Problems

These problems have come from a wide variety of branches of the U.S. Department of Defense.

Additional problems may continue to be added - and problems posted here may experience minor changes - so check back regularly.

Do you see a problem you like, but need to find team members? Fill out the team formation form.

Please note: The problem set below is subject to change. Please sign up for updates by entering your email address on the homepage to receive email updates.

Teams with their own startup idea can apply as well!!

CHALLENGE: Assemble a team of 4-5 Stanford students from any college or department to develop a technology that DoD/IC needs to adopt (if accepted we will help you find a sponsor)
CATEGORY: Open
BACKGROUND: You and your teammates are convinced you have a great idea and technology to solve a problem for National Security. Apply as a team to Hacking4Defense and we will help you find a sponsor, validate your problem, and help you figure out if your idea is desirable, feasible, viable. We help you find product/mission market fit.

If you'd like to explore this route, please reach out to Nate Simon (simonnj@stanford.edu) or Jeff Decker (jmdecker@stanford.edu).

2020 Problems will be posted soon.

2019 Problems:
Updated March 15, 2019.

#1 Back End Visibility into Front End Usage

CHALLENGE
Marine Corps Network Efficiency Laboratory Engineers need a better way to understand the universe of services and features users most often employ in order to efficiently manage network traffic.

BACKGROUND
Global Combat Support System - Marine Corps (GCSS - MC) is a web-enabled system that serves as a single point of entry for all USMC logistics requirements. This system allows users to request parts, track service requests, access maintenance production reports, and view inventory data, as well as several other logistics needs.
Marine Corps Network Efficiency Laboratory (MCNEL) comprises of a group of 9 engineers that focus on supporting the GCSS-MC program office, in addition to several other programs of record (PoR). Their responsibilities include performing network traffic analysis on both the user and server side. On the user side, GCSS-MC is usually operated in a degraded (lower network quality) environment. USMC systems are allocated a small amount of bandwidth aboard Navy ships and GCSS-MC has an even smaller portion of bandwidth. MCNEL engineers are tasked with tending to the technical issues that arise as a consequence of this operational environment and increasing its efficiency.
Presently, the engineers only have visibility into user behavior in the form of clicks and other, non-descriptive data. In order to increase system efficiency by prioritizing bandwidth based on usage, MCNEL engineers need insight into user behavior to understand how best to perform tasks such as managing network traffic. Having visibility into what system features are most frequently used, what features are used in tandem, and how users access features (sequences), etc., would allow netw

OPERATIONAL CONSTRAINTS
User data is stored on local servers in Kansas, MS and would need to be shared with MCNEL engineers.
PROBLEM SPONSOR: U.S. Marine Corp

#2 Hit the Easy Button

CHALLENGE
Program Management Competency teams need the ability to quickly search commercial and government research in order to effectively respond to requirements requests.

BACKGROUND
The Marine Corps Systems Command (MCSC) is the acquisition authority for the US Marine Corps (USMC). The MCSC receives requirements from operating forces on what is needed for Marines to accomplish their missions. When these requests come in, program management competency teams search across a series of databases on technology that may meet the requirements. There are thousands of databases and search functions have varying degrees of strength. As a result, program management teams spend between five to ten hours each week digging through repositories, often times with inconclusive results.
Currently, the speed of the search is limited by identifiers in the existing documentation as well as the program management team’s ability to associate identifiers with the requirements documentation. Failing to align these key words will yield a false failure, where the data does exist but the operator fails to query it. Furthermore, in previous incidences where requirements of a similar nature have been requested, the previous assessment on the availability of that technology is undocumented and unsearchable. For example, if a requirements document requested a type of laser technology for Marines and a team ran a search across the databases to conclude no such technology was commercially available, this assessment would not be recoverable a year later. The current team unknowingly wastes time performing redundant searches, and querying new data that was not previously checked. Copious hours are spent diving into databases, resulting in a painstakingly slow assessment solutions to meet the Marine Corps’ current battlefield challenges.

OPERATIONAL CONSTRAINTS
Most of the databases are open source and are parse-able to Excel Sheets and SQL

PROBLEM SPONSOR: U.S. Marine Corp

#3 Tactical Software Defined Radios for Electronic Warfare

CHALLENGE
Infantry units and radio operators need a way to access an electronic warfare capacity in order to introduce electronic warfare into operations and training.

BACKGROUND
In the past several years, near peer-competitors Russia and China have heavily prioritized electronic warfare (EW). They have invested in EW assets, primarily software defined radios, that allows them the ability to use the electromagnetic spectrum – signals such as radio, infrared, or radar – to sense and communicate. Furthermore, they have distributed these capabilities down to the tactical level, meaning even small units have EW capabilities. For example, Russia has used EW to target Ukrainian troops in Crimea when the troops attempt to communicate on their radios. The U.S. has never trained to counter this kind of warfare. The Marine Corps has invested in high quality equipment, but this equipment is available only at the Division level, making it practically inaccessible to units for training and day-to-day use.
While the equipment is costly – sometimes upwards of $1 million – there is medium-quality commercially available gear for as low as $20, with high-quality commercial technology available for $200 – 300. However, radio operators and infantry platoons are stymied by a long acquisition process for acquiring commercial technology, which can take up to three years. For this reason, units have experimented with building rudimentary tools, in order to more quickly introduce EW systems at the tactical level and allow units to receive immediate feedback on their Radio Frequency (RF) signature.

OPERATIONAL CONSTRAINTS
• The system should be man-portable
• The cost must be under $1000 per unit
• Able to operate between 0-140o F
• Able to be stored in a waterproof case, but the system itself does not have to be waterproof
• Capable of detecting and providing Lines of Bearing (LOBs) of emissions between 30 MHz and 3000 MHz
• Capable of providing effective localized jamming on VHF

PROBLEM SPONSOR: U.S. Marine Corps

#4 Analyzing Activity across Multiple Objects

CHALLENGE
Geospatial analysts need a tool to automatically identify and analyze object activity that is detected from Remote Autonomous Systems (RAS) footage in order to effectively produce actionable intelligence.

BACKGROUND
When receiving reconnaissance Remote Autonomous Systems (RAS) footage, the United States Air Force Air Combat Command (ACC) has a computer program that automatically “classifies” or identifies the objects captured in the footage. While this computer program identifies objects in the RAS footage, analysts do not have the ability to understand activity from multiple detected objects. As such, the only way to provide any sort of insight regarding object activity is to have analysts manually watch and report back on the RAS footage.
For this reason, the ACC has a team of geospatial analysts manually scan the footage when analyzing for object activity and larger life-cycle trends, Each team is comprised of at least two analysts, with one watching the footage and calling out any suspicious objects, their movements and activity, while the second analyst notes down the observations. These notes are captured in a database where identified objects can be cross-referenced and automatically searched for any significance. This process is not only error-prone but also time-intensive, hindering the geospatial analysts' ability to produce timely and meaningful intelligence.

OPERATIONAL CONSTRAINTS
• Solutions should take into account the Motion Imagery Standards Board (MIS-B) quality of the video being analyzed.

PROBLEM SPONSOR: US Air Force

#5 Dirty Disposal

CHALLENGE
Develop an effective method for Explosive Ordnance Disposal technicians to identify obscured, dirty ordnance in order to properly dispose of it.

BACKGROUND
Explosive ordnance disposal technicians are responsible for identifying and disarming ordnance during missions. Ordnance are unexploded weapons that, when triggered (e.g. if a warfighter walks on a part of it), can be deadly.
To safely disarm ordnance, technicians must first identify the ordnance. In the field, they photograph visible parts and upload the photographs to the Ordnance Threat/Target Automated Recognition (OTTAR) system. The system relies on an algorithm that compares submitted photographs with already catalogued ordnance. Once the ordnance has been identified, OTTAR provides instructions to allow technicians to safely dispose of it.
However, ordnance is often partially obscured or dirty (covered in dirt or other materials), making it difficult for OTTAR to identify the ordnance in question. Additionally, OTTAR may not recognize photographed ordnance, even if previously catalogued, as it may not include all detection angles. When ordnance is difficult to identify, technicians must manually clear the debris or photograph it from another angle, to upload a recognizable picture to OTTAR. Manual clearing and/or re-photographing an easily triggered ordnance can put technicians’ lives at risk.
Therefore, technicians need a more effective method for identifying obscured, dirty ordnance from all angles.

OPERATIONAL CONSTRAINTS
• Classification must distinguish between similar ordnance types

PROBLEM SPONSOR: Naval Research Lab

#6 Hide My Signature

CHALLENGE
Unmanned systems pilots need to eliminate unmanned air and ground systems’ electromagnetic spectrum (EMS) signature in order to protect the mission and the locations of the force.

BACKGROUND
Unmanned air and ground systems are used on the battlefield to identify obstacles, detect the enemy, etc. Often, these systems must gather environmental and positional data in GPS-denied environments. To do so, most autonomous navigation systems emit signals on the electromagnetic spectrum (EMS) and use the response to increase their environmental awareness. However, broadcasted signals may be used to identify and target both the system and the pilot’s location.
There are unmanned systems that do not broadcast an EMS signal to determine their relative position – instead, they use two cameras to correlate the distance between themselves and objects in their sight. However, these unmanned systems are costly, and the cameras place a high processing power burden on the overall system.

OPERATIONAL CONSTRAINTS
• Solution must be operational in GPS-denied environments

PROBLEM SPONSOR : Joint Artificial Intelligence Center

#7 Is that video real or fake?

BACKGROUND
The ability to manipulate video, sound, and images to create wholly realistic fake files, or “deep fakes,” is of significant concern to the Intelligence Community. Controversial videos on social media spread quickly and can lead to popular unrest.

Advances in artificial intelligence technology make it easy to create fake videos with fake audio.

The Intelligence Community and In-Q-Tel, a not-for-profit strategic investor, is interested in exploring how emerging technology can be leveraged to detect a fake video. Ultimately, this would inform countermeasures and recommendations to quickly identify fake videos.

CHALLENGE
Analysts need to detect fake video content in order to determine its authenticity and credibility.

LIMITATIONS

Must be within existing legal and policy parameters
Necessary to consider repeatability and provide confidence levels

PROBLEM SPONSOR: IQT, strategic investor for national security

#8 Semi-Supervised Learning for Text Analytics Models

BACKGROUND
Within the Office of the Director of National Intelligence (ODNI), the Augmenting Intelligence using Machines (AIM) Initiative iHub seeks to ease the existing labor requirements for natural language processing (NLP) data labeling. Modern deep learning-based NLP models require extensive labeled training data. This quickly becomes a hard and expensive problem because data labeling farms such as Amazon’s crowdsourcing marketplace, Mechanical Turk, are not an option in the Intelligence Community, as much of the data is classified. As a result, intelligence analysts are routinely pulled off mission tasks to support data labeling efforts, necessitating alternatives for labeling text.

Recent research in semi-supervised learning has shown that engineers can train NLP models with significantly fewer labels with similar precision and recall. For this reason, ODNI seeks to explore alternatives such as distant supervision labeling, one example being open source libraries.

CHALLENGE
The Augmenting Intelligence using Machines Program Staff needs recommendations for quickly labeling training sets for entity extraction and entity linking models.

Long Term Goal: The AIM iHub seeks algorithm implementations for semi-supervised data labeling.

LIMITATIONS

Open source libraries like Snorkel (developed from Chris Re’s lab at Stanford) and spaCy are useful resources
High likelihood that IARPA could provide training data sets

PROBLEM SPONSOR: Office of the Director of National Intelligence