Organizational Unit:
School of Computational Science and Engineering
School of Computational Science and Engineering
Permanent Link
Research Organization Registry ID
Description
Previous Names
Parent Organization
Parent Organization
Organizational Unit
Includes Organization(s)
ArchiveSpace Name Record
Publication Search Results
Now showing
1 - 4 of 4
-
ItemAn ensemble-based data assimilation approach to the simulation and reconstruction of chaotic cardiac states(Georgia Institute of Technology, 2023-05-02) Badr, ShoaleComplexities in time-dependent real-world systems pose several difficulties when forecasting their future dynamics. Advancements in the field of meteorology, with the purpose of improving weather forecasting (which behaves chaotically), over the last few decades have led to the development of data assimilation, which is a technique that combines predictive numerical mathematical models with real measurements, or observations, to form more refined estimates of system states over time. As reconstruction of chaos in the tissue of the heart presents a similar forecasting problem, we apply data assimilation to the cardiac domain in this thesis. Within the assimilation algorithm, we use three widely-known mathematical cardiac models tuned to produce specific types of complex cardiac electrical dynamics, including stable spiral waves and spiral wave breakup, corresponding to tachy- cardia and fibrillation, respectively. We generate synthetic observations from each model by subsampling their solutions in space and time and restricting utilizing only one variable representing voltage, then adding Gaussian noise, and use the resulting datasets to test our implementation. By leveraging the public availability of data assimilation filtering algorithms (primarily Kalman filters) through the Parallel Data Assimilation Framework (PDAF) and adding extensions necessary for the cardiac setting, we present how two- dimensional chaotic electro-cardiac voltage behavior can be reconstructed with ensemble-based data assimilation in the presence of several experimental conditions including noise, sparse observations, and model error. This thesis presents the first application, to our knowledge, of ensemble Kalman filtering to the reconstruction of complex cardiac electrical dynamics in the 2-D domain. We found that the Error Subspace Transform Kalman Filter (ESTKF) we used is sensitive to model error and the frequency at which states are assimilated (assimilation interval). We also propose several possible improvements that can be made to our assimilation system so that it may improve state reconstruction accuracy. These preliminary findings suggest promising future experimental results, both using synthetic observations (with different model dynamics initialization) and with true experimental data.
-
ItemNEWS DATA VISUALIZATION INTERFACE DEVELOPMENT USING NMF ALGORITHM(Georgia Institute of Technology, 2022-05-03) Ahn, ByeongsooNews data is a super large-scale dataset. It covers a wide range of topics ranging from heavy topics such as politics and society to beauty and entertainment, relatively light topics. At the same time, it is also the most accessible source of information for the general public to obtain information. Thus, how is this large amount of data used by the general public being utilized? Currently, services provided by news platforms are just full article searches and related news recommendations. It uses only a fraction of the vast news dataset, and there is still a lack of systems to fully utilize and analyze it. As mentioned above, news datasets which contain a wide range of topics and super large scales of data, record everything that happened in the past and present, so analyzing and visualizing them can track how trends in real-world change over time and even discover what the topics of the large dataset are without reading the full text through topic modeling. For this objective, in this thesis, we propose a novel interactive visualization interface for the news data based on NMF to analyze, visualize, and utilize datasets more practically than simply searching the articles. Through this thesis, We first show the superior topic modeling performance of the NMF algorithm and the superior processing speed that can be used for interactive visual interface compared to other methods and then suggest the visual interface that contains various features to help users better analyze and intuitively understand the data. Finally, we present use cases on how this study can be used practically and present their applicability in various fields
-
ItemROBUST COUNTERFACTUAL LEARNING FOR CLINICAL DECISION-MAKING USING ELECTRONIC HEALTH RECORDS(Georgia Institute of Technology, 2020-12-07) Choudhary, AnirudhBuilding clinical decision support systems, which includes diagnosing patient’s disease states and formulating a treatment plan, is an important step toward personalized medicine. The counterfactual nature of clinical decision-making is a major challenge for machine learning-based treatment recommendation, i.e., we can only observe the outcome of the clinician’s actions while the outcome of alternative treatment options is unknown. The thesis is an attempt to formulate robust counterfactual learning frameworks for efficient offline policy evaluation and policy learning using observational data. We focus on the offline data scenario and leverage historically collected Electronic Health Records, since online policy testing can potentially adversely impact the patient’s well-being. The problem is compounded by the inherent uncertainty in clinical decision-making due to heterogeneous patient contexts, the presence of significant variability in patient-specific predictions, smaller datasets, and limited knowledge of the clinician’s intrinsic reward function and environment dynamics. This motivates the need to tackle uncertainty and enable improved clinical policy generalization via context-based policy learning. We propose counterfactual frameworks to tackle the highlighted challenges under two learning scenarios: contextual bandits and dynamic treatment regime. In the bandit setting, we focus on effectively tackling the model uncertainty inherent in inverse propensity weighting methods and highlight our approach’s efficacy on oral anticoagulant dosing task. In dynamic treatment regime, we focus on sequential treatment interventions and consider the problem of imitating the clinician’s policy for sepsis management. We formulate it as a multi-task problem and propose meta-Inverse Reinforcement Learning framework to jointly adapt policy and reward functions to diverse patient groups, thus enabling improved policy generalization.
-
ItemLearning from Multi-Source Weak Supervision for Neural Text Classification(Georgia Institute of Technology, 2020-07-28) Ren, WendiText classification is a fundamental text mining task with numerous real-life applications. While deep neural nets have achieved superior performance for text classification, they rely on large-scale labeled data to achieve strong performance. Obtaining large-scale labeled data, however, can be prohibitively expensive in many applications. In this project, we study the problem of learning neural text classifiers without using any labeled data, but only easy-to-provide heuristic rules as weak supervision. This problem is challenging because rule-induced weak labels are often noisy and incomplete. To address these challenges, we propose a model that can be learned from multiple weak supervision sources with two key components. The first component is a rule denoiser, which estimates conditional source reliability using a soft attention mechanism and reduces label noise by aggregating rule- induced noisy data. The second is a neural classifier that predicts soft labels for unmatchable samples to address the rule coverage issue. The two components are integrated into a co-training framework, which can be trained end-to-end to mutually enhance each other. We evaluate our model on five benchmarks for four popular text classification tasks, including sentiment analysis, topic classification, spam classification, and relation extraction. The results show that our model outperforms state-of-the-art weakly-supervised and semi-supervised methods, and achieves comparable performance with fully-supervised methods even without any labeled data.