Dartmouth Institute for Health Policy and Clinical Practice
Dartmouth Digital Commons (Dartmouth College)Not a member yet
8214 research outputs found
Sort by
Data-Driven Computing Methods for Nonlinear Physics Systems with Geometric Constraints
Get PDFIn a landscape where scientific discovery is increasingly driven by data, the integration of machine learning (ML) with traditional scientific methodologies has emerged as a transformative approach. This paper introduces a novel, data-driven framework that synergizes physics-based priors with advanced ML techniques to address the computational and practical limitations inherent in first-principle-based methods and brute-force machine learning methods. Our framework showcases four algorithms, each embedding a specific physics-based prior tailored to a particular class of nonlinear systems, including separable and nonseparable Hamiltonian systems, hyperbolic partial differential equations, and incompressible fluid dynamics. The intrinsic incorporation of physical laws preserves the system\u27s intrinsic symmetries and conservation laws, ensuring solutions are physically plausible and computationally efficient. The integration of these priors also enhances the expressive power of neural networks, enabling them to capture complex patterns typical in physical phenomena that conventional methods often miss. As a result, our models outperform existing data-driven techniques in terms of prediction accuracy, robustness, and predictive capability, particularly in recognizing features absent from the training set, despite relying on small datasets, short training periods, and small sample sizes
Evading Antivirus Detection by Abusing File Type Identification
Get PDFFile type identification is a vital step in automated file processing, especially in the realm of malware detection. The challenges with file type identification and evasion techniques that take advantage of them were pointed out over a decade ago. We show that this remains the case: file type identification implementations are still fragile, especially for files with ambiguous file types. We present a novel antivirus bypass technique via crafted tar archives that evades all detection from VirusTotal and numerous antiviruses: BitDefender, F-Secure, Kaspersky, Panda Dome, Trend Micro, Quick Heal, IKARUS, Avira. These crafted files evade detection by tricking file type identification implementations, but can still be unpacked on end-host machines using GNU tar or 7-zip. We show that these file type-masquerading archives are also incorrectly labeled by popular file type identifiers. We present a survey of publicly available tools for file type identification and shared file signature databases. Finally, we discuss countermeasures for this evasion technique by detecting files with ambiguous file types
Curating Familiarity within the Unfamiliar: Exploring Non-Native Mobile App Experiences to Create Cross-Cultural Design Frameworks
Get PDFGlobal mobility and markets are expanding, and as a result, countries are becoming less and less monocultural. With multiple cultural affinity groups to cater towards, companies often will deploy different versions of a website or app based on the country a user is accessing it from. This strategy of catering to geographic location results in a lack of accommodation for people living within a culture that is different from their native one. In order to increase accessibility and equal ease-of-use for all audiences, designers should understand and work towards the needs of a multicultural user base. This study investigates how cultural differences can be defined and measured, and how those dimensions of difference can be used to better understand cultural behaviors and thinking patterns in relation to the use of digital interfaces. Research was conducted on previous cultural behavior studies and on current trends in user interface and user experience (UI/UX) design based on country of origin. Additionally, a redesign of a current mobile app for an audience of two cultural backgrounds was created, prototyped, and tested with users in order to adapt and develop existing design methods and tools for a more effective cross-cultural creation process. The study concludes by presenting a set of heuristics and recommendations that designers should consider when designing products for users of different cultural backgrounds, to support increased cultural accessibility for user groups around the world
Sky\u27s the Limit: Seeing the Future in a Disturbing Summer Storm
Get PDFA disturbing electrical storm in July 2020 in Washington, D.C. demarcates a break with the past. A runner-up in the 2023 Waterman Fund Essay Contest
Cold Climate Heating Alternatives: A Feasibility Study for Heat Pumps in Far Northern Greenland
No full textBuilding heating in the Arctic is extremely energy intensive. The cold climate of northern Greenland requires heating year-round. Qaanaaq, the northernmost village in Greenland, has 10,128 average annual heating degree days, in Celsius. Current heating relies on fossil fuels, often from an oil boiler or from diesel-powered district heating. Amidst the global transition of energy supply and consumption to reduce the use of fossil fuels, electrically based heating via heat pumps is becoming an increasingly attractive option to heat buildings more efficiently. Several heat pumps are now designed for cold climates specifically. While studies have investigated the potential and performance of heat pumps in cold climates, no studies have investigated their feasibility in the extreme cold of northern Greenland. Qaanaaq sits on a fjord with year-round water temperatures of 0.5°C to 1.5°C. The ground has permafrost beginning 1m to 2.5m below the ground surface, and air temperatures reach -40°C in the winter. This thesis analyzes the feasibility of air source heat pumps (ASHPs), ground source heat pumps (GSHPs), and surface water heat pumps (SWHPs) based on four criteria: their overall efficiency compared to the efficiency of an oil boiler, cost, the total available energy in the energy source compared to the annual heat energy demand of a typical house, and power delivery capability compared to the peak hourly power demand of a typical house. Analysis finds that Coefficient of Performance (COP) values for ASHPs, GSHPs and SWHPs are likely near or below 3.1, the COP required for efficiency comparable to an oil heater. All sources can supply the energy and power necessary. However, the cost per kilowatt-hour (kWh) is higher than that from oil-based heating. While solar energy is readily available for half the year, the costs of installing a solar photovoltaic system for electricity would be larger than the savings derived from reduced use of oil for home heating, by approximately a factor of two. With current costs and subsidies, heat pumps powered from diesel-based or solar-based electricity are not economically competitive in the cold climate of far northern Greenland, and oil boilers remain a more cost-effective heating option
Resolving Low-Clarity Text Input on Mobile Platforms
No full textText input constitutes a critical and frequent interaction with mobile computing devices. However, the experiences on mobile text interfaces remain significantly inferior in terms of speed and accuracy, as input noise, jitteriness, and spatial variability in human input signals result in reduced clarity of the input. This hinders the ability of the text input system to accurately transform them into the intended text. The compact form factors of modern mobile and wearable devices further complicate precise input, as the input space is highly limited. The inadequate clarity of human input on these devices, therefore, necessitates precisely formatted input from users, imposing a significant physical and mental burden on them. In this thesis, we explore two different methodologies to tackle this problem. Firstly, we design and optimize miniature interfaces to enhance the clarity of text input on compact AR/VR wearable devices. We showcase the development of fingertip keyboards, a miniature text interface designed for wearable devices, allowing users to type with micro-finger gestures in an eyes-free manner. In the second methodology, we leverage advanced language modeling to decode ambiguous input paradigms where clarity is further diminished in exchange for reduced user effort. We present non-delimited phrase gestures on smartphones, where users can swipe through all letters of the words in a phrase using a single, continuous gesture. This technique provides more flexibility for multi-word gestural input, as users do not need to lift their fingers after each word. We also present LLM-powered abbreviated writing on tablets, a more ambiguous input form where most characters in a phrase can be skipped when abbreviating. With an intuitive interface and a fine-tuned LLM decoder, users can save significant physical effort while maintaining competitive input performance
Device Discovery in the Smart Home Environment
No full textWith the availability of Internet of Things (IoT) devices offering varied services, smart home environments have seen widespread adoption in the last two decades. Protecting privacy in these environments becomes an important problem because IoT devices may collect information about the home’s occupants without their knowledge or consent. Furthermore, a large number of devices in the home, each collecting small amounts of data, may, in aggregate, reveal non-obvious attributes about the home occupants. A first step towards addressing privacy is discovering what devices are present in the home. In this paper, we formally define device discovery in smart homes and identify the features that constitute discovery in that environment. Then, we propose an evaluative rubric that rates smart home technology initiatives on their device discovery capabilities and use it to evaluate four commonly deployed technologies. We find none cover all device discovery aspects. We conclude by proposing a combined technology solution that provides comprehensive device discovery tailored to smart homes
Go Green: A Carbon Tracking Tool for the Berlin, Green City FSP
No full textThe Berlin Green City Foreign Study Program (FSP) is co-hosted by the Dartmouth College German and Engineering Departments. On the FSP , students take courses including ENGS37 (Introduction to Environmental Engineering) and ENGS 45 (Sustainable Urban Systems), but the program has thus far lacked a formal curriculum and conversation around sustainable living and practices. The recommendations and deliverables in this report are intended to generate a culture of sustainability in Dartmouth students\u27 everyday lives, at home and abroad. Our solution is two-fold: a web-based tool for students to track their carbon behavior, and a curriculum designed to educate students on sustainability. The Go Green web platform includes a comparative carbon calculator for travel, home, and food habits, a goal-setting and tracking feature, and a team page with a leaderboard and group carbon reduction targets. The 10-week curriculum focuses on a different sustainability theme each week and encourages discussion about relevant practices in Berlin and at Dartmouth. Examples of weekly focus areas include transportation, heating and cooling, water and sewage, and waste management. Our project aims to motivate sustainable behavior by providing students with a platform to track both personal and team goals, emphasizing small, tangible behavioral changes that turn into lifelong habits
Bike Walk Census Tool
No full textThe Bike W alk Census Tool is a machine learning-based counting and analysis device. It is designed to gather data for municipal entities aiming to improve their pedestrian and bike infrastructure. It has two primary components: a portable recording device, to be placed where the user wishes to gather data, and a software package installed on a provided NVIDIA Jetson computer which generates census data on the recorded video. The recording device is modular, weather-resistant, and portable. The user places the device in a clear and unobstructed area, where they can either manually interface with the device using onboard buttons, or use a web application on their phone which also provides a video display of the area being captured and the ability to frame the shot. Once a recording is complete, the user transfers the video over to the computer running the analysis tool. There, the user will draw lines, and the analysis tool increments separate counters for each type of road or sidewalk user crossing each line. After the tool has completed its analysis, which takes approximately as long as the real time duration of the recorded video, the user is provided with count data along with other useful data analyses such as conglomerated tracks and time-segmented counts