Analogical reasoning is an increasingly popular, lightweight solution to enable large language model (LLM)-level reasoning without computational complexity. Still, it has yet to be adopted due to its reliance on strictly hand-formatted data. Therefore, we propose Analogy2KG (“Analogy to Knowledge Graph’’), as an automatic pipeline that transforms text into a KG format via a fine-tuned version of information extraction (IE) algorithms for long-text analogies. The need to verify that the complex underlying analogical structure of the data is maintained was done via paired samples tests in the creation and validation of this pipeline. Graph density was used to evaluate the structural quality of the resulting KGs. Lastly, causal relationships were optionally detected using a novel, question-and-answer-based method. Analogy2KG was validated on the Rattermann and Wharton long-text datasets, which suggested that the proposed methodology maintains analogical structure when transforming from text to KGs. The resulting RattermannKG and WhartonKG datasets were introduced to the literature, which is the first instance of a the conversion of long-text analogy dataset into a KG format in the literature. Finally, Analogy2KG had superior performance among three LLM-enabled information extraction algorithms: ChatIE, Code4UIE, and InstructUIE for maintaining analogical structure, despite operating without the need for an LLM backend and a pre-defined relation extractor list; thus, making it an ideal lightweight solution.

We propose a neuro-symbolic approach for learning causal complex event models from multi-source data, integrating causal discovery and temporal logic. Given resource constraints, we employ signal-level fusion by averaging the data from different antennas of the same WiFi receiver, followed by downsampling to reduce computational overhead. We consider a dataset of WiFi Channel State Information capturing human activities alongside video data from which we extract atomic symbolic activities such as “moving the upper arm.” The extracted symbolic information is processed through LPCMCI (Latent PCMCI). This causal discovery method extends PCMCI (Peter and Clark Momentary Conditional Independence) to handle latent dependencies across multiple time steps while mitigating false discoveries due to auto-correlations. The resulting causal structure is then translated into a temporal logic formula, which serves as a symbolic constraint in a neuro-symbolic learning pipeline. To efficiently process and learn from these structured constraints under resource limitations, we leverage Spiking Neural Networks, which offer energy-efficient computation while preserving temporal dynamics.

Spiking Neural Networks (SNN) are characterised by their unique temporal dynamics, but the properties and advantages of such computations are still not well understood. In order to provide answers, in this work we demonstrate how Spiking neurons can enable temporal feature extraction in feed-forward neural networks without the need for recurrent synapses, and how recurrent SNNs can achieve comparable results to LSTM with a smaller number of parameters. This shows how their bio-inspired computing principles can be successfully exploited beyond energy efficiency gains and evidences their differences with respect to conventional artificial neural networks. These results are obtained through a new task, DVS-Gesture-Chain (DVS-GC), which allows, for the first time, to evaluate the perception of temporal dependencies in a real event-based action recognition dataset. Our study proves how the widely used DVS Gesture benchmark can be solved by networks without temporal feature extraction when its events are accumulated in frames, unlike the new DVS-GC which demands an understanding of the order in which events happen. Furthermore, this setup allowed us to reveal the role of the leakage rate in spiking neurons for temporal processing tasks and demonstrated the benefits of ”hard reset” mechanisms. Additionally, we also show how time-dependent weights and normalisation can lead to understanding order by means of temporal attention. Code for the DVS-GC task is available. • Experiments show how voltage integration in SNN enables temporal feature extraction. • SNN extract temporal features without the need for recurrent connections. • SNN are compared to LSTM, demonstrating an accuracy vs parameter efficiency tradeoff • The DVS Gesture dataset can be solved by networks without temporal feature extraction. • An event-based action recognition task is proposed based on temporal order perception. • Voltage leak and the reset strategy in SNN determines adaptation speed to new inputs.

Abstract Mitigating catastrophic forgetting in continual learning is a long-standing challenge for artificial intelligence. Often, methods used to alleviate forgetting make use of either rehearsal buffers, pretrained backbones or task-id knowledge. However, these requirements result in severe limitations regarding scalability, privacy preservation, and efficient deployment. In this work, we explore how to eliminate the need for such requirements in incremental learning approaches based on parameter isolation. We propose Low Interference Feature Extraction Subnetworks (LIFES), a method that learns a subnetwork per task and uses all of them concurrently at inference time. This solution minimises requirements; however, it creates the need to address certain challenges. To formalize them, we break down the catastrophic forgetting problem into 4 distinct causes, and address them with a novel lateral classifiers regularization, weight standardization, and subnetwork interference connection pruning. Specifically, the use of lateral classification shows very promising results, forcing the model to learn distributions with higher inter-class distance. Using these components, LIFES achieves competitive results in standard task-agnostic scenarios, demonstrating the viability of this new perspective for parameter isolation, which has minimal requirements. Finally, we discuss how future work can improve this new paradigm further, and how the strategies defined can be complementary to other approaches.

Wi-Fi devices, akin to passive radars, can discern human activities within indoor settings due to the human body’s interaction with electromagnetic signals. Current Wi-Fi sensing applications predominantly employ data-driven learning techniques to associate the fluctuations in the physical properties of the communication channel with the human activity causing them. However, these techniques often lack the desired flexibility and transparency. This paper introduces DeepProbHAR, a neuro-symbolic architecture for Wi-Fi sensing, providing initial evidence that Wi-Fi signals can differentiate between simple movements, such as leg or arm movements, which are integral to human activities like running or walking. The neuro-symbolic approach affords gathering such evidence without needing additional specialised data collection or labelling. The training of DeepProbHAR is facilitated by declarative domain knowledge obtained from a camera feed and by fusing signals from various antennas of the Wi-Fi receivers. DeepProbHAR achieves results comparable to the state-of-the-art in human activity recognition. Moreover, as a by-product of the learning process, DeepProbHAR generates specialised classifiers for simple movements that match the accuracy of models trained on finely labelled datasets, which would be particularly costly.

Radio frequency (RF) fingerprinting extracts fingerprint features from RF signals to protect against masquerade attacks by enabling reliable authentication of communication devices at the “serial number” level. Facilitating the reliable authentication of communication devices are machine learning (ML) algorithms which find meaningful statistical differences between measured data. The Generalized Relevance Learning Vector Quantization-Improved (GRLVQI) classifier is one ML algorithm which has shown efficacy for RF fingerprinting device discrimination. GRLVQI extends the Learning Vector Quantization (LVQ) family of “winner take all” classifiers that develop prototype vectors (PVs) which represent data. In LVQ algorithms, distances are computed between exemplars and PVs, and PVs are iteratively moved to accurately represent the data. GRLVQI extends LVQ with a sigmoidal cost function, relevance learning, and PV update logic improvements. However, both LVQ and GRLVQI are limited due to a reliance on squared Euclidean distance measures and a seemingly complex algorithm structure if changes are made to the underlying distance measure. Herein, the authors (1) develop GRLVQI-D (distance), an extension of GRLVQI to consider alternative distance measures and (2) present the Cosine GRLVQI classifier using this framework. To evaluate this framework, the authors consider experimentally collected Z-wave RF signals and develop RF fingerprints to identify devices. Z-wave devices are low-cost, low-power communication technologies seen increasingly in critical infrastructure. Both classification and verification, claimed identity, and performance comparisons are made with the new Cosine GRLVQI algorithm. The results show more robust performance when using the Cosine GRLVQI algorithm when compared with four algorithms in the literature. Additionally, the methodology used to create Cosine GRLVQI is generalizable to alternative measures.

As research and development (R&D) in autonomous systems progresses further, more interdisciplinary knowledge is needed from domains as diverse as artificial intelligence (AI), bi-ology, psychology, modeling and simulation (M&S), and robotics. Such R&D efforts are necessarily interdisciplinary in nature and require technical as well as further soft skills of teamwork, communication and integration. In this paper, we introduce a 14 week, summer long internship for developing these skills in undergraduate science and engineering interns through R&D. The internship was designed to be modular and divided into three parts: training, innovation, and application/integration. The end result of the internship was 1) the development of an M&S ecosystem for autonomy concepts, 2) development and robotics testing of reasoning methods through both Bayesian methods and cognitive models of the basal ganglia, and 3) a process for future internships within the modular construct. Through collaboration with full-time professional staff, who actively learned with the interns, this internship incorporates a feedback loop to educate and per-form fundamental R&D. Future iterations of this internship can leverage the M&S ecosystem and adapt the modular internship framework to focus on different innovations, learning paradigms, and/or applications.

The proliferation of low-cost IEEE 802.15.4 ZigBee wireless devices in critical infrastructure applications presents security challenges. Network security commonly relies on bit-level credentials that are easily replicated and exploited by hackers. Unauthorized access can be mitigated by physical layer (PHY) security measures that exploit device-dependent emission characteristics that are sufficiently unique to discriminate devices. RF distinct native attribute (RF-DNA) fingerprinting is a PHY-based security measure, which computes statistical features extracted from such device emissions. However, the RF-DNA fingerprints can be numerous, correlated, and noisy, therefore, a dimensional reduction analysis (DRA) via feature selection is, therefore, of interest. Device classification with DRA feature subsets is evaluated using a multiple discriminant analysis (MDA) classifier. Determining feature relevance from MDA was generally dismissed in prior RF fingerprinting work and is seldom considered in other applications. Here, the MDA feature relevance is revisited using a proposed eigen-based MDA loadings fusion (MLF) methodology. The MDA classification models are adopted and used to assess device identification (ID) classification and verification performance for both the authorized and unauthorized (rogue) devices using a claimed versus actual biometric methodology. Performance is compared for six DRA methods using: 1) a two-sample Kolmogorov-Smirnov test; 2) one-way analysis of variance F-test statistics; 3) a Wilk's lambda ratio; 4) generalized relevance learning vector quantized-improved relevance; 5) randomly selected; and 6) the proposed MLF method. Quantitative and qualitative dimensionality assessment methods are compared and contrasted to establish upper bounds on the number of retained features. Experimentally collected ZigBee emissions are considered and ZigBee device classification and ID verification performance using DRA subsets are compared with a full-dimensional feature set. Results show that DRA via the proposed MLF method is superior and more robust than competing methods.

In this letter, a reliable, simple, and intuitive approach for hyperspectral imagery (HSI) anomaly detection (AD) is presented. This method, namely, the global iterative principal component analysis (PCA) reconstruction-error-based anomaly detector (GIPREBAD), examines AD by computing errors (residuals) associated with reconstructing the original image using PCA projections. PCA is a linear transformation and feature extraction process commonly used in HSI and frequently appears in operation prior to any AD task. PCA features represent a projection of the original data into lower-dimensional subspace. An iterative approach is used to mitigate outlier influence on background covariance estimates. GIPREBAD results are provided using receiver-operating-characteristic curves for HSI from the hyperspectral digital imagery collection experiment. Results are compared against the Reed-Xiaoli (RX) algorithm, the linear RX (LRX) algorithm, and the support vector data description (SVDD) algorithm. The results show that the proposed GIPREBAD method performs favorably compared with RX, LRX, and SVDD and is both intuitively and computationally simpler than either RX or SVDD.