Search

This paper contributes with a pragmatic evaluation framework for explainable Machine Learning (ML) models for clinical decision support. The study revealed a more nuanced role for ML explanation models, when these are pragmatically embedded in the clinical context. Despite the general positive attitude of healthcare professionals (HCPs) towards explanations as a safety and trust mechanism, for a significant set of participants there were negative effects associated with confirmation bias, accentuating model over-reliance and increased effort to interact with the model. Also, contradicting one of its main intended functions, standard explanatory models showed limited ability to support a critical understanding of the limitations of the model. However, we found new significant positive effects which repositions the role of explanations within a clinical context: these include reduction of automation bias, addressing ambiguous clinical cases (cases where HCPs were not certain about their decision) and support of less experienced HCPs in the acquisition of new domain knowledge., Comment: supplementary information in the main pdf

Background: Pre-marketing authorisation estimates of survival are generally restricted to those observed directly in randomised controlled trials (RCTs). However, for regulatory and Health Technology Assessment (HTA) decision-making a longer time horizon is often required than is studied in RCTs. Therefore, extrapolation is required to estimate long-term treatment effect. Registry data can provide evidence to support extrapolation of treatment effects from RCTs, which are considered the main sources of evidence of effect for new drug applications. A number of methods are available to extrapolate survival data, such as Exponential, Weibull, Gompertz, log-logistic or log-normal parametric models. The different methods have varying functional forms and can result in different survival estimates. Methods: The aim of this paper was to use registry data to supplement the relatively short term RCT data to obtain long term estimates of effect. No formal hypotheses were tested. We explore the above parametric regression models as well as a nonparametric regression model based on local linear (parametric) regression. We also explore a Bayesian model constrained to the long term estimate of survival reported in literature, a Bayesian power prior approach on the variability observed from published literature, and a Bayesian Model Averaging (BMA) approach. The methods were applied to extrapolate overall survival of a RCT in metastatic melanoma. Results: The results showed that the BMA approach was able to fit the RCT data well, with the lowest variability of the area under the curve up to 72 months with or without the SEER Medicare registry. Conclusion: the BMA approach is a viable approach to extrapolate overall survival in the absence of long term data., Comment: 21 Pages, 7 named figures, 3 tables

POMDPs are standard models for probabilistic planning problems, where an agent interacts with an uncertain environment. We study the problem of almost-sure reachability, where given a set of target states, the question is to decide whether there is a policy to ensure that the target set is reached with probability 1 (almost-surely). While in general the problem is EXPTIME-complete, in many practical cases policies with a small amount of memory suffice. Moreover, the existing solution to the problem is explicit, which first requires to construct explicitly an exponential reduction to a belief-support MDP. In this work, we first study the existence of observation-stationary strategies, which is NP-complete, and then small-memory strategies. We present a symbolic algorithm by an efficient encoding to SAT and using a SAT solver for the problem. We report experimental results demonstrating the scalability of our symbolic (SAT-based) approach., Comment: Full version of "A Symbolic SAT-based Algorithm for Almost-sure Reachability with Small Strategies in POMDPs" AAAI 2016

Successive elimination of candidates is often a route to making manipulation intractable to compute. We prove that eliminating candidates does not necessarily increase the computational complexity of manipulation. However, for many voting rules used in practice, the computational complexity increases. For example, it is already known that it is NP-hard to compute how a single voter can manipulate the result of single transferable voting (the elimination version of plurality voting). We show here that it is NP-hard to compute how a single voter can manipulate the result of the elimination version of veto voting, of the closely related Coombs' rule, and of the elimination versions of a general class of scoring rules., Comment: To appear in Proceedings of Twenty-Sixth Conference on Artificial Intelligence (AAAI-12)

We prove that it is NP-hard for a coalition of two manipulators to compute how to manipulate the Borda voting rule. This resolves one of the last open problems in the computational complexity of manipulating common voting rules. Because of this NP-hardness, we treat computing a manipulation as an approximation problem where we try to minimize the number of manipulators. Based on ideas from bin packing and multiprocessor scheduling, we propose two new approximation methods to compute manipulations of the Borda rule. Experiments show that these methods significantly outperform the previous best known %existing approximation method. We are able to find optimal manipulations in almost all the randomly generated elections tested. Our results suggest that, whilst computing a manipulation of the Borda rule by a coalition is NP-hard, computational complexity may provide only a weak barrier against manipulation in practice.

We study the problem of coalitional manipulation in elections using the unweighted Borda rule. We provide empirical evidence of the manipulability of Borda elections in the form of two new greedy manipulation algorithms based on intuitions from the bin-packing and multiprocessor scheduling domains. Although we have not been able to show that these algorithms beat existing methods in the worst-case, our empirical evaluation shows that they significantly outperform the existing method and are able to find optimal manipulations in the vast majority of the randomly generated elections that we tested. These empirical results provide further evidence that the Borda rule provides little defense against coalitional manipulation., Comment: To appear in Proceedings of the Third International Workshop on Computational Social Choice