150813 research outputs found
Sort by
Flex-Route Transit for Smart Cities: A Reinforcement Learning Approach to Balance Ridership and Performance
No full textA major challenge for modern transit systems relying on traditional fixed-route designs is providing broad accessibility to users. Flex-route transit can enhance accessibility in low-density areas, since it combines the directness of fixed-route transit with the coverage of on-demand mobility. Although deviating for optional pickups can increase ridership and transit accessibility, it also deteriorates the service performance for fixed-route riders. To balance this inherent trade-off, this paper proposes a reinforcement learning approach for deviation decisions. The proposed model is used in a case study of a proposed flex-route service in the city of Boston. The performance on competing objectives is evaluated for reward configurations that adapt to peak and off-peak scenarios. The analysis shows a significant improvement of our method compared to a heuristic derived from industry practice as a baseline. To evaluate robustness, we assess performance across scenarios with varying demand compositions (fixed and requested riders). The results show that the method achieves greater improvements than the baseline in scenarios with increased request ridership, i.e., where decision-making is more complex. Our approach improves service performance under dynamic demand conditions and varying priorities, offering a valuable tool for smart cities to operate flex-route services
Discovering and Engineering the Computation Underlying Large Intelligent Agents
No full textThe richness of language and intelligent behavior has often been attributed to latent compositional structure. Can we build tools for discovering how deep networks learn and represent this latent structure implicitly? And more importantly, can we use this knowledge to improve generalization in largely structure-less general purpose models or refine our understanding of the world they describe? In this dissertation, I present three perspectives to answer these questions. First, I present experimental methods to functionally characterize the space of learnt solutions in LLMs and demonstrate how this understanding can be used to improve their empirical generalization in a gradient free manner, sometimes by as much as 30% points on language understanding benchmarks. Following that, I show how to decipher the structure of another (black box) language-like system, the naturally arising communication system of sperm whales in the wild, discovering for the first time a unique combinatorial communication system. Finally, I apply insights from these results to equip embodied agents with a latent language of thought—hierarchical and compositional—and show how it can enable long-horizon reasoning and planning in these systems. This dissertation ultimately aims to bridge the gap between natural and artificial intelligence, offering new insights into both the fundamental nature of communication in complex biological organisms in the wild and the development of more powerful, and improved AI systems. A key pattern in the discoveries in this thesis has been how simple structures enable complex externalized behaviors in both biological organisms and AI systems.Ph.D
TutorUp: What If Your Students Were Simulated? Training Tutors to Address Engagement Challenges in Online Learning
No full textCHI ’25, Yokohama, JapanWith the rise of online learning, many novice tutors lack experience engaging students remotely. We introduce TutorUp, a Large Language Model (LLM)-based system that enables novice tutors to practice engagement strategies with simulated students through scenario-based training. Based on a formative study involving two surveys (N1 = 86, N2 = 102) on student engagement challenges, we summarize scenarios that mimic real teaching situations. To enhance immersion and realism, we employ a prompting strategy that simulates dynamic online learning dialogues. TutorUp provides immediate and asynchronous feedback by referencing tutor-students online session dialogues and evidence-based teaching strategies from learning science literature. In a within-subject evaluation (N = 16), participants rated TutorUp significantly higher than a baseline system without simulation capabilities regarding effectiveness and usability. Our findings suggest that TutorUp provides novice tutors with more effective training to learn and apply teaching strategies to address online student engagement challenges
Technology-Policy Handbook for Trans-Atlantic Nuclear Maritime Corridors: Ports, Infrastructure, and Safety
No full textAdvised by Jacopo Buongiorno, Eric Forrest, Fotini Christia, Koroush Shirvanm and Themis Sapsis.On September 18, 2025, the United States and the United Kingdom published a Memorandum of Understanding (MoU) on scientific and technological advancement. This new partnership focuses on understanding and deploying disruptive technologies in Artificial Intelligence, quantum, and civil nuclear energy. Less highlighted was a single sentence within the MoU outlining efforts to "explore opportunities" for establishing a "maritime shipping corridor" between the US and UK. So far, research on civilian nuclear ships has generally prioritized ship design and operation analysis. This paper will instead analyze port, regulatory, and infrastructural issues within this space and provide a path forward for technology policy solutions supporting systems safety
Design and Testing of a Hadal Sediment Sampling System
No full textSediment cores are conventionally gathered to collect data on seabed chemical and mineral composition. These samples are important in determining the health of the surrounding environment and, if sediment layers are preserved, assessing the environmental history and trajectory of our ocean floors. Over the last few decades, some attempts to collect sediment core samples have been made from as deep as the Mariana Trench, which contains the deepest known point on Earth’s surface at 11,000 meters [1] [2]. However, the hadal region remains one of the most underexplored areas of our oceans [3].
Working in conjunction with Inkfish, a submersible technology company, we developed a deep-sea sediment core sampler that will travel to the Mariana Trench aboard one of Inkfish’s submersible landers and will collect four inches of sediment in ambient pressures of 110.32 MPa [4].
The first phase of this project was to design and fabricate a prototype sediment core sampling device. We designed an entirely mechanical sampler because underwater actuators suitable for use in hadal conditions are difficult to source and require additional communication and power resources. Furthermore, our device aims to preserve the layers of the sediment core samples as the lander ascends to preserve relative time scales within the sample.
The next phase involved testing the functionality of different subsystems of our device. In this paper, we considered two different sediment collection apparatuses and one-way valves for our collection tube. We performed field testing at the Charles River in Cambridge, MA to assess which apparatus and valve combination would provide the best results based on the volume of sediment collected and retained.
Building on our mechanical proof-of-concept, we improved the design and fabricated a second iteration prototype for 2-3 km depths. In this version, we addressed the two main weaknesses of the initial proof of concept design: substantial friction during sliding and the complexity of assembly and maintenance due to the numerous parts.
Lastly, we engineered a deployment system that made our device compatible with Inkfish’s lander deployment procedure. The deployment system was designed to lower the sediment sampler device from within the bay to below the lander once the lander was submerged in water.
In early October, the improved sediment sampler and the deployment system were tested during an engineering expedition near the shores of Tonga. Ultimately, we believe that our sediment sampler presents a viable purely mechanical solution to collecting deep-sea sediment from profoundly unexplored areas at hadal depths like the Mariana Trench. Our sampler can easily be mounted onto any surface where it would touch the ocean floor, requiring no electronics or controls. Though we were constrained by the particular seafloor lander used by Inkfish, the size of the sampler is scalable, allowing both the sample diameter and depth to be adjusted for a given mission. By making these sediment samples more accessible, we believe we will have an impact across a number of marine research areas.S.B
The Kigali story, the Singapore model, and rights to the city
No full textThree recent ethnographies of Kigali's urban planning and development provide a welcome addition to a long tradition of such ethnographies, including Lisa Redfield Peattie's famous fieldwork in the planning of Ciudad Guayana (1968; 1987), Grace Goodell's ethnographic account of the disjunction between planning offices in Tehran and the urban settlements (sharaks) of the Khuzistan Development Project modelled on the Tennessee Vally Authority (1986), and Gökce Günel's ethnographic analysis of the disjunction between plans for, and implementation of, Mazdar City and Mazdar Institute in Abu Dhabi (2019)
Diverging global incidence trends of early-onset cancers: comparisons with incidence trends of later-onset cancers and mortality trends of early-onset cancers
No full textBackground The global increase in the incidence of early-onset cancers (defined as cancers diagnosed at 20–49 years old) is a serious public health problem. We investigated 1) whether the incidence trend of early-onset cancers differs from that of later-onset cancers and 2) whether both the incidence and mortality of early-onset cancers have increased concurrently. Methods We utilized age-standardized incidence and mortality rates for early-onset and later-onset cancers diagnosed between 2000 and 2017 from the Cancer Incidence in Five Continents and World Health Organization (WHO) mortality databases. The national obesity prevalence among adults aged 20–49 years was obtained from the National Clinical Database. Using joinpoint regression models, we calculated average annual percentage changes (AAPCs) for cancer incidence and mortality by cancer types and countries. We additionally conducted human development index (HDI)-stratified analyses and assessed the correlation between the obesity prevalence in younger populations and early-onset cancer incidence by country. To investigate the more recent trend of early-onset cancer mortality, we extended our mortality analysis after 2017 for cancer types and countries with statistically significant positive AAPCs in both incidence and mortality of early-onset cancers between 2000 and 2017. Results Our analysis showed that 10 early-onset cancer types (thyroid cancer, breast cancer, melanoma, uterine cancer, colorectal cancer, kidney cancer, cervical cancer, pancreatic cancer, multiple myeloma, Hodgkin lymphoma) in females and 7 early-onset cancer types (thyroid cancer, kidney cancer, testis cancer, prostate cancer, colorectal cancer, melanoma, leukemia) in males had statistically significant positive AAPCs in at least 10 countries. Among these, the following early-onset cancer types had significantly higher AAPCs than later-onset cancer types in females: colorectal cancer (6 countries; AAPC range: 1.8–3.8%), cervical cancer (6 countries; AAPC range: 1.2–3.3%), pancreatic cancer (5 countries; AAPC range: 2.3–13.0%), and multiple myeloma (5 countries; AAPC range: 3.1–9.8%); in males: prostate cancer (12 countries; AAPC range: 3.9–18.4%), colorectal cancer (8 countries; AAPC range: 1.8–3.2%), and kidney cancer (6 countries; AAPC range: 2.0–6.0%). We observed statistically significant positive AAPCs in both the incidence and mortality of the following early-onset cancer types: uterine cancer (5 countries) and colorectal cancer (3 countries in females and 5 countries in males). The steeper increases in early-onset cancers compared with later-onset cancers were mainly observed in the very high-HDI country group, including early-onset colorectal cancer (AAPC = 2.4%, 95% CI 2.1–2.6 in females; AAPC = 2.0%, 95% CI 1.7–2.4 in males) to later-onset colorectal cancer (AAPC = −0.1%, 95% CI −0.2 to 0 in females; AAPC = −0.2%, 95% CI −0.3 to 0 in males). We observed strong positive correlations between the increasing obesity prevalence and the rising incidence of early-onset obesity-related cancers in several countries, including Australia (7 cancer types), United Kingdom (7 cancer types), Canada (7 cancer types), Republic of Korea (7 cancer types), and USA (6 cancer types) in females and United Kingdom (7 cancer types), Canada (6 cancer types), Australia (5 cancer types), Sweden (5 cancer types), and Republic of Korea (4 cancer types) in males. Although we did not observe an apparent spike after 2017 in many countries, we observed continued increases in the mortality of certain cancer types, such as uterine cancer (Japan, Republic of Korea, United Kingdom, USA, and Ecuador) in females and colorectal cancer (Argentina, Canada, United Kingdom, and USA) in males. Conclusions The increase in many early-onset cancer types was significantly higher than that of later-onset cancers, and the incidence and mortality of certain early-onset cancer types (such as colorectal cancer) increased simultaneously. Our study highlights global differences in cancer incidence and mortality trends of early-onset and later-onset cancers
Characterization of pGaN-gate power HEMTs
No full textThis thesis presents a comprehensive study of p-GaN gate GaN High Electron Mobility Transistors (HEMTs) with a focus on understanding how fabrication process variations and gate structural designs impact key electrical performance metrics. Five industry-fabricated wafers, each processed with distinct etch depths, contact strategies, and p-GaN surface configurations, were characterized using a combination of DC and pulsed I–V measurements. Full-transistor modules were evaluated alongside specialized test structures to enable both system-level and localized analysis. DC measurements using the Keysight B1505A system revealed that more aggressive gate contact schemes improved ON-resistance and transconductance, but often at the cost of increased gate leakage and reduced threshold control. Pulsed-IV characterization with the Auriga AU4750 system uncovered dynamic Ron degradation behavior and charge trapping effects, especially under high drain bias conditions. Extracted time constants demonstrated process-dependent trends, with wafers retaining more of the p-GaN surface exhibiting slower charge detrapping and more severe transient effects. Specialized test structures provided additional insights into gate lateral conduction, sheet resistance, and contact asymmetry, reinforcing the connection between device layout, processing, and observed variability. These findings highlight critical trade-offs in the design and fabrication of p-GaN gate GaN HEMTs and offer design-aware strategies for optimizing performance and reliability.S.M
Modeling the Impact of Helicopter Vibrations on the Musculoskeletal Health of US Army Blackhawk Helicopter Pilots
No full textThe UH-60, used for troop transport, MEDVAC, and mission control, has evolved over the last 45 years from the Alpha Model to the Lima and Mike models that are currently utilized. Previous studies investigated the impact of Whole-Body Vibrations (WBV) on aviators and the resulting musculoskeletal injury, but none have investigated the efficacy of the Mike model’s Active Vibration Control System (AVCS) on reducing the impact of helicopter vibrations on musculoskeletal health.
Computational analyses of a biomechanical model using OpenSim and motion capture at varying levels of vibration was conducted. This quantifies the response of the spine and the surrounding muscles when vibratory loads are applied while positioned to manipulate the flight controls. A musculoskeletal model was developed to represent the aviator in the seated posture required to effectively manipulate the flight controls. To develop the model, the team recorded motion capture data with a pilot in a pilot test for concept validation. This data was then processed and input in the OpenSim inverse kinematics tool to determine joint angle and to demonstrate the muscle-tendon length of several muscles in the back. Unlike the initial predictions, the muscles in the right side of the back were not consistently longer than those of the left side.
A survey was also developed that builds upon previous efforts, seeking to understand the aviator’s perspective on musculoskeletal injury and prevention, with a focus on the back. Aviators are asked to describe the cause of their injury, methods of injury prevention, and recovery techniques encompassing numerous subpopulations of flight experience: Lima-majority, Mike-only, Mike-majority, and an even mixture of L/M. The data attempts to characterize the impact of the AVCS on aviator spine health. The AVCS should decrease the rate of injury by reducing the vibratory loads experienced by the aviator. This survey is unique to previous questionnaires as it focuses on the user’s perspective of differences between the two models, and the injury or pain felt by each service member.
While it was expected to see a trend of reduced injury occurrence amongst the Mike-only aviators versus those with Lima-majority flight hours, this was not the case. Injury prevalence was consistent across most populations, indicating the potential inefficacy of the AVCS. Analysis of open-ended responses, particularly from the hybrid group, provide some context for the perceived impacts of using the AVCS. Some population demographics were not represented in this survey due to the nature of the unit being surveyed, which may impact the validity of some results.
By quantifying the perceived efficacy of the AVCS as it relates to chronic musculoskeletal injury using a survey of pilot experience factors (flight hours, airframes, operating theatres, etc.), and by representing the maladaptive posture of the pilots with a computational simulation based on experimental pilot data; a full picture is developed of the risk of issue related to the near and long-term health of US Army Aviators. The aim is to expand the overall understanding of how vibration is impacting the musculoskeletal health of aviators and their perceived impact on lifelong health from the profession. The ultimate goal is to aid in the design of additional countermeasures to improve aviator spine health and to serve as a platform for optimization of systems like AVCS.S.M
A gravity-based mounting approach for large-scale cryogenic calorimeter arrays
No full textCryogenic calorimeters are among the leading technologies for searching for rare events. The CUPID experiment is exploiting this technology to deploy a tonne-scale detector to search for neutrinoless double-beta decay of 100 Mo. The CUPID collaboration proposed an innovative approach to assembling cryogenic calorimeters in a stacked configuration, held in position solely by gravity. This gravity-based assembly method is unprecedented in the field of cryogenic calorimeters and offers several advantages, including relaxed mechanical tolerances and simplified construction. To assess and optimize its performance, we constructed a medium-scale prototype hosting 28 Li 2 MoO 4 crystals and 30 Ge light detectors, both operated as cryogenic calorimeters at the Laboratori Nazionali del Gran Sasso (Italy). Despite an unexpected excess of noise in the light detectors, the results of this test proved (i) a thermal stability better than ±0.5 mK at 10 mK, (ii) a good energy resolution of Li 2 MoO 4 cryogenic calorimeters, (6.6 ± 2.2) keV FWHM at 2615 keV, and (iii) a Li 2 MoO 4 light yield measured by the closest light detector of 0.36 keV/MeV, sufficient to guarantee the particle identification requested by CUPID