LLM Multimodal Traffic Accident Forecasting.

With the rise in traffic congestion in urban centers, predicting accidents has become paramount for city planning and public safety. This work comprehensively studied the efficacy of modern deep learning (DL) methods in forecasting traffic accidents and enhancing Level-4 and Level-5 (L-4 and L-5) driving assistants with actionable visual and language cues. Using a rich dataset detailing accident occurrences, we juxtaposed the Transformer model against traditional time series models like ARIMA and the more recent Prophet model. Additionally, through detailed analysis, we delved deep into feature importance using principal component analysis (PCA) loadings, uncovering key factors contributing to accidents. We introduce the idea of using real-time interventions with large language models (LLMs) in autonomous driving with the use of lightweight compact LLMs like LLaMA-2 and Zephyr-7b-α. Our exploration extends to the realm of multimodality, through the use of Large Language-and-Vision Assistant (LLaVA)-a bridge between visual and linguistic cues by means of a Visual Language Model (VLM)-in conjunction with deep probabilistic reasoning, enhancing the real-time responsiveness of autonomous driving systems. In this study, we elucidate the advantages of employing large multimodal models within DL and deep probabilistic programming for enhancing the performance and usability of time series forecasting and feature weight importance, particularly in a self-driving scenario. This work paves the way for safer, smarter cities, underpinned by data-driven decision making.

LLM Adaptive PID Control for B5G Truck Platooning Systems.

This paper presents an exploration into the capabilities of an adaptive PID controller within the realm of truck platooning operations, situating the inquiry within the context of Cognitive Radio and AI-enhanced 5G and Beyond 5G (B5G) networks. We developed a Deep Learning (DL) model that emulates an adaptive PID controller, taking into account the implications of factors such as communication latency, packet loss, and communication range, alongside considerations of reliability, robustness, and security. Furthermore, we harnessed a Large Language Model (LLM), GPT-3.5-turbo, to deliver instantaneous performance updates to the PID system, thereby elucidating its potential for incorporation into AI-enabled radio and networks. This research unveils crucial insights for augmenting the performance and safety parameters of vehicle platooning systems within B5G networks, concurrently underlining the prospective applications of LLMs within such technologically advanced communication environments.

Simulation of germanium detector calibration using the Monte Carlo method: comparison between point and surface source models.

Simulation of detector calibration using the Monte Carlo method is very convenient. The computational calibration procedure using the MCNP code was validated by comparing results of the simulation with laboratory measurements. The standard source used for this validation was a disc-shaped filter where fission and activation products were deposited. Some discrepancies between the MCNP results and laboratory measurements were attributed to the point source model adopted. In this paper, the standard source has been simulated using both point and surface source models. Results from both models are compared with each other as well as with experimental measurements. Two variables, namely, the collimator diameter and detector-source distance have been considered in the comparison analysis. The disc model is seen to be a better model as expected. However, the point source model is good for large collimator diameter and also when the distance from detector to source increases, although for smaller sizes of the collimator and lower distances a surface source model is necessary.

Developing a virtual reality application for training nuclear power plant operators: setting up a database containing dose rates in the refuelling plant.

Operators in Nuclear Power Plants can receive high doses during refuelling operations. A training programme for simulating refuelling operations will be useful in reducing the doses received by workers as well as minimising operation time. With this goal in mind, a virtual reality application is developed within the framework of the CIPRES project. The application requires doses, both instantaneous and accumulated, to be displayed at all times during operator training. Therefore, it is necessary to set up a database containing dose rates at every point in the refuelling plant. This database is based on radiological protection surveillance data measured in the plant during refuelling operations. Some interpolation routines have been used to estimate doses through the refuelling plant. Different assumptions have been adopted in order to perform the interpolation and obtain consistent data. In this paper, the procedures developed to set up the dose database for the virtual reality application are presented and analysed.