We present a technique to leverage annotator rationale an- notations for ventilator assisted pneumonia (VAP) classifi- cation. Given an annotated training corpus of 1344 narrative chest X-ray reports, we report results for two supervised classification tasks: Critical Pulmonary Infection Score (CPIS) and the likelihood of Pneumonia (PNA). For both tasks, our training data contain annotator rationale snippets (i.e., spans of text that are relevant to annotator decisions). Because we assume that the snippet is not marked in the test data, we first built a sequential labeler to detect the location of snippets. The detected snippets are then used by the CPIS and PNA classifiers. Our experiments demonstrate that having access to detected annotator rationale leads to an incremental improvement in classification accuracy from 0.858 to 0.871 for CPIS, and from 0.785 to 0.821 for PNA.

There is a pressing need in clinical practice to mitigate (identify and address) adverse interactions that occur when a comorbid patient is managed according to multiple concurrently applied disease-specific clinical practice guidelines (CPGs). We describe an automatic algorithm for mitigating undesirable interactions for pairs of CPGs. The algorithm constructs logical models of processed CPGs and employs constraint logic programming to solve them. It handles two important issues frequently occurring in CPGs - iterative actions forming a cycle and numerical measurements. Dealing with these two issues in practice relies on a physician's knowledge and the manual analysis of CPGs. Yet for guidelines to be considered stand-alone and an easy to use clinical decision support tool this process needs to be automated. In this paper we present our algorithm that aims to build such a tool by mitigating multiple CPGs while handling cycles and numerical measurements. The application of the mitigation algorithm is illustrated with a clinical case study involving a comorbid patient suffering from atrial fibrillation in the setting of Wolff-Parkinsons-White syndrome.

Practitioners and researchers in health services and public health routinely estimate population health indicators from a range of data sources. These indicators are used in many settings to describe health status, monitor quality of care, and evaluate the effect of interventions. The data and knowledge necessary to calculate indicators, however, are scattered across different health settings, resulting in inconsistent and fragmented indicators and an inefficient use of population health information in research and practice. The Population Health Record (PopHR) described in this paper is an informatics platform for semi-automated integration of disparate data to enable measurement and monitoring of population health status and determinants. The research and development to build the PopHR uses AI methods to perform many tasks, including calculation of indicators and interaction with users.

In this paper we offer an approach for reasoning about resource allocation and scheduling in multiagent systems that takes into consideration the costs of preempting an agent from its current task. We apply our methodology to the motivating medical application of allocating doctors to patients in hospitals during mass casualty incidents and demonstrate noticeable improvements in performance (generating far fewer problem patients) over competing approaches that do not model the costs of preemption in sufficient detail. In particular, our approach offers a method for addressing the challenges of cyclical dependencies in the estimation of preemption costs by localized agents through a combination of planning techniques.

Multi-disciplinary shared care is based around a single, patient-centred health record. A key driver for storing that record electronically is the need to gather data once (for clinical care) and to reuse it for secondary purposes, including clinical studies. However, physicians working in different specialties may have different perspectives on that record, both when entering new data for clinical use and when reusing those data in clinical studies. The ORCHID classification scheme in use at the Nottingham University Hospitals NHS Trust in the UK, is an ontology-based model which supports multiple, simultaneous clinical perspectives yet allows data to be stored as standard HL7 CDA documents in an immutable, patient-centred record. This paper describes the basic mechanisms used to support those multiple perspectives and the solution to specific problems of recording diagnosis with co-morbidities and recording different levels of detail in disease phenotypes.

Some of the most successful Poker agents that participate in the Annual Computer Poker Competition (ACPC) use an almost zero regret strategy: a strategy that approximates a Nash Equilibrium. However, it is still unfeasible to efficiently compute a Nash Equilibrium without some sort of information set abstraction due to the size of Poker’s search tree. One popular technique for abstracting Poker information sets is to group hands with similar Expected Hand Strength ( E [ HS ]) and thus play them in the same way. For large Poker variants, algorithms like CFR might need to calculate E [ HS ] billions of times, when the game abstraction is so large that it cannot be pre-computed, implying that E [ HS ] must be determined online. This way, improving the efficiency of this method would certainly reduce the computation time needed by CFR for these cases. In this paper we describe Average Rank Strength; a technique based on a pre-computed lookup table that speeds up E [ HS ] computation. Ours results demonstrate speed improvements of about three orders of magnitude and negligible results difference, when compared to the original E [ HS ].

Poker is increasingly becoming an area of interest in AI research, partly because of the complex qualities it exhibits which are absent from more traditionally studied games, such as chess. One of the most difficult but also most important aspects of poker is the need to infer information about your opponent while also handling his attempts at disinformation. This problem of ``opponent modelling" is a central aspect of poker agent design and has been approached in many different ways. In this paper we focus on one subset of the opponent modelling problem, namely that of bluff detection. We explore the effectiveness of different feature sets towards this task and test the ease with which the bluffs of various poker agents can be detected.

Opponents are characterized by a Bayesian network intended to guide Monte-Carlo Tree Search through the game tree of No-Limit Texas Hold'em Poker. By using a probabilistic model of opponents, the network is able to integrate all available sources of information, including the infrequent revelations of hidden beliefs. These revelations are biased, and as such are difficult to incorporate into action prediction. The proposed network mitigates this bias via the expectation maximization algorithm and a probabilistic characterization of the hidden variables that generate observations. 

In poker, players tend to play sub-optimally due to theuncertainty in the game. Payoffs can be maximized byexploiting these sub-optimal tendencies. One way of realizingthis is to acquire the opponent strategy by recognizingthe key patterns in its style of play. Existing studieson opponent modeling in poker aim at predicting opponent’sfuture actions or estimating opponent’s hand.In this study, we propose a machine learning methodfor acquiring the opponent’s behavior for the purpose ofpredicting opponent’s future actions.We derived a numberof features to be used in modeling opponent’s strategy.Then, an ensemble learning method is proposed forgeneralizing the model. The proposed approach is testedon a set of test scenarios and shown to be effective.

Accurate and detailed measurement of an individual's physical activity is a key requirement for helping researchers understand the relationship between physical activity and health. Accelerometers have become the method of choice for measuring physical activity due to their small size, low cost, convenience and their ability to provide objective information about physical activity. However, interpreting accelerometer data once it has been collected can be challenging. In this work, we applied machine learning algorithms to the task of physical activity recognition from triaxial accelerometer data. We employed a simple but effective approach of dividing the accelerometer data into short non-overlapping windows, converting each window into a feature vector, and treating each feature vector as an i.i.d training instance for a supervised learning algorithm. In addition, we improved on this simple approach with a multi-scale ensemble method that did not need to commit to a single window size and was able to leverage the fact that physical activities produced time series with repetitive patterns and discriminative features for physical activity occurred at different temporal scales.