Browsing by Author "Tang L"

Now showing 1 - 2 of 2
  • Thumbnail Image
    Item
    A Framework to Assess Multilingual Vulnerabilities of LLMs
    (Association for Computing Machinery, 2025-05-23) Tang L; Bogahawatta N; Ginige Y; Xu J; Sun S; Ranathunga S; Seneviratne S
    Large Language Models (LLMs) are acquiring a wider range of capabilities, including understanding and responding in multiple languages. While they undergo safety training to prevent them from answering illegal questions, imbalances in training data and human evaluation resources can make these models more susceptible to attacks in low-resource languages (LRL). This paper proposes a framework to automatically assess the multilingual vulnerabilities of commonly used LLMs. Using our framework, we evaluated six LLMs across eight languages representing varying levels of resource availability. We validated the assessments generated by our automated framework through human evaluation in two languages, demonstrating that the framework's results align with human judgments in most cases. Our findings reveal vulnerabilities in LRL; however, these may pose minimal risk as they often stem from the model's poor performance, resulting in incoherent responses.
  • Thumbnail Image
    Item
    A novel load cell-supported research platform to measure vertical and horizontal motion of a horse’s centre of mass during trailer transport
    (Elsevier, 4/02/2021) Colborne G; Tang L; Adams B; Gordon BI; McCabe BE; Riley CB
    During transport, horses are subjected to acceleration in three dimensions, rapid braking, turning, noise, and other stressors. The animal's ability to make postural corrections may be insufficient to prevent injury or distress, and so knowledge of the compensatory motion patterns of the horse in the trailer is a necessary precondition for smart design of transport systems. A custom two-horse trailer was built for this project. It had a horse compartment 1.85 m wide by 3.95 m long, with adjustable bulkheads and a centre divider separating the horses. The floor was instrumented with 24 shearbeam load cells to measure the vertical load imposed by each horse and its horizontal motion. Two horses were driven on a 56 km trip on both rural and urban roads. Load data were collected at 100 Hz for the 58-minute trip and were filtered with a cut-off frequency of 5 Hz using a Butterworth low-pass filter and then vertical acceleration computed. A pivot table counted sign reversals in the vertical acceleration signal, and vertical displacement was calculated using the fundamental frequency of the resulting acceleration data. Total vertical motion was calculated by making the negative displacements absolute and summing these with the positive displacements, and vertical work done was calculated by multiplying the force by the displacement measures. Horizontal motion was calculated by averaging the transverse and cranio-caudal position of the centre of pressure every second and adding the resultant displacements. Absolute vertical displacement of the two horses was 69.55 m and 97.56 m. In addition to the work done by standing, vertical work done in response to vibration was 322.4 kJ and 443.2 kJ. Horizontal excursion was 227.1 m and 243.0 m. This is a first effort to quantify the additional workload imposed on animals during transport, which will aid in the design of smart transport vehicles that will minimize the stress to horses.