150813 research outputs found
Sort by
Transformer-Based Prediction of Coronary Artery Lumen Expansion Post Angioplasty Using Optical Coherence Tomography
No full textCoronary artery disease is the leading cause of mortality globally, resulting in an urgent and critical need to better understand both vessel morphology and the processes of intervention. Angioplasty is an intervention which causes a previously constricted vessel to expand via placement of a stent, and is affected by numerous characteristics of the vessel such as calcium eccentricity and size, wall thickness, and prior lumen size. Being able to accurately assess whether a stent will properly expand allows cardiologists to pursue pre-stenting calcium lesion modification strategies that help avoid dangerous complications of improper stenting. This work introduces a pipeline for post-stenting lumen area prediction from pre-stenting optical coherence tomography (OCT) images. This pipeline includes morphological correction of OCT image segmentations, explainable feature extraction from OCT segmentations, and a predictive transformer network that combines morphological features with injected stent information. The aim is for such a pipeline to be used to support clinical decision making.M.Eng
Foundation Models for Protein Phenotype Prediction
No full textUnderstanding the roles of human proteins remains a major challenge, with approximately 20% of human proteins lacking known functions and more than 40% missing context-specific functional insights. Even well-annotated proteins are often poorly characterized in diverse biological contexts, disease states, and perturbations. We present ProCyon, a foundation model for modeling, generating, and predicting protein phenotypes across five interrelated knowledge domains: molecular functions, therapeutic mechanisms, disease associations, functional protein domains, and molecular interactions. To support this, we created ProCyon-Instruct, a dataset of 33 million protein phenotype instructions, representing a comprehensive resource for multiscale protein phenotypes. By co-training a large language model with multimodal molecular encoders, ProCyon integrates phenotypic and protein data. A novel architecture and instruction tuning strategy allow ProCyon to process arbitrarily interleaved proteinand-phenotype inputs, achieve zero-shot task transfer, and generate free-form text phenotypes interleaved with retrieved protein sequence, structure, and drug modalities in a single unified model.S.M
Deep-learning models for forecasting financial risk premia and their interpretations
No full textThe measurement of financial risk premia, the amount that a risky asset will outperform a risk-free one, is an important problem in asset pricing. The noisiness and non-stationarity of asset returns makes the estimation of risk premia using machine learning (ML) techniques challenging. In this work, we develop ML models that solve the problems associated with risk premia forecasting by separating risk premia prediction into two independent tasks, a time series model and a cross-sectional model, and using neural networks with skip connections to enable their deep neural network training. These models are tested robustly with different metrics, and we observe that our models outperform several existing standard ML models. A known issue with ML models is their ‘black box’ nature, i.e. their opaqueness to interpretability. We interpret these deep neural networks using local approximation-based techniques that provide explanations for our model's predictions
A biomimetic chip to assess subcutaneous bioavailability of monoclonal antibodies in humans
No full textSubcutaneous (subQ) injection is a common route for delivering biotherapeutics, wherein pharmacokinetics is largely influenced by drug transport in a complex subQ tissue microenvironment. The selection of good drug candidates with beneficial pharmacokinetics for subQ injections is currently limited by a lack of reliable testing models. To address this limitation, we report here a Subcutaneous Co-Culture Tissue-on-a-chip for Injection Simulation (SubCuTIS). SubCuTIS possesses a 3D coculture tissue architecture, and it allows facile quantitative determination of relevant scale independent drug transport rate constants. SubCuTIS captures key in vivo physiological characteristics of the subQ tissues, and it differentiates the transport behavior of various chemically distinct molecules. We supplemented the transport measurements with theoretical modeling, which identified subtle differences in the local absorption rate constants of seven clinically available mAbs. Accounting for first-order proteolytic catabolism, we established a mathematical framework to assess clinical bioavailability using the local absorption rate constants obtained from SubCuTIS. Taken together, the technology described here broadens the applicability of organs-on-chips as a standardized and easy-to-use device for quantitative analysis of subQ drug transport
Mixed-Variable Bayesian Optimization using Prior-Data Fitted Networks
No full textBayesian optimization (BO) is a powerful framework for optimizing expensive blackbox functions, widely used in domains such as materials science, engineering design, and hyperparameter tuning. Traditional BO relies on Gaussian processes (GPs) as surrogate models, but GPs face limitations in flexibility and scalability. Prior-Data Fitted Networks (PFNs) have recently emerged as a promising alternative, leveraging transformer architectures and in-context learning to approximate posterior predictive distributions (PPDs) in a single forward pass. By training on large amounts of synthetically generated data from sample-able function priors, PFNs can learn to rapidly predict PPDs across a wide range of function classes. In this thesis, we investigate the application of PFNs to mixed-variable BO, a particularly challenging setting due to the interplay between continuous and discrete inputs and the combinatorial complexity of the search space. We evaluate how PFNs perform when integrated with a range of mixed-variable BO strategies, including various encoding schemes and discrete-aware acquisition optimization. Additionally, we explore how finetuning PFNs on targeted function priors can enhance performance when prior knowledge about the objective is available. Our contributions include empirical evaluations of mixed-BO techniques, insights into PFN training, and a suite of mixed-variable benchmark problems.M.Eng
Enhancing resilience with natural growth targeting
No full textDespite a number of helpful changes, including the adop-tion of an inflation target, the Fed's monetary policy strat-egy proved insufficiently resilient in recent years. Whilethe Fed eased policy appropriately during the pandemic,it fell behind the curve during the post-pandemic recov-ery. During 2021, the Fed kept easing policy while theinflation outlook was deteriorating and the economy wasgrowing considerably faster than the economy's naturalgrowth rate—the sum of the Fed's 2% inflation goal andthe growth rate of potential output. The resilience of theFed's monetary policy strategy could be enhanced, andsuch errors be avoided with guidance from a simple natu-ral growth targeting rule that prescribes that the federalfunds rate during each quarter be raised (cut) when pro-jected nominal income growth exceeds (falls short) of theeconomy's natural growth rate. An illustration with real-time data and forecasts since the early 1990s shows thatFed policy has not persistently deviated from this simplerule with the notable exception of the period coincidingwith the Fed's post-pandemic policy error
Biomechanical Validation of Skeletal Tracking Data and Developing Action Recognition Models for Basketball: A Baseline for NBA Officiating Tools
No full textOptical tracking technology in sports has advanced rapidly in recent years, enabling new opportunities for data-driven analysis and tools to enhance the game. This study presents a framework for processing and analyzing a new skeletal tracking dataset collected from NBA basketball games. The methodology includes biomechanical joint validation, anomaly detection, and region-based consistency analysis to assess the integrity of player motion data. Joint movement anomalies are used to detect tracking errors, while court region and stadium-level evaluations help identify where the optical tracking system may be underperforming. These patterns can guide data providers toward specific areas that require refinement, offering a clearer starting point for improving system accuracy. After cleaning the dataset of 117 NBA games, two action recognition models—a transformer-based model and a temporal graph neural network—are implemented to classify player actions, specifically dribbling, passing, shooting, and rebounding, from sequences of skeletal tracking frames. The objective is to establish a baseline for developing tools to support officiating decisions in the NBA. By leveraging spatiotemporal representations of joint motion, this work improves the reliability of skeletal tracking data and contributes to the advancement of automated decision support in professional sports officiating.M.Eng
Race Car Reverse Gear Design
No full textThis thesis presents the design, simulation, and installation of a reverse gear system for the RUSH SR, a lightweight, motorcycle-engine-powered race car that lacks built-in reverse capability. The proposed solution repurposes a high-torque automotive starter motor to drive the car in reverse through engagement with a custom ring gear on the rear differential. Analytical modeling and time-domain simulation were used to evaluate performance under varying loads, including the effect of incline angle on terminal velocity and motor current draw. Simulated results show that the system can reliably move the car in reverse on slopes up to 10° before stalling, with current draw remaining within safe operational limits. The mechanical design includes a new differential carrier, gear coupler, and ring gear, while the electrical system explores both off-the-shelf and custom high-side switching controllers to manage power and solenoid activation. The final hardware was bench tested and installed on a working vehicle. Recommendations for future validation include current-limited incline testing and dynamic vehicle response trials. This modular and cost-effective system demonstrates a practical solution to a common race car limitation while preserving the RUSH SR’s lightweight performance characteristics.S.B
Interdependence of Solvent and Catalyst Selection on Low Pressure Hydrogen-Free Reductive Catalytic Fractionation
No full textHydrogen-free reductive catalytic fractionation (RCF) is a promising method to produce aromatic compounds directly from native biomass without the use of external hydrogen gas. In this work, we show that by using high boiling point diols as a solvent in hydrogen-free RCF, reaction pressures can be reduced by an order of magnitude compared to conventional RCF with methanol and hydrogen gas, while still producing appreciable aromatic monomer yields. Importantly, the use of diols with secondary alcohol functional groups increases hydrogenation activity on Ru/C, Pt/C, and Ni/C, measured by the yield of aromatic compounds with saturated propyl side chains, compared to processing in ethylene glycol, indicating that the choice of solvent and catalyst together can be tuned to control product selectivity of aromatic monomers in RCF
Private, Verifiable, and Auditable AI Systems
No full textThe growing societal reliance on artificial intelligence necessitates robust frameworks for ensuring its security, accountability, and trustworthiness. This thesis addresses the complex interplay between privacy, verifiability, and auditability in modern AI, particularly in foundation models. It argues that technical solutions that integrate these elements are critical for responsible AI innovation. Drawing from international policy contributions and technical research to identify key risks in the AI pipeline, this work introduces novel technical solutions for critical privacy and verifiability challenges. Specifically, the research introduces techniques for enabling verifiable and auditable claims about AI systems using zero-knowledge cryptography; utilizing secure multi-party computation and trusted execution environments for auditable, confidential deployment of large language models and information retrieval; and implementing enhanced delegation mechanisms, credentialing systems, and access controls to secure interactions with autonomous and multi-agent AI systems. Synthesizing these technical advancements, this dissertation presents a cohesive perspective on balancing privacy, verifiability, and auditability in foundation model-based AI systems, offering practical blueprints for system designers and informing policy discussions on AI safety and governance.Ph.D