Saliency-map explanations are intended to make computer-vision models more transparent, but it is unclear whether they help people recognise biased behaviour. We conducted a controlled on-line study with 40 participants who compared Layer-wise Relevance Propagation maps from convolutional face-recognition models. A fair model was trained on a balanced synthetic dataset; two biased models were trained on data in which either light- or dark-skinned faces appeared only in frontal pose. Each participant completed 32 comparison trials. When the fair model was paired with the dark-skinned-pose-biased model, selections were near chance (52.8% favouring the fair model, binomial p = .36). When the fair model was paired with the light-skinned-pose-biased model, participants chose the biased model significantly more often (58.1%, p = .005). Confidence ratings varied with condition and did not systematically track model fairness. These results indicate that pixel-level attribution alone does not reliably expose training bias and can, in some settings, mislead non-expert users.

We develop and apply several novel methods quantifying dynamic multi-agent team interactions. These interactions are detected information-theoretically and captured in two ways: via (i) directed networks (interaction diagrams) representing significant coupled dynamics between pairs of agents, and (ii) state-space plots (coherence diagrams) showing coherent structures in Shannon information dynamics. This model-free analysis relates, on the one hand, the information transfer to responsiveness of the agents and the team, and, on the other hand, the information storage within the team to the team's rigidity and lack of tactical flexibility. The resultant interaction and coherence diagrams reveal implicit interactions, across teams, that may be spatially long-range. The analysis was verified with a statistically significant number of experiments (using simulated football games, produced during RoboCup 2D Simulation League matches), identifying the zones of the most intense competition, the extent and types of interactions, and the correlation between the strength of specific interactions and the results of the matches.

We study the order-chaos phase transition in random Boolean networks (RBNs), which have been used as models of gene regulatory networks. In particular we seek to characterize the phase diagram in information-theoretic terms, focusing on the effect of the control parameters (activity level and connectivity). Fisher information, which measures how much system dynamics can reveal about the control parameters, offers a natural interpretation of the phase diagram in RBNs. We report that this measure is maximized near the order-chaos phase transitions in RBNs, since this is the region where the system is most sensitive to its parameters. Furthermore, we use this study of RBNs to clarify the relationship between Shannon and Fisher information measures.

We investigate the problem of identifying anomalies in monitoring critical gas concentrations using a sensor network in an underground coal mine. In this domain, one of the main problems is a provision of mine specific anomaly detection, with cyclical (moving) instead of flatline (static) alarm threshold levels. An additional practical difficulty in modelling a specific mine is the lack of fully labelled data of normal and abnormal situations. We present an approach addressing these difficulties based on echo state networks learning mine specific anomalies when only normal data is available. Echo state networks utilize incremental updates driven by new sensor readings, thus enabling a detection of anomalies at any time during the sensor network operation. We evaluate this approach against a benchmark -- Bayesian network based anomaly detection, and observe that the quality of the overall predictions is comparable to the benchmark. However, the echo state networks maintain the same level of predictive accuracy for data from multiple sources. Therefore, the ability of echo state networks to model dynamical systems make this approach more suitable for anomaly detection and predictions in sensor networks.

In this paper, we combined the application of a non-linear dimensionality reduction technique, isomap, with expectation maximisation in graphical probabilistic models for learning and classification of hyperspectral image. Hyperspectral image spectroscopy gives much greater information content per pixel on the image than a normal colour image. This should greatly help with the autonomous identification of natural and man-made objects in unfamiliar terrains for robotic vehicles. However, the large information content of such data makes interpretation of hyperspectral images time-consuming and user-intensive. Isomap is used to find the underlying manifold of the training data. This low dimensional representation of the hyperspectral data facilitates the learning of a mixture of linear models representation similar to a mixture of factor analysers, the joint probability distributions of the model can be calculated offline. The learnt model is then applied to the hyperspectral image at run-time and data classification can be performed. We also show the comparison with results from standard techniques

In this paper, we use the Bayesian Structural EM algorithm as a classification method to learn and interpret hyperspectral sensor data in robotic planetary missions. Hyperspectral image spectroscopy is an emerging technique for geological investigations from airborne or orbital sensors. Many spacecraft carry spectroscopic equipment as wavelengths outside the visible light in the electromagnetic spectrum give much greater information about an object. The algorithm presented combines the standard Expectation Maximisation (EM), which optimises parameters, with structure search for model selection. We use the Bayesian Information Criterion (BIC) score to learn the network struc ture. The procedure only converges to a local maxima, thus requiring a good initial graph structure. Two initial structures are used: the Naïve Bayes, and the Tree-Augmented-Naïve Bayes structures. Our preliminary experiments show that the former results in a structure that can correctly determine the presence and types of minerals with merely 13% accuracy while the latter results in a structure that has approximately 94% accuracy.