Daniel Hogan Email & Phone Number

Vienna, Va, Us

Washington D.C. Metro Area

• Now starting work with large language models (LLMs).• Proposed and implemented a novel way to geolocate photographs (i.e., figure out where a photo was taken) using artificial intelligence. The method builds on prior work with satellite imagery (see below) to enable worldwide coverage. A browser-based data visualization system, with a Leaflet JS frontend and a FastAPI/PyTorch backend, furnishes a user-friendly interface.• Designed an AI solution for locating objects of interest in sidescan sonar data, by creating and open-sourcing a Python module for computer vision anomaly detection.• Developed a dataset of overhead aircraft photographs, using field data collected with a “SkyScan” camera system that slues towards the origin of ADS-B broadcasts. Created two versions of an object detection model for aircraft type: one for conventional GPUs and one for AI at the edge using a Coral USB accelerator.• Conducted the data analysis for a study comparing the quality of aircraft image labels from human labelers versus from concurrently-collected ADS-B information.• Rapidly adapted prior work to demonstrate geolocating photographs from Ukraine.• Led a team in executing a project that I proposed to geolocate photographs with cross-view image geolocalization. The project entailed creating a dataset of photograph/satellite image pairs, implementing a model, and doing comparative studies with different versions of the model and dataset.Throughout In-Q-Tel / SpaceNet tenure thus far:• Authored or coauthored 19 blog posts to share work with public• Major contributor to six open-source GitHub repos• Verbal and written communication with stakeholders in various contexts: customer briefings, internal updates, coordinating with vendors, presenting at conferences

Washington D.C. Metro Area

• Member of CosmiQ Works, the In-Q-Tel Labs branch focused on geospatial analytics with machine learning.• Developed an image processing library with Python/GDAL that can reduce an entire image processing workflow to a single line of code.• Undertook a study of training data needs for geospatial deep learning. Insights included a practical heuristic for estimating the effect of increased training data.• [See also “SpaceNet,” below.]

Washington D.C. Metro Area

• Supported SpaceNet 6 and SpaceNet 7 public geospatial data science challenges.• For SpaceNet 7, co-created a new evaluation metric for geospatial time series data.• For SpaceNet 7, set up a baseline model demonstration with cloud computing.• For SpaceNet 6, preprocessed synthetic aperture radar (SAR) imagery for a multimodal dataset, following self-directed learning about working with SAR.• For SpaceNet 6, built baseline model for extracting building footprints from SAR.

Greater Boston Area

• Developed software to identify a bird's species from an audio recording of its call. The audio is cleaned and processed using SciPy to produce spectrograms, which are then classified using a convolutional neural network in Keras. The technique can identify 150 bird species from the midwestern and eastern United States to 44% accuracy.• Built an Ajax webapp for bird-watchers based on this software.

San Francisco Bay Area

Graduate Student Researcher:• Implemented an interactive data visualization web portal with an Ajax frontend and Python/MySQL backend, which became a widely-used analysis tool across LUX, a 100-member physics collaboration.• Built and maintained a pipeline using GridFTP for transferring and archiving the collaboration's experimental data, of which there was in excess of 500TB.• Evaluated various algorithms for using the pulse shape of photomultiplier output to distinguish signal from background in a rare-event search particle physics experiment.• Deployed code at NERSC (National Energy Research Scientific Computer Center) supercomputer facility and worked with NERSC staff to solve technical issues.• Calculated the expected neutrino spectrum for a nuclear reactor experiment, using Matlab.• Developed Monte Carlo simulation of energy deposition in a new type of particle detector, using C++ with the GEANT4 library.Graduate Student Instructor:• Taught physics to diverse audiences ranging from engineering students to non-scientists, educating nearly 200 students over six semesters.• Administered a class of 500+ students, coordinating the efforts of lecturers, students, teaching assistants, and others.• Held leadership roles in two science policy student organizations.

Lawrence, Ks

• Developed a C++ Monte Carlo simulation of hypothetical subatomic particles (ultrarelativistic magnetic monopoles) interacting with the RICE particle physics experiment, constraining a key parameter with 100 times more precision than before.• Conducted a search for new resonance particles by analyzing four years of data from the CLEO particle detector.• Developed static and animated visualizations of the output of a climate model for a study of ancient climate change and mass extinctions.• Presented my work internationally (Italy, Sweden, US) and gave an invited talk (Penn State).• Lead author of a paper, and coauthor of five others, as an undergraduate.

Doctor Of Philosophy - Phd, Physics

Master Of Arts - Ma, Physics

Bachelor Of Science - Bs, Physics And Mathematics