Wildfires cause major damages to forests and infrastructures every year. The cost of wildfire control and damage repair are rapidly increasing, leading to significant financial burdens for many communities, especially remote and rural communities. An effective way to mitigate the damage wildfires cause is by detecting the fire early. Recently, a combination of Unmanned Aerial Vehicles (UAVs) and deep learning has been proposed as a method for early detection. However, a major challenge is the lack of data to train learning models. This causes current models to lack the ability to generalize the task of wildfire detection. In this paper, we present One Class Classification (OCC) as an alternative solution to alleviate the lack of data. Our models achieved F1 scores of 0.94 when trained and tested on the FLAME data set and 0.97 when trained and tested on the FLAME2 data set.

Cardiovascular diseases, such as heart attack and congestive heart failure, are the leading cause of death in the United States and worldwide. The current medical practice for diagnosing cardiovascular diseases is not suitable for long-term, out-of-hospital use. A key to long-term, at-home cardiac care is the ability to provide continuous monitoring, and detect abnormal cardiac rhythms, i.e., arrhythmia, in real-time. Various big data and deep learning based approaches have been developed to analyze electrocardiogram data to identify arrhythmia conditions. However, most existing studies only focus on the accuracy of arrhythmia classification, instead of runtime performance of the workflow, which is critical for real-time detection. In this paper, we propose progressive resolution shrinking, a new method for supporting efficient execution of deep learning models for arrhythmia detection, without compromising the detection accuracy. Specifically, we explored multidimensional methods in reducing the amount of information needed for the learning task, and developed a new training method to leverage the advantage of reduced resolution. We have evaluated this approach using real electrocardiogram data, and the experimental results show that it effectively improves the efficiency of arrhythmia detection while preserving high accuracy. We expect this approach will pave the way for real-time arrhythmia detection on resource-constrained wearable devices.

Modern computing applications increasingly rely on technologies in artificial intelligence, machine learning, and big data analytics. These applications often demand more powerful and energy-efficient hardware. Resistive switching random access memory (ReRAM) has emerged as a promising solution to satisfy both the storage and computing needs. In this paper, a natural organic honey film embedded with carbon nanotube (CNT) was fabricated into a resistive switching device, and the resistive switching behaviors were investigated. Endurance test results show the cycle-to-cycle variation of set and reset voltages. On/Off ratio in retention test was found to be in the order of ~105which proves its potential as a non-volatile memory device to support neuromorphic computing applications. This research opens up opportunities to execute big data and machine learning applications with modest energy consumption and minimal electronic waste.

The growing popularity of big data and machine learning applications call for a more powerful and energy-efficient way to execute deep learning workflows. Neuromorphic chips provide a potential solution, as they attempt to mimic the neuronal architectures in human brain and show great potentials in reducing energy consumption in the order of magnitude and also improve the computational performance. However, the fabrication process for neuromophic chips is costly and currently based on trial-and-error, which adds complexity to the design process. In this paper, we address this challenge by designing and developing machine learning guided microfabrication process for Resistive Random Access Memory (RRAM), which is a key device in neuromorphic chips. Experimental results show that our approach is effective in terms of predicting the performance of RRAM devices fabricated under various process conditions.

Cardiovascular diseases, such as heart attack and congestive heart failure, are the leading cause of death both in the United States and worldwide. The current medical practice for diagnosing cardiovascular diseases is not suitable for long-term, out-of-hospital use. A key to long-term monitoring is the ability to detect abnormal cardiac rhythms, i.e., arrhythmia, in real-time. In this paper, we present our work in designing real-time sensing, and evaluating machine learning algorithms for real-time arrhythmia detection. Most of the existing work applies machine learning algorithms to electrocardiogram (ECG) images to detect abnormal patterns. These approaches are not suitable for real-time processing due to high processing overhead. In our work, we treat data as time series, and evaluate various machine learning algorithms in terms of both learning and computational performance. Our experimental results show that the long short-term memory network (LSTM) has both high accuracy and efficiency, demonstrating great potential for online detection of arrhythmia.

Electronically controllable magnetorheological brakes (MRB) can be used in haptic devices to apply forces/torques to the user in a virtual reality (VR) simulation to improve realism. Precise control of the braking torque is possible with a control system using a Hall sensor which measures the magnetic field. Machine learning models can be used to predict the output torque using the input from the Hall sensor. However, over time the fluid leaks out of the MRB due to failure of rubber seals, which degrades the haptic device performance and presents challenges in torque prediction. In this paper, we present our efforts in developing machine learning based approaches that can capture the dynamic behavior of an MRB and its changing torque output as the fluid leaks out. Extensive experiments have been carried out using data collected from the device, and results show that our 2-Step-RN approach can accurately predict the output torque. Notably, it even outperforms the baseline models which are trained for and operate at a stable fluid level, indicating its great potential for enabling torque control of MRB devices with high fidelity.

Today's online social networks produce a significant amount of data containing rich information. A major challenge is to analyze and make sense of the data, so that useful information can be retrieved efficiently. In many application scenarios, this requires analyzing both the network topology and the textual information contained in the network. However, existing network analysis tools usually focus on one of these aspects, instead of providing end-to-end solutions for this particular research scenario. In this paper, we present NetNet, a social network analysis tool that is specifically designed to simplify the workflow of analyzing social networks. In NetNet, we model social networks as interconnected user nodes with text nodes associated with them, and leverage network analysis and text mining algorithms to perform both tasks seamlessly. We evaluate the performance of NetNet using social network data, and the experimental results show that our design is practically efficient and scalable.

Phasor measurement units (PMUs) provide high-fidelity situational awareness of electric power grid operations. PMU data are used in real-time to inform wide area state estimation, monitor area control error, and event detection. As PMU data becomes more reliable, these devices are finding roles within control systems such as demand response programs and early fault detection systems. As with other cyber physical systems, maintaining data integrity and security are significant challenges for power system operators. In this paper, we present a comprehensive study of multiple machine learning techniques for detecting malicious data injection within PMU data streams. The two datasets used in this study are from the Bonneville Power Administration's PMU network and an inter-university PMU network among three universities, located in the U.S. Pacific Northwest. These datasets contain data from both the transmission level and the distribution level. Our results show that both SVM and ANN are generally effective in detecting spoofed data, and TensorFlow, the newly released tool, demonstrates potential for distributing the training workload and achieving higher performance. We expect these results to shed light on future work of adopting machine learning and data analytics techniques in the electric power industry.