Writings

All of my publications, patents, and blogs spanning Machine Learning, Software Engineering, Systems, and Fun listed in reverse chronological order.

PUBLICATION: Leveraging Deep Learning for High-Resolution Optical Satellite Imagery From Low-Cost Small Satellite Platforms

Published in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing: In this work, we detail a deep-learning-based approach, which improves optical satellite imagery to five times the original pixel-based resolution without the need or expense of increasing the capabilities of the imager through larger telescope apertures. The approach—demonstrated on Terran Orbital's GEOStare SV2 mission imagery—is generally applicable to any optical satellite image and is agnostic to the mission, satellite manufacturer, optical payload specifications, or data source. This capability provides a basis for small satellite missions and constellations—and their optical payloads—to rival the native hardware-based resolutions available through larger satellites with wider telescope apertures at a significantly reduced cost.

PATENT: (Intruder Estimation) Wireless device detection systems and methods incorporating streaming survival modeling for discrete rotating identifier data

This is one in a series of patents related to the RF signaling work at Ubiety Technologies. This particular patent contains claims related to the calculation of an intruder probability given our streaming survival modeling subsystem for device uniqueness from psuedo-randomization of underlying RF data.

BLOG: The Great Chicago Oyster Happy Hour Journey

Published on Medium: As I was taking a break from working between jobs, I wanted to hit all of the Chicago Oyster Happy hours in an optimal mannar. This writing documents that journey.

PATENT: Correlation of WiFi, Bluetooth, cellular signaling to one mobile device

This is one in a series of patents related to the RF signaling work at Ubiety Technologies. This particular patent contains claims related to the correlation of multi model RF signaling such as WiFi, Bluetooh, and Cellular signals together for determining uniqueness of electronic devices.

PATENT: IMSI / TMSI / RNTI mapping to a particular mobile device

This is one in a series of patents related to the RF signaling work at Ubiety Technologies. This particular patent contains claims related to the reverse engineering of the uplink and downlink LTE signals to signal unique cellular devices.

PUBLICATION: Scheduling the NASA Deep Space Network with Deep Reinforcement Learning

Published in 2021 IEEE Aerospace Conference: With three complexes spread evenly across the Earth, NASA's Deep Space Network (DSN) is the primary means of communications as well as a significant scientific instrument for dozens of active missions around the world. A rapidly rising number of spacecraft and increasingly complex scientific instruments with higher bandwidth requirements have resulted in demand that exceeds the network's capacity across its 12 antennae. The existing DSN scheduling process operates on a rolling weekly basis and is time-consuming; for a given week, generation of the final baseline schedule of spacecraft tracking passes takes roughly 5 months from the initial requirements submission deadline, with several weeks of peer-to-peer negotiations in which mission planners and DSN personnel iteratively update their respective schedules based on inputs from the other. This paper proposes a deep reinforcement learning (RL) approach to generate candidate DSN schedules from mission requests and spacecraft view period data with demonstrated capability to address real-world operational constraints. A deep RL agent is developed that takes mission requests for a given week as input, and interacts with a DSN scheduling environment to allocate tracks such that its reward signal is maximized. A comparison is made between an agent trained using Proximal Policy Optimization and its random, untrained counterpart. The results represent a proof-of-concept that, given a well-shaped reward signal, a deep RL agent can learn the complex heuristics used by experts to schedule the DSN. A trained agent can potentially be used to generate candidate schedules to bootstrap the scheduling process and thus reduce the turnaround cycle for DSN scheduling.

PATENT: Identifying mobile devices via IMSI, TMSI, and uplink and downlink RNTI; determining location via SINR & RSSI; geofencing

This is one in a series of patents related to the RF signaling work at Ubiety Technologies. This particular patent contains claims related to using the IMSI, TMSI, and RNTI from pieced together uplink and downlink reverse engineering to identify unique cellular devices on 4G. We also go into determining location with the SINR and RSSI data along with geofencing capabilities from our sensors.

PRESENTATION: Single Event Metadata Analysis

Presented at the 2020 NEPP Electronics Technology Workshop: Work on using metadata of COTS electronics to model Single Event Latchup (SEL) radiation events. Also delves into using image segmentation in order to quantify cross section of various silicon componets as inputs to SEL prediction system.

PUBLICATION: Vision-based estimation of driving energy for planetary rovers using deep learning and terramechanics

Published in IEEE Robotics and Automation Letters: This letter presents a prediction algorithm of driving energy for future Mars rover missions. The majority of future Mars rovers would be solar-powered, which would require energy-optimal driving to maximize the range with limited energy. The essential and arguably the most challenging technology for realizing energy-optimal driving is the capability to predict the driving energy, which is needed to construct an energy-aware cost function for path planning. In this letter, we propose vision-based algorithms to remotely predict the driving energy consumption using machine learning. Specifically, we develop and compare two machine-learning models in this letter, namely VeeGer-EnergyNet and Veeger-TerramechanicsNet, respectively. The former is trained directly using recorded power, while the latter estimates terrain parameters from the images using a simplified-terramechanics model, and calculate the power based on the model. The two approaches are fully automated self-supervised learning algorithms. To combine RGB and depth images efficiently with high accuracy, we propose a new network architecture called Two-PNASNet-5, which is based on PNASNet-5. We collected a new dataset to verify the effectiveness of the proposed approaches. Comparison of the two approaches showed that Veeger-TerramechanicsNet had better performance than VeeGer-EnergyNet.

PUBLICATION: Automated Machine Learning as a Service for the Earth Sciences

Published in AGU Fall Meetings: NASA's JPL team is developing an Automated Machine Learning (AutoML) environment, named MARVIN, to handle large Earth or Space science datasets. MARVIN, part of the DARPA Data Driven Discovery of Models (D3M) program, automates the composition of ML pipelines. It includes a library of ML 'primitives' for tasks like preprocessing and feature extraction, and automates the creation of Docker containers for execution on a Kubernetes cluster. The system functions like an 'app store' for ML, allowing new capabilities to be added easily. Currently, MARVIN contains over 400 datasets/problems and over 250 primitives.

BLOG: Exploring Stochastic Gradient Descent with Restarts (SGDR)

Published on Medium: This blog post discusses my journey into deep learning, focusing on the concept of Stochastic Gradient Descent with Restarts (SGDR). The author explains the basics of deep learning, including the use of loss functions and the idea of gradient descent to minimize these functions. The post then introduces SGDR, a method that resets the learning rate periodically to avoid getting stuck in a local minimum. The author suggests that this method, especially when combined with a progressively lengthening cycle, can improve the performance of deep learning models. The post emphasizes that many cutting-edge ideas in deep learning are simple modifications of base methodologies, and are relatively easy to understand and implement.

PUBLICATION: Population of 13Be in a nucleon exchange reaction

Published in Physical Review C: The neutron-unbound nucleus Be 13 was populated with a nucleon exchange reaction from a 71 MeV/u secondary B 13 beam. The decay-energy spectrum was reconstructed using invariant mass spectroscopy based on Be 12 fragments in coincidence with neutrons. The data could be described with an s-wave resonance at E r= 0.73 (9) MeV with a width of Γ r= 1.98 (34) MeV and a d-wave resonance at E r= 2.56 (13) MeV with a width of Γ r= 2.29 (73) MeV. The observed spectral shape is consistent with previous one-proton removal reaction measurements from B 14.

PUBLICATION: Low-lying neutron unbound states in 12Be

Published in Physical Review C: The neutron decay of an unbound resonance in Be 12 has been measured at 1243±21 keV decay energy with a width of 634±60 keV. This state was populated with a one-proton removal reaction from a 71 MeV/u B 13 beam incident upon a beryllium target. The invariant mass reconstruction of the resonance was achieved by measuring the daughter fragment in coincidence with neutrons. Despite being above the 2 n separation energy, the state decays predominantly by the emission of one neutron to Be 11, setting an upper limit on the branching ratio for the two-neutron decay channel to Be 10 of less than 5%. From the characteristics of the population and decay of the resonance, it is concluded that this state cannot correspond to the previously observed state at 4580±5 keV.