the states, controls, and adjoints are coupled to each other throughout the whole time domain. This computational difficulty motivates the use of parallel-in-time methods. For time-periodic problems our diagonalization efficiently transforms the discretized optimality system into nt (=number of time steps) decoupled complex valued 2ny × 2ny systems, where ny is the dimension of the state space. These systems resemble optimality systems corresponding to a steady-state version of the optimal control problem and they can be solved in parallel across the time steps, but are complex valued. For optimal control problems with initial value state equations a direct solution via diagonalization is not possible, but an efficient preconditioner can be constructed from the corresponding time periodic optimal control problem. The preconditioner can be efficiently applied parallel-in-time using the diagonalization technique. The observed number of preconditioned GMRES iterations is small and insensitive to the size of the problem discretization
Deep Learning for Healthcare: Empirical Studies and Methodological Advances in Physiological Time Series Data
No full textHuman physiological time series data, including electrocardiograms (ECGs) and wearable sensors data, provide essential information for healthcare applications. The inherent complexity and variability of these signals introduce both challenges and opportunities for deep learning research. This thesis systematically addresses several critical aspects of physiological time series analysis:
(1) We begin by empirically studying the behavior of the "Mixup" family of data augmentation methods, traditionally employed in computer vision, in the context of physiological time series. Through extensive experiments, we verify the unique properties of Mixup in this domain, extending known techniques to new applications.
(2) We introduce a novel two-stage curriculum learning approach for seizure forecasting, a critical task in epilepsy treatment. By employing an autoencoder-based scoring stage followed by a subsequent fine-tuning stage involving a pre-trained encoder and a ranked sample pool, we address the challenge of class imbalance in forecasting anomaly events. This scheduled sampling scheme demonstrates superior performance and faster model convergence in seizure prediction.
(3) We present a Transformer-based model specifically designed for ECG data analysis. Drawing inspiration from the inherent physiological components of ECG signals, the model adopts a representation learning approach, segmenting ECG signals into distinct fragments or "tokens." These tokens, characterized by physiological significance such as P/QRS/T components in the waveforms, are processed by the Transformer. Experiments demonstrated competitive performance in the arrhythmia classification task. This approach ensures that the model's learning process aligns closely with clinical insights, enhancing its interpretability.
We evaluate the proposed methods on various datasets, including standard 12-lead ECG signals for project 1 and 3, multiple biomedical time series datasets for project 1, and wristband-recorded physiological signals for project 2. The evaluations show promising results and significant improvements compared to baseline models. %Together,
Collectively, the projects form a cohesive narrative that reflects a progression from empirical exploration to the design of deep learning models and schemes, all centered around the unique challenges and opportunities of physiological time series modeling
Millimeter-sized battery-free epidural cortical stimulators
No full textRefractory neurological and psychiatric disorders are increasingly treated with brain stimulation therapies using implanted neuromodulation devices. Current commercially available stimulation systems, however, are limited by the need for implantable pulse generators and wired power; the complexity of this architecture creates multiple failure points including lead fractures, migration, and infection. Enabling less invasive approaches could increase access to these therapies. Here we demonstrate the first millimeter-sized leadless brain stimulator in large animal and human subjects. This Digitally programmable Over-brain Therapeutic (or DOT) is approximately 1 cm in width yet can produce sufficient energy to stimulate cortical activity on-demand through the dura. This extreme miniaturization is possible using recently developed magnetoelectric wireless power transfer that allows us to reach power levels required to stimulate the surface of the brain without direct contact to the cortical surface. This externally powered cortical stimulation (XCS) opens the possibility of simple minimally invasive surgical procedures to enable precise, long-lasting, and at-home neuromodulation with tiny implants that never contact the surface of the brain