Evaluating uncertainty in large language model (LLM) reasoning is challenging due to their vast parameter space, abstract knowledge representation, and limited transparency regarding training data. While normative formalisms, such as deductive logic, clearly define sound reasoning in the absence of uncertainty, reasoning under uncertainty admits multiple approaches, including probabilistic reasoning (e.g. Bayesian), belief function reasoning (e.g. Dempster-Shafer), or fuzzy logic, to name a few. This paper examines how LLMs handle uncertainty by analyzing outcomes based on an atomic fusion and reasoning problem. We establish a point of reference using the simplest of fusion topologies to facilitate transparency and understanding of how LLMs align with established theories. The reasoning approaches of different LLMs with varying complexities are compared to established normative frameworks, providing insights into which formalism best aligns with LLM reasoning and assessing its soundness and consistency. A deviation function for assessment is developed, and the results indicate that the tested LLMs' reasoning under uncertainty does not consistently align with established theories, even for the simplest information fusion topologies. These preliminary results form the basis for further investigations and LLM refinements.

The ability to offer long-range, high scalability, sustainability, and low-power wireless communication, are the key factors driving the rapid adoption of the LoRaWAN technology in large-scale Internet of Things applications. This situation has created high demand to incorporate location estimation capabilities into large-scale IoT applications to meaningfully interpret physical measurements collected from IoT devices. As a result, research aimed at investigating node localization in LoRaWAN networks is on the rise. The poor localization performance of classical range-based localization approaches in LoRaWAN networks is due to the long-range nature of LoRaWAN and the rich scattering nature of outdoor environments, which affects signal transmission. Because of the ability of fingerprint-based localization methods to effectively learn useful positional information even from noisy RSSI data, this work proposes a fingerprinting-based branched convolutional neural network (CNN) localization method enhanced with squeeze and excitation (SE) blocks to localize a node in LoRaWAN using RSSI data. Results from the experiments conducted to evaluate the performance of the proposed method using a publicly available LoRaWAN dataset prove its effectiveness and robustness in localizing a node with satisfactory results even with a 30% reduction in both the principal component analysis (PCA) variances on the training data and the size of the original sample. A localization accuracy of 284.57 m mean error on the test area was achieved using the Powed data representation, which represents an 8.39% increase in localization accuracy compared to the currently best-performing fingerprint method in the literature, evaluated using the same LoRaWAN dataset.

In recent years, low-power wide area networks (LPWANs), particularly Long-Range Wide Area Network (LoRaWAN) technology, are increasingly being adopted into large-scale Internet of Things (IoT) applications thanks to having the ability to offer cost-effective long-range wireless communication at low-power. The need to provide location-stamped communications to IoT applications for meaningful interpretation of physical measurements from IoT devices has increased demand to incorporate location estimation capabilities into LoRaWAN networks. Fingerprint-based localization methods are increasingly becoming popular in LoRaWAN networks because of their relatively high accuracy compared to range-based localization methods. This work proposes hybrid convolutional neural networks (CNNs)-transformer fingerprinting method to localize a node in a LoRaWAN network. CNNs are adopted to complement the strengths of the Transformer by adding the ability to capture local features from input data and consequently allow the Transformer, through the attention mechanism, to effectively learn global dependencies from the input data. Specifically, the proposed method works by first learning the local location features from the input data using the CNNs and passing the resulting information to the transformer encoder to learn global features from the input data. The output of the transformer encoder is then concatenated with information learned at the local level and then passed through the regressor for the final location estimation. With a localization performance of 290.71 m mean error achieved, the proposed method outperformed similar state-of-the-art works in the literature evaluated on the same publicly available LoRaWAN dataset.

This study explores the feasibility of constructing a small-scale speech recognition system capable of competing with larger, modern automated speech recognition (ASR) systems in both performance and word error rate (WER). Our central hypothesis posits that a compact transformer-based ASR model can yield comparable results, specifically in terms of WER, to traditional ASR models while challenging contemporary ASR systems that boast significantly larger computational sizes. The aim is to extend ASR capabilities to under-resourced languages with limited corpora, catering to scenarios where practitioners face constraints in both data availability and computational resources. The model, comprising a compact convolutional neural network (CNN) and transformer architecture with 2.214 million parameters, challenges the conventional wisdom that large-scale transformer-based ASR systems are essential for achieving high accuracy. In comparison, contemporary ASR systems often deploy over 300 million parameters. Trained on a modest dataset of approximately 3000 h—significantly less than the 50,000 h used in larger systems—the proposed model leverages the Common Voice and LibriSpeech datasets. Evaluation on the LibriSpeech test-clean and test-other datasets produced character error rates (CERs) of 6.40% and 16.73% and WERs of 16.03% and 35.51% respectively. Comparisons with existing architectures showcase the efficiency of our model. A gated recurrent unit (GRU) architecture, albeit achieving lower error rates, incurred a computational cost 24 times larger than our proposed model. Large-scale transformer architectures, while achieving marginally lower WERs (2%–4% on LibriSpeech test-clean), require 200 times more parameters and 53,000 additional hours of training data. Modern large language models are used to improve the WERs, but require large computational resources. To further enhance performance, a small 4-g language model was integrated into our end-to-end ASR model, resulting in improved WERs. The overarching goal of this work is to provide a practical solution for practitioners dealing with limited datasets and computational resources, particularly in the context of under-resourced languages.

ChatGPT, OpenAI’s chatbot, has gained consider-able attention since its launch in November 2022, owing to its ability to formulate articulated responses to text queries and comments relating to seemingly any conceivable subject. As impressive as the majority of interactions with ChatGPT are, this large language model has a number of acknowledged shortcomings, which in several cases, may be directly related to how ChatGPT handles uncertainty. The objective of this paper is to pave the way to formal analysis of ChatGPT uncertainty handling. To this end, the ability of the Uncertainty Representation and Reasoning Framework (URREF) ontology is assessed, to support such analysis. Elements of structured experiments for reproducible results are identified. The dataset built varies Information Criteria of Correctness, Non-specificity, Self-confidence, Relevance and Inconsistency, and the Source Criteria of Reliability, Competency and Type. ChatGPT’s answers are analyzed along Information Criteria of Correctness, Non-specificity and Self-confidence. Both generic and singular information are sequentially provided. The outcome of this preliminary study is twofold: Firstly, we validate that the experimental setup is efficient in capturing aspects of ChatGPT uncertainty handling. Secondly, we identify possible modifications to the URREF ontology that will be discussed and eventually implemented in URREF ontology Version 4.0 under development.

Video data is increasingly being provided by multiple sources. These sources are used to obtain reliable feature information within the context of training data shortages. As such it becomes crucial to interpret and refine such information. Recent research on video content analysis often use deep-learning features due to their outstanding performance in different domains. These features are aggregated over time to create a video-level descriptor. In this research, we explore the potential of statistical feature aggregation methods in combining deep-learning features that represent the visual content of videos. In particular, the contributions of this paper are two-fold: Firstly, we compared statistical feature aggregation methods by selecting movie sequels, calculating their in-sequel-mean-distance and out-of-sequel-mean-distance as well as the Bhattacharyya distance between them and using Principal Component Analysis (PCA) and T-distributed Stochastic Neighbour Embedding (t-SNE) to visualise their distribution in the feature space. This is important to easily comprehend the video content without the need to watch them and understand how well the statistical feature aggregation methods represent the videos. Secondly, the performance of these aggregation methods is explored in the context of a content-based video retrieval task which is evaluated in terms of relevance. The observed results show that the statistical feature aggregation method based on variance outperforms the methods based on maximum, mean, median, median absolute deviation and interquartile range. This outcome is supported when interpreting the t-SNE visualisation method as well as Bhattacharyya distance calculations. Overall statistical feature aggregation methods which measure the spread of a distribution achieve better performance compared to methods that are a measure of location.

In this paper, data from a simulated data set were fused with a significantly smaller measured data set for the purpose of training a deep neural network to identify animals from the footage of a dash cam. The trained networks were used to detect wildlife in an environment similar to game reserves in South Africa. To enable the automatic collection of data for the experiment, a simulated environment was created to simulate four classes of wildlife found in South Africa: buffalo, elephants, rhino and zebra. The network structure for the detector network selected was an adapted version of the tiny YOLOv3 network. It was discovered that using transfer learning and fine tuning resulted in two models with higher accuracy of 82.59% and 86.64% [email protected] respectively than models where no transfer learning was used. The results were achieved when tested on a testing set of digital photographic images. These networks, initialised using transfer learning, were also faster and easier to train than training using a combined data set of photographic and simulated images from scratch. The simulated environment can however not replace real-life data, as was proven by an accuracy of no better than chance for the model trained using only simulated data.

Extended object tracking has become an integral part of various autonomous systems in diverse fields. Although it has been extensively studied over the past decade, many complex challenges remain in the context of extended object tracking. In this paper, a new method for tracking multiple irregularly shaped extended objects using surface measurements is proposed. The Gaussian Process Convolution Particle Filter proposed in [1], designed to track a single extended/group object, is enhanced for tracking multiple extended objects. A convolution kernel is proposed to estimate the multi-object likelihood. A target birth/death model based on the proposed method is also introduced for automatic initiation and deletion of the objects. The proposed approach is validated on real-world LiDAR data which shows that the method is efficient in tracking multiple irregularly shaped extended objects in challenging scenarios involving occlusion, dense clutter and low object detection.

Sensormanagement is an important function of any data fusion center as the output of a fusion system is dependent on the quality of the information collected. In this paper, the scheduling aspect of sensor management function is implemented using Response Surface Modeling (RSM). Applying RSM requires formulating the sensor management function as an objective function. The benefit of RSM over prior global optimization approaches is the simplification of the evaluation of this objective function to find global optima. This leads to either reduced computational requirements and/ or shorter due times for creating sensor schedules. This work shows the utility of RSM towards scheduling multiple sensors, and seeks to introduce RSM to the sensor management community. It is shown that the RSM scheduler provides a significant improvement towards reducing the number of missed targets in a surveillance radar network. This is compared to performing a uniform scanning regime (or sequential stepped scan) often employed. Very few iterations are required to provide this gain. The RSM technique also quickly determines where the most effective use of sensor resources needs to be applied. Consequently, it spends more radar dwell time on these beam locations.

Advances in sensor systems have resulted in the availability of high resolution sensors, capable of generating massive amounts of data. For complex systems to run online, the primary focus is on computationally efficient filters for the estimation of latent states related to the data. In this paper a novel method for efficient state estimation with the unscented Kalman Filter is proposed. The focus is on applications consisting of a massive amount of data. From a modelling perspective, this amounts to a measurement vector with dimensionality significantly greater than the dimensionality of the state vector. The efficiency of the filter is derived from a parallel filter structure which is enabled by the expectation propagation algorithm. A novel parallel measurement processing expectation propagation unscented Kalman filter is developed. The primary advantage of the novel algorithm is in the ability to achieve computational improvements with negligible loses in filter accuracy. An example of robot localization with a high resolution laser rangefinder sensor is presented. A 47.53% decrease in computational time was exhibited for a scenario with a processing platform consisting of 4 processors, with a negligible loss in accuracy.