Grandmaster level in StarCraft II using multi-agent reinforcement learning.

Many real-world applications require artificial agents to compete and coordinate with other agents in complex environments. As a stepping stone to this goal, the domain of StarCraft has emerged as an important challenge for artificial intelligence research, owing to its iconic and enduring status among the most difficult professional esports and its relevance to the real world in terms of its raw complexity and multi-agent challenges. Over the course of a decade and numerous competitions1-3, the strongest agents have simplified important aspects of the game, utilized superhuman capabilities, or employed hand-crafted sub-systems4. Despite these advantages, no previous agent has come close to matching the overall skill of top StarCraft players. We chose to address the challenge of StarCraft using general-purpose learning methods that are in principle applicable to other complex domains: a multi-agent reinforcement learning algorithm that uses data from both human and agent games within a diverse league of continually adapting strategies and counter-strategies, each represented by deep neural networks5,6. We evaluated our agent, AlphaStar, in the full game of StarCraft II, through a series of online games against human players. AlphaStar was rated at Grandmaster level for all three StarCraft races and above 99.8% of officially ranked human players.

The Primary Physis.

The primary physis is responsible for long bone growth in children and adolescents. Injury and physiologic or metabolic stress to the primary physis present unique radiologic findings that are important for radiologists to recognize and diagnose. Appreciation of the anatomy and histology of the primary physis forms the basis for understanding the imaging findings associated with pathologic conditions affecting the primary physis. Salter-Harris injuries, physeal bars, growth arrest lines, rickets, and focal periphyseal edema zones are common conditions with recognizable radiologic presentations. Proper diagnosis of these primary physeal conditions will aid in the treatment of affected pediatric patients.

Comparing Performance of Risk Scores for Combined Aortic Valve Replacement and Coronary Bypass Grafting Surgery.

BACKGROUND: Aortic valve replacement (AVR) and/or coronary artery bypass grafting (CABG) make up the majority of cardiac surgery with increasing demand as the population ages. Accuracy of risk stratification is important, in predicting adverse outcomes and selecting modality of intervention, but has been rarely studied for the combined AVR+CABG operation. We compared the prognostic utility of EuroSCORE, EuroSCORE II and Society of Thoracic Surgeons' (STS) Score for AVR+CABG. METHODS: All patients (n=450) undergoing AVR+CABG at Auckland City Hospital during 2005-2012 with mean follow-up of 4.7+/-2.5 years were included. The three risk scores were calculated and their discrimination and calibration for mortality and morbidities assessed. RESULTS: Operative mortality was 6.4% (29), and mean scores were EuroSCORE 12.5+/-11.1%, EuroSCORE II 6.6+/-6.1% and STS Score 5.5+/-4.4%. C-statistics were 0.587, 0.669 and 0.699 respectively for operative mortality, Hosmer-Lemeshow test P-values were 0.064, 0.718 and 0.567, and Brier Score 0.716, 0.585 and 0.588. Independent predictors of operative mortality were history of myocardial infarction and impaired renal function. Society of Thoracic Surgeons' score also was the most accurate score for predicting mortality during follow-up (c=0.663), composite morbidity (c=0.627), stroke (c=0.642), prolonged ventilation>24hours (c=0.642), and return to theatre (c=0.612). CONCLUSION: The STS score has the best discriminative ability for mortality and the majority of complications after AVR+CABG, while its calibration was similar to EuroSCORE II and superior to EuroSCORE. It should therefore be used for risk stratification and when considering surgical versus percutaneous intervention in those with concurrent aortic valve and coronary artery disease.

Comparison of four contemporary risk models at predicting mortality after aortic valve replacement.

OBJECTIVE: Risk stratification for aortic valve replacement (AVR) is desirable given the increased demand for intervention and the introduction of transcatheter aortic valve implantation. We compared the prognostic utility of the European System for Cardiac Operative Risk Evaluation (EuroSCORE), EuroSCORE II, Society of Thoracic Surgeons (STS) score, and an Australasian model (Aus-AVR score) for AVR. METHODS: We retrospectively calculated the 4 risk scores for patients undergoing isolated AVR at Auckland City Hospital from 2005 to 2012 and assessed their discrimination and calibration for short- and long-term mortality. RESULTS: A total of 620 patients were followed up for 3.8 ± 2.4 years, with an operative mortality of 2.9% (n = 18). The mean EuroSCORE, EuroSCORE II, STS score, and Aus-AVR score was 8.7% ± 8.3%, 3.8% ± 4.7%, 2.8% ± 2.7%, and 3.2% ± 4.8%, respectively. The corresponding C-statistics for operative mortality were 0.752 (95% confidence interval [CI], 0.652-0.852), 0.711 (95% CI, 0.607-0.815), 0.716 (95% CI, 0.593-0.837), and 0.684 (95% CI, 0.557-0.811). The corresponding Hosmer-Lemeshow test P and chi-square values for calibration were .007 and 21.1, .125 and 12.6, .753 and 5.0, and .468 and 7.7. The corresponding Brier scores were 0.0348, 0.0278, 0.0276, and 0.0294. Independent predictors of operative mortality included critical preoperative state, atrial fibrillation, extracardiac arteriopathy, and mitral stenosis. The log-rank test P values were all <.001 for mortality during follow-up for all 4 scores, stratified by quintile. CONCLUSIONS: All 4 risk scores discriminated operative mortality after isolated AVR. The EuroSCORE had poor calibration, overestimating operative mortality, although the other 3 scores fitted well with contemporary outcomes. The STS score was the best calibrated in the highest quintile of operative risk.