Language Models for Veteran Suicide Prevention (LM4VSP)
On September 22, 2021, the US House Veterans Affairs Committee held a hearing entitled, “Veteran Suicide Prevention: Innovative Research and Expanded Public Health Efforts” [1]. The hearing followed the release of annual data from the US Department of Veterans Affairs showing that the disproportionate rate of veteran suicide is a public health crisis [2]. Although there is no single reason why veterans commit suicide, evidence suggests that stable housing, financial security, access to healthcare, addressing social isolation and loneliness, and treating the effects of trauma are important components of a comprehensive suicide prevention strategy; all of which require coordination and cooperation across families, communities, and at all levels of government. Recent advances in AI, and specifically LMs, have the potential to help lessen the effects of social isolation/loneliness and trauma. According to a 2019 report on “Sleep and timing of death by suicide among US Veterans 2006-2015” [McCarthy, et al. 2019], the raw proportion of veteran suicides peaks between the hours of 1000 and 1200; however, the peak prevalence of suicide, after accounting for the population being awake, is between the hours of 0000 and 0300 (p < 0.00001, F = 0.88). The highest Standardized Incidence Ratio (SIR) is at midnight; US Veterans are eight times more likely to die by suicide than expected given the population awake (SIR = 8.17; 95% Confidence Interval = 7.45-8.94). In other words, when clinical help is likely unavailable or difficult to access, technology has the potential to provide critical assistance. Recent advances in the field of natural language processing have allowed LMs (for example, Chat GPT (Generative Pre-trained Transformer)) to be fine-tuned using reinforcement learning based on human feedback [3]. Prior efforts have shown that it is possible to create a highly conversational model based on 40,000 pieces of feedback [4]. Other recent research work in the field suggests promising results in prompt engineering [5], using memory-based machine learning with dramatic improvements in the LM’s ability to stay on task, and return more accurate and precise results. [6] LM4VSP seeks to develop a clinical co-pilot based on LMs specific to the mental health subdomain, and in close collaboration with mental health subject matter experts (SMEs). The goal is for the LM4VSP clinical co-pilot to enable caregivers to offer around-the-clock assistance and accelerate their understanding and assessments and improve the effectiveness of intervention.
Multi-factor Continuous Authentication for Wearable Defense Equipment (MCAWDE)
Internet-of-Things (IoT) devices have seen unprecedented growth [1] and yet remain one of the weakest links when it comes to cybersecurity [2]. User and device authentication for battery-operated IoT devices (e.g., smartphones and wearables) is challenging due to limitations on the available energy, user interface, and processing power [3]. Over the last few years, multiple authentication techniques have been developed to address these challenges, for example, location-based authentication techniques [4] and gait-based authentication techniques [5]. However, existing techniques face challenges in terms of performance overhead, power consumption, and overall efficiency of cryptographic operations [6]. To address these challenges, DARPA seeks novel, continuous [7], multi-factor authentication [8] solutions for small weight and power devices.
Non-destructive Viral ID
Warfighters are often deployed to emerging disease hotspots. To help mitigate potential exposure risks, DoD entities tasked with force health protection [1] rapidly assess, often on-site, a range of sample types for potential biological threats. Current rapid on-site identification (ID) methods include sequencing [2,3] and lateral flow assays (aka dipstick tests) [4], which destroy both the sample & the organism(s), inhibiting further analysis. While sequencing costs continue to decrease and cheap lateral flow assays continue to increase in organism scope, forward operators often triage the number of sites and samples collected due to resource, personnel, and time constraints [5]. Recent advances in metasurfaces [6,7]7, optical waveguides [8], microfluidics [9], and (super)high resolution imaging [10], now suggest accurate organism ID and viability maintenance can co-occur; however, current non-destructive systems lack field utility due to their lab-centric designs [7]. This SBIR will address both the significant limitations to rapid, on-site biosurveillance in resource constrained environments and the lab-centric designs for non-destructive pathogen ID by developing human-portable, low size, weight, and power (SWaP) technology to rapidly and non-destructively ID viruses on-site and in the field. Final Phase II prototypes must be: ≤ 1 ft3; ≤ 5 lbs; and ≤ 200-Watt peak power input, with all SWaP requirements inclusive of power delivery mechanisms, software/data processing, consumables, and reagents, and should function with a wide range of simulated clinical samples (e.g., blood, saliva, nasal swabs, etc.) and contrived environmental samples (swipes/wipes, chicken rinse, etc.). Systems should non-destructively ID viruses faster than current state-of-the-art sequencing and lateral flow assays (≤ 15 min per sample, not including sample pre-processing) while maintaining viral infectivity for downstream lab-based analysis. By the end of Phase II systems should ID viruses faster than current state-of-the-art sequencing and lateral flow assays (≤ 15 min per sample), in the field sample pre-processing, if any, should be no more than 20 minutes per sample, and viral ID should be independent of cloud connectivity (e.g., database access, analysis and ID can occur on the device without cloud access).
Request for Information (RFI) - Strategic Technology Office (STO)
This RFI is to provide a mechanism to inform the office of new capabilities and concepts and the potential performers who will provide them. DARPA will use this information to determine the companies and individuals that STO would invite to have further substantive discussions to inform future projects and programs in appropriately classified settings when necessary.
TurboFCL
The Department of Defense (DOD) aims to increase small business participation within the industrial base and has boosted its involvement in providing goods, services, and research and technology to support national defense. However, significant obstacles persist that dissuade non-traditional vendors from entering the defense ecosystem of contracting and acquisition. Prominent among them is the complex industrial facility security clearance (FCL) process. The FCL process is instrumental in vetting businesses for their capability to handle classified information and bid on classified contracts. However, this process can be daunting for small businesses and non-traditional vendors due to fragmented information, heavy reliance on “government speak”, and the significant time and resources required to complete the process.DARPA’s Bringing classified Innovation to Defense and Government Systems (BRIDGES) has sought to connect innovation from small and nontraditional companies to classified Department of Defense (DoD) research and development (R&D) efforts, specifically by sponsoring innovative companies for facility clearances so that they can directly interact with DoD customers at classified levels. During execution of the BRIDGES program, DARPA observed firsthand the challenges that these small companies faced while going through the FCL process, most notably:• Misunderstanding what information was required on the required forms• Filling out forms incorrectly• Not fully understanding submission timelinesThis resulted in additional workload on DARPA security to answer questions and provide guidance, delays in processing for FCLs, and some FCL packages being rejected by DCSA.To address these challenges, DARPA is looking to develop a software application, leveraging recent advancements in language model technology, that can assist small and nontraditional companies apply for their FCLs, as well as their associated personnel clearances (PCLs). Information on the FCL and PCL processes, while disaggregated, is publicly available. A targeted language model can be trained on the processes, forms, and other related information, and distill this information into more easily understandable wording. Then, just like the popular and commercially available TurboTax software, this model can provide appropriate prompts to users to gather the required information, and then autocomplete the required forms. Additionally, it could provide insights into the deadlines and timelines, and provide a “chat bot” feature to answer questions about the processes.In this SBIR, the proposer will develop a prototype system ready to be tested by representatives from the target audience (small businesses and nontraditional contractors who are not familiar with the FCL and/or PCL processes) during the base effort. If successful, they will then oversee test and evaluation of the prototype by end users, ideally performers on DARPA’s BRIDGES program. During this test and evaluation period, the proposer will verify both the inputs and outputs of the application, collect feedback from users, and incorporate appropriate updates. The end goal will be a system that removes errors from FCL packages, improves processing timelines by removing delays/extension requests, while also improving the user experience and reducing the time and effort required by the end user.
