A5065521265- Atomic and Subatomic Physics Research
- Robotics and Sensor-Based Localization
- Particle physics theoretical and experimental studies
- Software Engineering Research
- Advanced Malware Detection Techniques
- Advanced Image and Video Retrieval Techniques
- Software Reliability and Analysis Research
- Target Tracking and Data Fusion in Sensor Networks
- Atomic and Molecular Physics
- Dark Matter and Cosmic Phenomena
- Quantum, superfluid, helium dynamics
- Radiomics and Machine Learning in Medical Imaging
- COVID-19 diagnosis using AI
- Lung Cancer Diagnosis and Treatment
- Muon and positron interactions and applications
- Radiation Detection and Scintillator Technologies
- Human-Automation Interaction and Safety
- Risk and Safety Analysis
- Video Surveillance and Tracking Methods
- Particle Accelerators and Free-Electron Lasers
- Robot Manipulation and Learning
- Quantum Chromodynamics and Particle Interactions
- Anomaly Detection Techniques and Applications
- Advanced Data Storage Technologies
- Robotic Path Planning Algorithms
Draper Laboratory
2018-2025
Massachusetts Institute of Technology
2012-2021
California Institute of Technology
2010-2019
Brown University
1993
Increasing numbers of software vulnerabilities are discovered every year whether they reported publicly or internally in proprietary code. These can pose serious risk exploit and result system compromise, information leaks, denial service. We leveraged the wealth C C++ open-source code available to develop a largescale function-level vulnerability detection using machine learning. To supplement existing labeled datasets, we compiled vast dataset millions functions it with carefully-selected...
We introduce a new computer aided detection and diagnosis system for lung cancer screening with low-dose CT scans that produces meaningful probability assessments. Our is based entirely on 3D convolutional neural networks achieves state-of-the-art performance both nodule malignancy classification tasks the publicly available LUNA16 Kaggle Data Science Bowl challenges. While systems are typically designed optimized their own, we find it important to consider coupling between components....
Thousands of security vulnerabilities are discovered in production software each year, either reported publicly to the Common Vulnerabilities and Exposures database or internally proprietary code. often manifest themselves subtle ways that not obvious code reviewers developers themselves. With wealth open source available for analysis, there is an opportunity learn patterns bugs can lead directly from data. In this paper, we present a data-driven approach vulnerability detection using...
The OLYMPUS Collaboration reports on a precision measurement of the positron-proton to electron-proton elastic cross section ratio, R_{2γ}, direct measure contribution hard two-photon exchange section. In measurement, 2.01 GeV electron and positron beams were directed through hydrogen gas target internal DORIS storage ring at DESY. A toroidal magnetic spectrometer instrumented with drift chambers time-of-flight scintillators detected elastically scattered leptons in coincidence recoiling...
Deep learning has the potential to dramatically impact navigation and tracking state estimation problems critical autonomous vehicles robotics. Measurement uncertainties in systems based on Kalman other Bayes filters are typically assumed be a fixed covariance matrix. This assumption is risky, particularly for "black box" deep models, which uncertainty can vary unexpectedly. Accurate quantification of multivariate will allow full used more safely reliably these applications. We show how...
A precise measurement of the neutron decay β asymmetry A₀ has been carried out using polarized ultracold neutrons from pulsed spallation source at Los Alamos Neutron Science Center. Combining data obtained in 2008 and 2009, we report = -0.119 66±0.000 89{-0.001 40}{+0.001 23}, which determine ratio axial-vector to vector weak coupling nucleon g{A}/g{V}=-1.275 90{-0.004 45}{+0.004 09}.
We present a detailed report of measurement the neutron $\beta$-asymmetry parameter $A_0$, parity-violating angular correlation between spin and decay electron momentum, performed with polarized ultracold neutrons (UCN). UCN were extracted from pulsed spallation solid deuterium source via transport through 7-T magnetic field. The then transported an adiabatic-fast-passage spin-flipper field region, prior to storage in cylindrical volume situated within 1-T $2 \times 2\pi$ solenoidal...
In this work, we introduce pose interpreter networks for 6-DoF object estimation. contrast to other CNN-based approaches estimation that require expensively annotated data, our network is trained entirely on synthetic data. We use masks as an intermediate representation bridge real and synthetic. show when combined with a segmentation model RGB images, synthetically able generalize Our end-to-end system runs in real-time (20 Hz) live without using depth information or ICP refinement.
Motivated by the problem of automated repair software vulnerabilities, we propose an adversarial learning approach that maps from one discrete source domain to another target without requiring paired labeled examples or and domains be bijections. We demonstrate proposed is effective technique for repairing performing close seq2seq approaches require pairs. The Generative Adversarial Network application-agnostic in it can applied other problems similar code repair, such as grammar correction...
Introduction Future concepts for airborne autonomy point toward human operators moving out of the cockpit and into supervisory roles. Urban air mobility, package delivery, military intelligence, surveillance, reconnaissance (ISR) are all actively exploring such or currently undergoing this transition. Supervisors these systems will be faced with many challenges, including platforms that operate outside visual range need to decipher complex sensor telemetry data in order make informed safe...
A puzzling discrepancy exists between the values of proton charge radius obtained using different experimental techniques: elastic electron-proton scattering and spectroscopy electronic muonic hydrogen. The is defined through slope electric form factor, $G_E(Q^2)$, at zero four-momentum transfer, which inaccessible in experiments. We propose a novel method for extracting from data over broad $Q^2$ range rather than attempting to directly determine $G_E$ $Q^2 = 0$. This relates its transverse...
Deep learning has the potential to dramatically impact navigation and tracking state estimation problems critical autonomous vehicles robotics. Measurement uncertainties in systems based on Kalman other Bayes filters are typically assumed be a fixed covariance matrix. This assumption is risky, particularly for "black box" deep models, which uncertainty can vary unexpectedly. Accurate quantification of multivariate will allow full used more safely reliably these applications. We show how...
We report the first measurement of average electron-proton and positron-proton elastic scattering cross sections. This lepton charge-averaged section is insensitive to leading effects hard two-photon exchange, giving more robust access proton’s electromagnetic form factors. The was extracted from data taken by OLYMPUS experiment at DESY, in which alternating stored electron positron beams were scattered a windowless gaseous hydrogen target. Elastic events identified coincident detection...
Received 10 November 2010DOI:https://doi.org/10.1103/PhysRevLett.105.219903© 2010 The American Physical Society
Due to the increased role of autonomous robots in accomplishing a variety challenging tasks alongside humans, it is essential for human operator establish appropriate trust towards these systems. To this end, we present step generating competency-aware agents that are able communicate their self-confidence given task. We develop and analyze an model-based reinforcement learning UAV ISR agent uses neural network based learned model world uncertain planner generate series simulated...
For autonomous agents to act as trustworthy partners human users, they must be able reliably communicate their competency for the tasks are asked perform. Towards this objective, we develop probabilistic world models based on deep generative modelling that allow simulation of agent trajectories and accurate calculation tasking outcome probabilities. By combining strengths conditional variational autoencoders with recurrent neural networks, model can probabilistically forecast over long...
Cross-view geolocalization, a supplement or replacement for GPS, localizes an agent within search area by matching images taken from ground-view camera to overhead satellites aircraft. Although the viewpoint disparity between ground and makes crossview geolocalization challenging, significant progress has been made assuming that access panoramic camera. For example, our prior work (WAG) introduced changes in discretization, training loss, particle filter weighting enabled city-scale...
The UCNA experiment was designed to measure the neutron $\beta$-asymmetry parameter $A_0$ using polarized ultracold neutrons (UCN). UCN produced via downscattering in solid deuterium were transport through a 7 T magnetic field, and then directed 1 solenoidal electron spectrometer, where decay electrons detected detector packages located on two ends of spectrometer. A value for extracted from asymmetry numbers counts packages. We summarize all results experiment, obtained during run periods...
Increasing numbers of software vulnerabilities are discovered every year whether they reported publicly or internally in proprietary code. These can pose serious risk exploit and result system compromise, information leaks, denial service. We leveraged the wealth C C++ open-source code available to develop a large-scale function-level vulnerability detection using machine learning. To supplement existing labeled datasets, we compiled vast dataset millions functions it with carefully-selected...