Given the pressing need for assuring algorithmic transparency, Explainable AI (XAI) has emerged as one of the key areas of AI research. In this paper, we develop a novel Bayesian extension to the LIME framework, one of the most widely used approaches in XAI – which we call BayLIME. Compared to LIME, BayLIME exploits prior knowledge and Bayesian reasoning to improve both the consistency in repeated explanations of a single prediction and the robustness to kernel settings. BayLIME also exhibits better explanation fidelity than the state-of-the-art (LIME, SHAP and GradCAM) by its ability to integrate prior knowledge from, e.g., a variety of other XAI techniques, as well as verification and validation (V&V) methods. We demonstrate the desirable properties of BayLIME through both theoretical analysis and extensive experiments.

Recent probabilistic model checking techniques can verify reliability and performance properties of software systems affected by parametric uncertainty. This involves modelling the system behaviour using interval Markov chains, i.e., Markov models with transition probabilities or rates specified as intervals. These intervals can be updated continually using Bayesian estimators with imprecise priors, enabling the verification of the system properties of interest at runtime. However, Bayesian estimators are slow to react to sudden changes in the actual value of the estimated parameters, yielding inaccurate intervals and leading to poor verification results after such changes. To address this limitation, we introduce an efficient interval change-point detection method, and we integrate it with a state-of-the-art Bayesian estimator with imprecise priors. Our experimental results show that the resulting end-to-end Bayesian approach to change-point detection and estimation of interval Markov chain parameters handles effectively a wide range of sudden changes in parameter values, and supports runtime probabilistic model checking under parametric uncertainty.

Demonstrating high reliability and safety for safety-critical systems (SCSs) remains a hard problem. Diverse evidence needs to be combined in a rigorous way: in particular, results of operational testing with other evidence from design and verification. Growing use of machine learning in SCSs, by precluding most established methods for gaining assurance, makes evidence from operational testing even more important for supporting safety and reliability claims. We revisit the problem of using operational testing to demonstrate high reliability. We use Autonomous Vehicles (AVs) as a current example. AVs are making their debut on public roads: methods for assessing whether an AV is safe enough are urgently needed. We demonstrate how to answer 5 questions that would arise in assessing an AV type, starting with those proposed by a highly-cited study. We apply new theorems extending our Conservative Bayesian Inference (CBI) approach, which exploit the rigour of Bayesian methods while reducing the risk of involuntary misuse associated (we argue) with now-common applications of Bayesian inference; we define additional conditions needed for applying these methods to AVs. Prior knowledge can bring substantial advantages if the AV design allows strong expectations of safety before road testing. We also show how naive attempts at conservative assessment may lead to over-optimism instead; why extrapolating the trend of disengagements (take-overs by human drivers) is not suitable for safety claims; use of knowledge that an AV has moved to a “less stressful” environment. While some reliability targets will remain too high to be practically verifiable, our CBI approach removes a major source of doubt: it allows use of prior knowledge without inducing dangerously optimistic biases. For certain ranges of required reliability and prior beliefs, CBI thus supports feasible, sound arguments. Useful conservative claims can be derived from limited prior knowledge.

Robots are increasingly used to carry out critical missions in extreme environments that are hazardous for humans. This requires a high degree of operational autonomy under uncertain conditions, and poses new challenges for assuring the robot’s safety and reliability. In this paper, we develop a framework for probabilistic model checking on a layered Markov model to verify the safety and reliability requirements of such robots, both at pre-mission stage and during runtime. Two novel estimators based on conservative Bayesian inference and imprecise probability model with sets of priors are introduced to learn the unknown transition parameters from operational data. We demonstrate our approach using data from a real-world deployment of unmanned underwater vehicles in extreme environments.

There is an urgent societal need to assess whether autonomous vehicles (AVs) are safe enough. From published quantitative safety and reliability assessments of AVs, we know that, given the goal of predicting very low rates of accidents, road testing alone requires infeasible numbers of miles to be driven. However, previous analyses do not consider any knowledge prior to road testing - knowledge which could bring substantial advantages if the AV design allows strong expectations of safety before road testing. We present the advantages of a new variant of Conservative Bayesian Inference (CBI), which uses prior knowledge while avoiding optimistic biases. We then study the trend of disengagements (take-overs by human drivers) by applying Software Reliability Growth Models (SRGMs) to data from Waymo's public road testing over 51 months, in view of the practice of software updates during this testing. Our approach is to not trust any specific SRGM, but to assess forecast accuracy and then improve forecasts. We show that, coupled with accuracy assessment and recalibration techniques, SRGMs could be a valuable test planning aid.

In this paper the challenges of creating accurate, scalable and usable energy demand models are discussed, in the context of existing simulation and data driven energy demand models. Results from high resolution bottom-up data and simulation-based energy demand analysis from a community energy project are provided. A novel Hidden Markov Modelling and Generalised Pareto (HMM-GP) methodology for simulating synthetic electrical demand profiles is validated for residential buildings at a temporal resolution of five minutes. The corresponding dynamic thermal demands for the various building archetypes within the community are also modelled. This is achieved using automated externally driven IES-VE (building simulation) models for arrays of control profiles, and is also compared against in-situ thermal measurements.

In this paper we present a review into data driven prognostics and its relevance to resilience in energy systems. A data driven remaining useful life prediction for Li-ion batteries utilizing data analysis via a relevance vector machine (RVM) model is shown to be within 5% accuracy when applied to large lifecycle datasets. Results demonstrate that due to the agile nature of prognostic models and their accuracy, prognostics and health management methods will be vital to resilient and sustainable energy systems.

Power companies such as Southern California Edison (SCE) uses Demand Response (DR) contracts to incentivize consumers to reduce their power consumption during periods when demand forecast exceeds supply. Current mechanisms in use offer contracts to consumers independent of one another, do not take into consideration consumers' heterogeneity in consumption profile or reliability, and fail to achieve high participation. We introduce DR-VCG, a new DR mechanism that offers a flexible set of contracts (which may include the standard SCE contracts) and uses VCG pricing. We prove that DR-VCG elicits truthful bids, incentivizes honest preparation efforts, and enables efficient computation of allocation and prices. With simple fixed-penalty contracts, the optimization goal of the mechanism is an upper bound on probability that the reduction target is missed. Extensive simulations show that compared to the current mechanism deployed by SCE, the DR-VCG mechanism achieves higher participation, increased reliability, and significantly reduced total expenses.

This paper introduces a novel intention-aware routing system (IARS) for electric vehicles. This system enables vehicles to compute a routing policy that minimizes their expected journey time while considering the policies, or intentions, of other vehicles. Considering such intentions is critical for electric vehicles, which may need to recharge en route and face potentially significant queueing times if other vehicles choose the same charging stations. To address this, the computed routing policy takes into consideration predicted queueing times at the stations, which are derived from the current intentions of other electric vehicles. The efficacy of IARS is demonstrated through simulations using realistic settings based on real data from The Netherlands, including charging station locations, road networks, historical travel times, and journey origin-destination pairs. In these settings, IARS is compared with a number of state-of-the-art benchmark routing algorithms and achieves significantly lower average journey times. In some cases, IARS leads to an over 80% improvement in waiting times at charging stations and a more than 50% reduction in overall journey times.

Current electricity tariffs do not reflect the real cost that customers incur to suppliers, as units are charged at the same rate, regardless of how predictable each customer's consumption is. A recent proposal to address this problem are prediction-of-use tariffs. In such tariffs, a customer is asked in advance to predict her future consumption, and is charged based both on her actual consumption and the deviation from her prediction. Prior work {aamas2014} studied the cost game induced by a single such tariff, and showed customers would have an incentive to minimize their risk, by joining together when buying electricity as a grand coalition. In this work we study the efficient (i.e. cost-minimizing) structure of buying groups for the more realistic setting when multiple, competing prediction-of-use tariffs are available. We propose a polynomial time algorithm to compute efficient buyer groups, and validate our approach experimentally, using a large-scale data set of domestic electricity consumers in the UK.

In this paper we present a smart portfolio management methodology, which advances existing portfolio management techniques at two distinct levels. First, we develop a set of investment models that target regimes found in the data over different time horizons. We then build a meta-model which uses the Kelly criterion to determine an optimal allocation over these investment strategies, thus simultaneously capturing regimes operating in the data over different time horizons. Finally, in order to detect changes in the relevant data regime, and hence investment allocations, we use a forecasting algorithm which relies on a Kalman filter. We call our combined method, that uses both the Kelly criterion and the Kalman filter, the K2 algorithm. Using a large-scale historical dataset of both stocks and indices, we show that our K2 algorithm gives better risk adjusted returns in terms of the Sharpe ratio, better average gain to average loss ratio and higher probability of success compared to existing benchmarks, when measured in out-of-sample tests.

We develop an online mechanism for the allocation of an expiring resource to a dynamic agent population. Each agent has a non-increasing marginal valuation function for the resource, and an upper limit on the number of units that can be allocated in any period. We propose two versions on a truthful allocation mechanism. Each modifies the decisions of a greedy online assignment algorithm by sometimes cancelling an allocation of resources. One version makes this modification immediately upon an allocation decision while a second waits until the point at which an agent departs the market. Adopting a prior-free framework, we show that the second approach has better worst-case allocative efficiency and is more scalable. On the other hand, the first approach (with immediate cancellation) may be easier in practice because it does not need to reclaim units previously allocated. We consider an application to recharging plug-in hybrid electric vehicles (PHEVs). Using data from a real-world trial of PHEVs in the UK, we demonstrate higher system performance than a fixed price system, performance comparable with a standard, but non-truthful scheduling heuristic, and the ability to support 50% more vehicles at the same fuel cost than a simple randomized policy.

Virtual Power Plants (VPPs) are fast emerging as a suitable means of integrating small and distributed energy resources (DERs), like wind and solar, into the electricity supply network (Grid). VPPs are formed via the aggregation of a large number of such DERs, so that they exhibit the characteristics of a traditional generator in terms of predictability and robustness. In this work, we promote the formation of such "cooperative'' VPPs (CVPPs) using multi-agent technology. In particular, we design a payment mechanism that encourages DERs to join CVPPs with large overall production. Our method is based on strictly proper scoring rules and incentivises the provision of accurate predictions from the CVPPs---and in turn, the member DERs---which aids in the planning of the supply schedule at the Grid. We empirically evaluate our approach using the real-world setting of 16 commercial wind farms in the UK. We show that our mechanism incentivises real DERs to form CVPPs, and outperforms the current state of the art payment mechanism developed for this problem.

We propose a priced options model for solving the exposure problem of bidders with valuation synergies participating in a sequence of online auctions. We consider a setting in which complementary-valued items are offered sequentially by different sellers, who have the choice of either selling their item directly or through a priced option. In our model, the seller fixes the exercise price for this option, and then sells it through a first-price auction. We analyze this model from a decision-theoretic perspective and we show, for a setting where the competition is formed by local bidders (which desire a single item), that using options can increase the expected profit for both sides. Furthermore, we derive the equations that provide minimum and maximum bounds between which the bids of the synergy buyer are expected to fall, in order for both sides of the market to have an incentive to use the options mechanism. Next, we perform an experimental analysis of a market in which multiple synergy buyers are active simultaneously. We show that, despite the extra competition, some synergy buyers may benefit, because sellers are forced to set their exercise prices for options at levels which encourage participation of all buyers.

In multi-issue automated negotiation against unknown opponents, a key part of effective negotiation is the choice of concession strategy. In this paper, we develop a principled concession strategy, based on Gaussian processes predicting the opponent's future behaviour. We then use this to set the agent's concession rate dynamically during a single negotiation session. We analyse the performance of our strategy and show that it outperforms the state-of-the-art negotiating agents from the 2010 Automated Negotiating Agents Competition, in both a tournament setting and in self-play, across a variety of negotiation domains.

In this paper we introduce a new option pricing mechanism for reducing the exposure problem encountered by bidding agents with complementary valuations when participating in sequential, second-price auction markets. Existing option pricing models have two main drawbacks: they either apply fixed exercise prices, which may deter bidders with low valuations, thereby decreasing allocative efficiency, or options are offered for free, in which case bidders are less likely to exercise them, thereby reducing seller revenues. The proposed mechanism involving flexibly priced options addresses these problems by calculating the exercise price as well as the option price based on the bids received during an auction. For this new model, which extends and encompasses all the previous models examined, we derive the optimal strategies for a bidding agent with complementary preferences. Finally, we use these strategies to evaluate the proposed option mechanism through Monte-Carlo simutions, and compare it to existing mechanisms, both in terms of the seller revenue and the social welfare. We show that our new mechanism achieves higher market efficiency compared to having no options and free options, while achieving higher revenues for the seller than any existing option mechanism.

This article uses data from the social bookmarking site del.icio.us to empirically examine the dynamics of collaborative tagging systems and to study how coherent categorization schemes emerge from unsupervised tagging by individual users. First, we study the formation of stable distributions in tagging systems, seen as an implicit form of “consensus” reached by the users of the system around the tags that best describe a resource. We show that final tag frequencies for most resources converge to power law distributions and we propose an empirical method to examine the dynamics of the convergence process, based on the Kullback-Leibler divergence measure. The convergence analysis is performed for both the most utilized tags at the top of tag distributions and the so-called long tail. Second, we study the information structures that emerge from collaborative tagging, namely tag correlation (or folksonomy) graphs. We show how community-based network techniques can be used to extract simple tag vocabularies from the tag correlation graphs by partitioning them into subsets of related tags. Furthermore, we also show, for a specialized domain, that shared vocabularies produced by collaborative tagging are richer than the vocabularies which can be extracted from large-scale query logs provided by a major search engine. Although the empirical analysis presented in this article is based on a set of tagging data obtained from del.icio.us, the methods developed are general, and the conclusions should be applicable across other websites that employ tagging.

The debate within the Web community over the optimal means by which to organize information often pits formalized classifications against distributed collaborative tagging systems. A number of questions remain unanswered, however, regarding the nature of collaborative tagging systems including whether coherent categorization schemes can emerge from unsupervised tagging by users. This paper uses data from the social bookmarking site delicio. us to examine the dynamics of collaborative tagging systems. In particular, we examine whether the distribution of the frequency of use of tags for "popular" sites with a long history (many tags and many users) can be described by a power law distribution, often characteristic of what are considered complex systems. We produce a generative model of collaborative tagging in order to understand the basic dynamics behind tagging, including how a power law distribution of tags could arise. We empirically examine the tagging history of sites in order to determine how this distribution arises over time and to determine the patterns prior to a stable distribution. Lastly, by focusing on the high-frequency tags of a site where the distribution of tags is a stabilized power law, we show how tag co-occurrence networks for a sample domain of tags can be used to analyze the meaning of particular tags given their relationship to other tags.

A component-based generic agent architecture for multi-attribute (integrative) negotiation is introduced and its application is described in a prototype system for negotiation about cars, developed in cooperation with, among others, Dutch Telecom KPN. The approach can be characterized as cooperative one-to-one multi-criteria negotiation in which the privacy of both parties is protected as much as desired. We model a mechanism in which agents are able to use any amount of incomplete preference information revealed by the negotiation partner in order to improve the efficiency of the reached agreements. Moreover, we show that the outcome of such a negotiation can be further improved by incorporating a “guessing” heuristic, by which an agent uses the history of the opponent’s bids to predict his preferences. Experimental evaluation shows that the combination of these two strategies leads to agreement points close to or on the Pareto-efficient frontier. The main original contribution of this paper is that it shows that it is possible for parties in a cooperative negotiation to reveal only a limited amount of preference information to each other, but still obtain significant joint gains in the outcome.