Kevin Leach's research while affiliated with Vanderbilt University and other places

... In addition, the attacker is capable of spoofing system IDs. The work by Highnam et al. [117] described a realistic scenario of many drones operating under the MAVLink protocol being compromised. By capturing a flight mission's system ID and spoofing MAVLink packets, the considered specimen threat scenario exhibits an attacker's capacity to carry out a stealthy assault. ...

... Because Zipr's technology is agnostic to the source language, it was the only solution in the program that was able to handle the ADA Web Server application (obviously written in ADA.) in addition these projects, Zipr has been used to do antifragility work, and is the basis for effective binary-only fuzzing with tools like Zafl and the binary-only based version of Untracer, called HeXcite. (26)(27)(28)(29)(30) Hardened Registries in UVA ACCORD. As a practical example of how to use hardened registries in a real-world environment, we have partnered with the ACCORD team at the University of Virgnia (UVA). ...

... Heterogeneous TEEs. A line of research has been conducted to extend confidential computing to devices such as GPUs, including Graviton [45], HIX [27], HETEE [53], Strongbox [21], and Cronus [29]. In particular, GPU TEE is currently available on the commercial Nvidia H100 GPU [39]. ...

... Wider developments in TOD have been hampered by the two conflicting requirements: 1) largescale in-domain datasets are crucially required in order to unlock the potential of deep learning-based TOD components and systems to handle complex dialog patterns (Budzianowski et al., 2018;Lin et al., 2021b); at the same time 2) data collection for TOD is known to be notoriously difficult as it is extremely time-consuming, expensive, and requires expert and domain knowledge (Shah et al., 2018;Larson and Leach, 2022). Put simply, the creation of TOD datasets for new domains and languages incurs significantly higher time and budget costs than for most other NLP tasks (Casanueva et al., 2022). ...

... While there are some existing prompt-based approaches for DST with different designs of prompts such as using slot name [20,21,22,23], slot description [24], slot type [25], possible values [25], priming examples [26] and/or slot-specific question [4,27,28,29,8,30] in prompt sentences, they all fine-tune the entire LM along with the prompt tokens for a new domain, which requires a significant amount of training time, system resources, and annotated data [31,32]. The computing and data resource-hungry issues are more severe in the real-world deployment where LMs tuned for different domains and tasks need to be trained and hosted, and a typical dialogue system has to serve dozens of such LMs [33,34,35]. ...

... Thus, these results support the evidence that code and prose writing have different behaviour at mental level. Sharafi et al. (2021) performed two controlled experiments with 112 students during a series of development activities, i.e., code comprehension, code review, and data structure manipulations. During coding activities, students were monitored through neuroimaging activities, i.e., functional near-infrared spectroscopy (fNIRS), functional magnetic resonance imaging (fMRI) and eye tracking. ...

... Thus, we motivate to establish a new benchmark dataset that provides annotation of the fine-grained reasoning steps to automatically measure the reasoning consistency. However, collecting such a dataset is challenging due to high underlying human effort and may contain inconsistencies among annotators for the reasoning chains [11,22,51]. ...

... Advances in technology to accurately measure neural activity in the last three decades have revealed the existence of specialized regions for a variety of cognitive functions like language processing, face recognition, navigation etc. [39] The use of neuroimaging techniques to study the cognitive responses to programming has gained momentum recently. Prior works have investigated the neural processes involved in debugging [14], variable tracking when reading programs [36,46], semantic cues or program layout [18,51], program generation [41], manipulating data structures [34], biases in code review processes [33], and programming expertise [27,35,48]. ...

... determining the purpose behind a phrase or sentence). When crowdsourcing intent examples, workers typically create new phrases for some scenario (Wang et al., 2012;Larson et al., 2020). However, a data augmentation approach can also be used: by workers paraphrasing existing sentences with known intents. ...

... In last decade, neuroimaging techniques have been used to understand the cognitive processes of programming (Ebisch et al., 2013;Floyd et al., 2017;Huang et al., 2019;Peitek et al., 2018;Siegmund et al., 2014Siegmund et al., , 2017Krueger et al., 2020;Karas et al., 2021). Specifically, Peitek et al. (2018) and Siegmund et al. (2017) use these techniques in the program comprehension and Krueger et al. (2020) use these techniques in the code writing. ...