A natural way to characterize the cluster structure of a dataset is by finding regions containing a high density of data. This can be done in a nonparametric way with a kernel density estimate, whose modes and hence clusters can be found using mean-shift algorithms. We describe the theory and practice behind clustering based on kernel density estimates and mean-shift algorithms. We discuss the blurring and non-blurring versions of mean-shift; theoretical results about mean-shift algorithms and Gaussian mixtures; relations with scale-space theory, spectral clustering and other algorithms; extensions to tracking, to manifold and graph data, and to manifold denoising; K-modes and Laplacian K-modes algorithms; acceleration strategies for large datasets; and applications to image segmentation, manifold denoising and multivalued regression.

Neural network models of early sensory processing typically reduce the dimensionality of streaming input data. Such networks learn the principal subspace, in the sense of principal component analysis (PCA), by adjusting synaptic weights according to activity-dependent learning rules. When derived from a principled cost function these rules are nonlocal and hence biologically implausible. At the same time, biologically plausible local rules have been postulated rather than derived from a principled cost function. Here, to bridge this gap, we derive a biologically plausible network for subspace learning on streaming data by minimizing a principled cost function. In a departure from previous work, where cost was quantified by the representation, or reconstruction, error, we adopt a multidimensional scaling (MDS) cost function for streaming data. The resulting algorithm relies only on biologically plausible Hebbian and anti-Hebbian local learning rules. In a stochastic setting, synaptic weights converge to a stationary state which projects the input data onto the principal subspace. If the data are generated by a nonstationary distribution, the network can track the principal subspace. Thus, our result makes a step towards an algorithmic theory of neural computation.

Saving energy is a major concern. Hence, it is fundamental to design and construct buildings that are energy-efficient. It is known that the early stage of architectural design has a significant impact on this matter. However, it is complex to create designs that are optimally energy efficient, and at the same time balance other essential criterias such as economics, space, and safety. One state-of-the art approach is to create parametric designs, and use a genetic algorithm to optimize across different objectives. We further improve this method, by aggregating the solutions of multiple agents. We evaluate diverse teams, composed by different agents; and uniform teams, composed by multiple copies of a single agent. We test our approach across three design cases of increasing complexity, and show that the diverse team provides a significantly larger percentage of optimal solutions than single agents.

Vehicle-sharing (ex: bike sharing, car sharing) is widelyadopted in many cities of the world due to concernsassociated with extensive private vehicle usage, whichhas led to increased carbon emissions, traffic conges-tion and usage of non-renewable resources. In vehicle-sharing systems, base stations are strategically placedthroughout a city and each of the base stations containa pre-determined number of vehicles at the beginningof each day. Due to the stochastic and individualisticmovement of customers, typically, there is either con-gestion (more than required) or starvation (fewer thanrequired) of vehicles at certain base stations. As demon-strated in our experimental results, this happens oftenand can cause a significant loss in demand. We proposeto dynamically redeploy idle vehicles using carriers soas to minimize lost demand or alternatively maximizerevenue of the vehicle sharing company. To that end,we contribute an optimization formulation to jointly ad-dress the redeployment (of vehicles) and routing (of car-riers) problems and provide two approaches that rely ondecomposability and abstraction of problem domains toreduce the computation time significantly. Finally, wedemonstrate the utility of our approaches on two realworld data sets of bike-sharing companies.

The timely detection of leaks in water distribution systems is critical to the sustainable provision of clean water to consumers. Increasingly, water companies are deploying remote sensors to measure water flow in real-time in order to detect such leaks. However, in practice, for typical District Metering Zones (DMZ), financial constraints limit the number of deployable real time flow sensors/meters to one or two, thus constraining leak detection to be based on the aggregated flow being monitored at these point. Such aggregated flow data typically exhibits input signal dependence whereby both noise and leaks are dependent on the flow being measured. This limited monitoring and input signal dependance make conventional approaches based on simple thresholds unreliable for real time leak detection. To address this, we propose a Gaussian process (GP) model with an additive diagonal noise covariance that is able to handle the input dependant noise observed in this setting. A parameterised mean step change function is used to detect leaks and to estimate their size. Using prior water distribution systems (WDS) knowledge we dynamically bound and discretize the detection parameters of the step change mean function, reducing and pruning the parameter search space considerably. We evaluate the proposed noise scaled GP (NSGP) against both the latest researchwork on GP based fault detection methods and the current state of the art and applied leak detection approaches in water distribution systems. We show that our proposed method outperforms other approaches, on real water network data with synthetically generatedvtime varying leaks, with a detection accuracy of 99%, almost zero false positive detections and the lowest root mean squared error in leak magnitude estimation (0.065 l/s).

Skills are generally designed by a domain expert, but designing a `good' set of skills can be challenging in high-dimensional, complex domains. In some cases, the skills may contain useful prior knowledge but cannot solve the task, resulting in a sub-optimal solution or no solution at all. Given a `poor' set of skills, we would like to dynamically improve them. The main contribution of this paper is showing that Interrupting Options (IO) can improve the initial skill set in a high-dimensional, complex domain by learning when to switch between skills. Furthermore, we discuss some of the pitfalls we ran into while trying to get IO to work.

Western societies are facing demographic challenges due the rapid aging of their population. In this context, economic and social issues are emerging, such as an increasing number of elderly in need of home cares and a shortage of caregivers. Smart home technology has imposed itself as a potential avenue of solution to these important issues. Its goal is to provide adapted assistance to a semi-autonomous resident in the form of hints, suggestions, reminders, and to take preventive actions, for instance turning off the oven, in the case of an emergency. The main scientific challenge related to this kind of assistance concerns the problem of recognizing, in real time, of the on-going activities of the resident in order to provide punctual guidance for the completion of everyday tasks. In the literature, the majority of the proposed solutions for activity recognition exploit a complex and expensive network of intrusive sensors (i.e. infrared, radio-identification, electromagnetic, pressure, cameras, etc.). A recent and innovative way of performing activity recognition is based on the monitoring of electrical household appliances by analyzing the electrical signals solely at the main panel. This approach is less intrusive and required only one sensor. In this paper, we present new advancements in that field, which take the form of an efficient method for recognizing electrical appliances within smart home based on the analysis of the features of the load signatures (active and reactive power, FFT) and on the use of the C4.5 algorithm to extract decision trees. This method has been implemented and tested in real smart home infrastructure showing that it is economical, simple and efficient.

We are concerned with controlling dynamical systems, such as self-driving cars and smart buildings, in a manner that guarantees that they satisfy complex task specifications. Mixed integer linear programming has recently proven to be a powerful tool for such problems, enabling the computation of optimal plans that satisfy complex temporal constraints for high-dimensional, dynamical systems. These optimization-based approaches find solutions quickly for challenging (and previously unsolvable) planning problems. Framing temporal logic planning as constrained optimization also presents exciting new areas of research.

In this paper we present a novel online transfer learning approach to determine the set of tests to perform, and the sequence in which they need to be performed, in order to develop an accurate diagnosis while minimizing the cost of performing the tests. Our learning approach can be incorporated as part of a clinical decision support system (CDSS) with which clinicians can interact. The approach builds on a contextual bandit framework and uses online transfer learning to overcome limitations with the availability of rich training data sets that capture different conditions, context, test results as well as outcomes. We provide confidence bounds for our recommended policies, which is essential in order to build the trust of clinicians. We evaluate the algorithm against different transfer learning approaches on real-world patient alarm datasets collected from Neurological Intensive Care Units (with reduced costs by 20%).

We built a conversational agent performing social promotion (CASP) to assist in automation of interacting with Facebook friends. CASP relies on a domain-independent natural language relevance technique which filters web mining results to support a conversation with friends and other network members. In this study we focus on recognizing friends’ intents to better support automated conversation with them. We learn the plausible sequences of communicative actions and mental states as they are expressed in text to support plausible dialogue. We evaluate the relevance of the constructed conversations with respect to suitability of topicality and communicative actions, measuring how human users loose trust in the system. It is confirmed that maintaining a plausible sequences of communicative actions in automated postings is important for retaining trust of human peers and efficient social promotion by means of CASP.