Search results

Title: ACTIVE INFERENCE FOR PREDICTIVE MODELS OF SPATIO-TEMPORAL DOMAINS
Creator: Komurlu, Caner
Date: 2019
Description: Active inference is the method of selective information gathering during prediction in order to increase a predictive machine learning model's...
Show moreActive inference is the method of selective information gathering during prediction in order to increase a predictive machine learning model's prediction performance. Unlike active learning, active inference does not update the model, but rather provides the model with useful information during prediction to boost the prediction performance. To be able to work with active inference, a predictive model needs to exploit correlations among variables that need to be predicted. Then the model, while being provided with true values for some of the variables, can make more accurate predictions for the remaining variables.In this dissertation, I propose active inference methods for predictive models of spatio-temporal domains. I formulate and investigate active inference in two different domains: tissue engineering and wireless sensor networks. I develop active inference for dynamic Bayesian networks (DBNs) and feed-forward neural networks (FFNNs).First, I explore the effect of active inference in the tissue engineering domain. I design a dynamic Bayesian network (DBN) model for vascularization of a tissue development site. The DBN model predicts probabilities of blood vessel invasion in regional scale through time. Then utilizing spatio-temporal correlations between regions represented as variables in the DBN model, I develop an active inference technique to detect the optimal time to stop a wet lab experiment. The empirical study shows that the active inference is able to detect the optimal time and the results are coherent with domain simulations and lab experiments.In the second phase of my research, I develop variance-based active inference techniques for dynamic Bayesian networks for the purpose of battery saving for wireless sensor networks (WSN). I propose the expected variance reduction active inference method to detect variables that reduce the overall variance the most. I first propose a DBN model of a WSN. I then compare the prediction performance of the DBN with Gaussian processes and linear chain graphical models on three different WSN data using several baseline active inference methods. After showing that DBNs perform better than the baseline predictive models, I compare the performance of expected variance reduction active inference method with the performances of baseline methods on the DBN, and show the superiority of the expected variance reduction on the three WSN data sets.Finally, to address the inference complexity and the limitation of representing linear correlations due to Gaussian assumption, I replace the DBN representation with a feed-forward neural network (FFNN) model. I first explore techniques to integrate observed values into predictions on neural networks. I adopt the input optimization technique. Finally, I discover two problems: model error and optimization overfitting. I show that the input optimization can mitigate the model error. Lastly, I propose a validation-based regularization approach to solve the overfitting problem.
Show less

Title: Removing Confounds in Text Classification for Computational Social Science
Creator: Landeiro Dos Reis, Virgile
Date: 2018
Description: Nowadays, one can use social media and other online platforms to communicate with friends and family, write a review for a product, ask...
Show moreNowadays, one can use social media and other online platforms to communicate with friends and family, write a review for a product, ask questions about a topic of interest, or even share details of private life with the rest of the world. The ever-increasing amount of user-generated content has provided researchers with data that can offer insights on human behavior. Because of that, the field of computational social science - at the intersection of machine learning and social sciences - has soared in the past years, especially within the field of public health research. However, working with large amounts of user-generated data creates new issues. In this thesis, we propose solutions for two problems encountered in computational social science and related to confounding bias.First, because of the anonymity provided by online forums, social networks, or other blogging platforms through the common usage of usernames, it is hard to get accurate information about users such as gender, age, or ethnicity. Therefore, although collecting data on a specific topic is made easier, conducting an observational study with this type of data is not simple. Indeed, when one wishes to run a study to measure the effect of a variable on another variable, one needs to control for potential confounding variables. In the case of user-generated data, these potential confounding variables are at best noisily observed or inferred and at worst not observed at all. In this work, we wish to provide a way to use these inferred latent attributes in order to conduct an observational study while reducing the effect of confounding bias as much as possible. We first present a simple matching method in a large-scale observational study. Then, we propose a method to retrieve relevant and representative documents through adaptive query building in order to build the treatment and control groups of an observational study.Second, we focus on the problem of controlling for confounding variables when the influence of these variables on the target variable of a classification problem changes over time. Although identifying and controlling for confounding variables has been assiduously studied in empirical social science, it is often neglected in text classification. This can be understood by the fact that, if we assume that the impact of confounding variables does not change between the training and the testing data, then prediction accuracy should only be slightly affected. Yet, this assumption often does not hold when working with user-generated text. Because of this, computational science studies are at risk of reaching false conclusions when based on text classifiers that are not controlling for confounding variables. In this document, we propose to build a classifier that is robust to confounding bias shift, and we show that we can build such a classifier in different situations: when there are one or more observed confounding variables, when there is one noisily predicted confounding variable, or when the confounding variable is unknown but can be detected through topic modeling.
Show less

Title: A SCALABLE SIMULATION AND MODELING FRAMEWORK FOR EVALUATION OF SOFTWARE-DEFINED NETWORKING DESIGN AND SECURITY APPLICATIONS
Creator: Yan, Jiaqi
Date: 2019
Description: The world today is densely connected by many large-scale computer networks, supporting military applications, social communications, power...
Show moreThe world today is densely connected by many large-scale computer networks, supporting military applications, social communications, power grid facilities, cloud services, and other critical infrastructures. However, a gap has grown between the complexity of the system and the increasing need for security and resilience. We believe this gap is now reaching a tipping point, resulting in a dramatic change in the way that networks and applications are architected, developed, monitored, and protected. This trend calls for a scalable and high-fidelity network testing and evaluation platform to facilitate the transformation from in-house research ideas to real-world working solutions. With this objective, we investigate means to build a scalable and high-fidelity network testbed using container-based emulation and parallel simulation; our study focuses on the emerging software-defined networking (SDN) technology. Existing evaluation platforms facilitate the adoption of the SDN architecture and applications to production systems. However, the performance of those platforms is highly dependent on the underlying physical hardware resources. Insufficient resources would lead to undesired results, such as low experimental fidelity or slow execution speed, especially with large-scale network settings. To improve the testbed fidelity, we first develop a lightweight virtual time system for Linux container and integrate the system into a widely-used SDN emulator. A key issue with an ordinary container-based emulator is that it uses the system clock across all the containers even if a container is not being scheduled to run, which leads to the issue of both performance and temporal fidelity, especially with high workloads. We investigate virtual time approaches by precisely scaling the time of interactions between containers and physical devices. Our evaluation results indicate a definite improvement in fidelity and scalability. To improve the testbed scalability, we investigate how the centralized paradigm of SDN can be utilized to reduce the simulation workload. We explore a model abstraction technique that effectively transforms the SDN network devices to one virtualized switch model. While significantly reducing the model execution time and enabling the real-time simulation capability, our abstracted model also preserves the end-to-end forwarding behavior of the original network.With enhanced fidelity and scalability, it is realistic to utilize our network testbed to perform a security evaluation of various SDN applications. We notice that the communication network generates and processes a huge amount of data. The logically-centralized SDN control plane, on the one hand, has to process both critical control traffic and potentially big data traffic, and on the other hand, enables many efficient security solutions, such as intrusion detection, mitigation, and prevention. Recently, deep neural networks achieve state-of-the-art results across a range of hard problem spaces. We study how to utilize the big data and deep learning to secure communication networks and host entities. For classifying malicious network traffic, we have performed the feasibility study of off-line deep-learning based intrusion detection by constructing the detection engine with multiple advanced deep learning models. For malware classification on individual hosts, another necessity to secure computer systems, existing machine learning-based malware classification methods rely on handcrafted features extracted from raw binary files or disassembled code. The diversity of such features created has made it hard to build generic malware classification systems that work effectively across different operational environments. To strike a balance between generality and performance, we explore new graph convolutional neural network techniques to effectively yet efficiently classify malware programs represented as their control flow graphs.
Show less

Title: AI IN MEDICINE: ENABLING INTELLIGENT IMAGING, PROGNOSIS, AND MINIMALLY INVASIVE SURGERY
Creator: Getty, Neil
Date: 2022
Description: While an extremely rich research field, compared to other applications of AI such as natural language processing (NLP) and image processing...
Show moreWhile an extremely rich research field, compared to other applications of AI such as natural language processing (NLP) and image processing/generation, AI in medicine has been much slower to be applied in real-world clinical settings. Often the stakes of failure are more dire, the access of private and proprietary data more costly, and the burden of proof required by expert clinicians is much higher. Beyond these barriers, the often typical data-driven approach towards validation is interrupted by a need for expertise to analyze results. Whereas the results of a trained Imagenet or machine translation model are easily verified by a computational researcher, analysis in medicine can be much more multi-disciplinary demanding. AI in medicine is motivated by a great demand for progress in health-care, but an even greater responsibility for high accuracy, model transparency, and expert validation.This thesis develops machine and deep learning techniques for medical image enhancement, patient outcome prognosis, and minimally invasive robotic surgery awareness and augmentation. Each of the works presented were undertaken in di- rect collaboration with medical domain experts, and the efforts could not have been completed without them. Pursuing medical image enhancement we worked with radiologists, neuroscientists and a neurosurgeon. In patient outcome prognosis we worked with clinical neuropsychologists and a cardiovascular surgeon. For robotic surgery we worked with surgical residents and a surgeon expert in minimally invasive surgery. Each of these collaborations guided priorities for problem and model design, analysis, and long-term objectives that ground this thesis as a concerted effort towards clinically actionable medical AI. The contributions of this thesis focus on three specific medical domains. (1) Deep learning for medical brain scans: developed processing pipelines and deep learn- ing models for image annotation, registration, segmentation and diagnosis in both traumatic brain injury (TBI) and brain tumor cohorts. A major focus of these works is on the efficacy of low-data methods, and techniques for validation of results without any ground truth annotations. (2) Outcome prognosis for TBI and risk prediction for Cardiovascular Disease (CVD): we developed feature extraction pipelines and models for TBI and CVD patient clinical outcome prognosis and risk assessment. We design risk prediction models for CVD patients using traditional Cox modeling, machine learning, and deep learning techniques. In this works we conduct exhaustive data and model ablation study, with a focus on feature saliency analysis, model transparency, and usage of multi-modal data. (3) AI for enhanced and automated robotic surgery: we developed computer vision and deep learning techniques for understanding and augmenting minimally invasive robotic surgery scenes. We’ve developed models to recognize surgical actions from vision and kinematic data. Beyond model and techniques, we also curated novel datasets and prediction benchmarks from simulated and real endoscopic surgeries. We show the potential for self-supervised techniques in surgery, as well as multi-input and multi-task models.
Show less

Title: SOLID-STATE SMART PLUG DEVICE
Creator: Deng, Zhixi
Date: 2022
Description: Electrical faults are a leading cause of residential fire, and flexible power cords are particularly susceptible to metal or insulation...
Show moreElectrical faults are a leading cause of residential fire, and flexible power cords are particularly susceptible to metal or insulation degradation that may lead to a variety of electrical faults. Smart Plugs are a type of plug-in device controlling electrical loads via wireless communication for consumer market. However, there is lack of circuit protection features in existing Smart Plug products. Moreover, there is no previous product or research on Smart Plug with circuit protection features. This thesis introduces a new Smart Plug 2.0 concept which offers all-in-one protection against over-current, arc, and ground faults in addition to the smart features in Smart Plug products. It aims at preventing fire and shock hazards caused by degraded or damaged power cords and electrical connections in homes and offices. It offers microsecond-scale time resolution to detect and respond to a fault condition, and significantly reduces the electrothermal stress on household electrical wires and loads. A new arc fault detection method is developed using machine learning models based on load current di/dt events. The Smart Plug 2.0 concept has been validated experimentally. A 120V/10A solid-state Smart Plug 2.0 prototype using power MOSEFTs is designed and tested. It has experimentally demonstrated the comprehensive protection features against all types of electrical faults.
Show less

Title: Exploiting contextual information for deep learning based object detection
Creator: Zhang, Chen
Date: 2020
Description: Object detection has long been an important research topic in computer vision area. It forms the basis of many applications. Despite the great...
Show moreObject detection has long been an important research topic in computer vision area. It forms the basis of many applications. Despite the great progress made in recent years, object detection is still a challenging task. One of the keys to improving the performance of object detection is to utilize the contextual information from the image itself or from a video sequence. Contextual information is defined as the interrelated condition in which something exists or occurs. In object detection, such interrelated condition can be related background/surroundings, support from image segmentation task, and the existence of the object in the temporal domain for video-based object detection. In this thesis, we propose multiple methods to exploit contextual information to improve the performance of object detection from images and videos.First, we focus on exploiting spatial contextual information in still-image based object detection, where each image is treated independently. Our research focuses on extracting contextual information using different approaches, which includes recurrent convolutional layer with feature concatenation (RCL-FC), 3-D recurrent neural networks (3-D RNN), and location-aware deformable convolution. Second, we focus on exploiting pixel-level contextual information from a related computer vision task, namely image segmentation. Our research focuses on applying a weakly-supervised auxiliary multi-label segmentation network to improve the performance of object detection without increasing the inference time. Finally, we focus on video object detection, where the temporal contextual information between video frames are exploited. Our first research involves modeling short-term temporal contextual information using optical flow and modeling long-term temporal contextual information using convLSTM. Another research focuses on creating a two-path convLSTM pyramid to handle multi-scale temporal contextual information for dealing with the change in object's scale. Our last work is the event-aware convLSTM that forces convLSTM to learn about the event that causes the performance to drop in a video sequence.
Show less

repository.iit

Search the repository

Enabled Filters

Refine Results

Type

Date

Subject

Creator

Rights