Win Rate Research Articles

SpecMCTS: Accelerating Monte Carlo Tree Search Using Speculative Tree Traversal

Monte Carlo Tree Search (MCTS) algorithms show outstanding strengths in decision-making problems such as the game of Go. However, MCTS requires significant computing loads to evaluate many nodes in the decision tree to make a good decision. Parallelizing MCTS node evaluations is challenging because MCTS is a sequential process that each round of tree traversal depends on the previous node evaluations. In this work, we present SpecMCTS , a new approach for accelerating MCTS by speculatively traversing the search tree. Many MCTS applications, such as AlphaGo Zero, use a deep neural network (DNN) model to evaluate the tree nodes during the search. SpecMCTS uses a pair of DNN models, the speculation model and the main model . The faster (but less accurate) speculation model accelerates the sequential tree search while the more accurate main model improves the decision quality. SpecMCTS accelerates MCTS for the game of Go by up to $2.09\times {}$ on the NVIDIA T4 GPU. This performance improvement can be translated into a better decision quality by performing a larger number of tree traversals within the time limit. For a fixed decision time, SpecMCTS shows stronger gameplay (higher win rate) than the original sequential MCTS and state-of-the-art MCTS parallelization approaches.

Do fee-shifting rules affect plaintiffs’ win rates? A theoretical and empirical analysis

This article studies whether fee-shifting rules can affect plaintiffs’ win rates. Beyond theoretical modeling, this study goes a step further and provides empirical evidence on this issue, thanks to the study of a real change in Spanish legislation. Spain applied the so-called English rule in 2011 in the administrative jurisdiction. This study explores whether plaintiffs were more or less successful when litigating against public administrations since then. After controlling for several other factors, it is shown that the new rule considerably increased plaintiffs’ win rate (while also reducing litigation). The paper also presents a theoretical model explaining that the English rule can increase the prevalence of judgments in favor of plaintiffs through two channels: the selection of cases brought before the courts by interested parties and the level of deference paid by courts to the decisions under review.

Human-AI Collaboration in a Cooperative Game Setting

Human-AI interaction is pervasive across many areas of our day to day lives. In this paper, we investigate human-AI collaboration in the context of a collaborative AI-driven word association game with partially observable information. In our experiments, we test various dimensions of subjective social perceptions (rapport, intelligence, creativity and likeability) of participants towards their partners when participants believe they are playing with an AI or with a human. We also test subjective social perceptions of participants towards their partners when participants are presented with a variety of confidence levels. We ran a large scale study on Mechanical Turk (n=164) of this collaborative game. Our results show that when participants believe their partners were human, they found their partners to be more likeable, intelligent, creative and having more rapport and use more positive words to describe their partner's attributes than when they believed they were interacting with an AI partner. We also found no differences in game outcome including win rate and turns to completion. Drawing on both quantitative and qualitative findings, we discuss AI agent transparency, include design implications for tools incorporating or supporting human-AI collaboration, and lay out directions for future research. Our findings lead to implications for other forms of human-AI interaction and communication.

Word Autobots: Using Transformers for Word Association in the Game Codenames

Winning the social game Codenames involves combining cooperative and language understanding capabilities. We developed six cooperative bots designed to play the Codemaster and Guesser roles in the Codenames AI Competition and tested them using the provided framework and a round-robin tournament set. The bots are based on term frequency - inverse document frequency (TF-IDF), Naive-Bayes and GPT-2 Transformer word embedding. Additionally, Transformer-based bots were assessed and compared with the concatenation of word2vec and GloVe baseline bot developed by Codenames AI Competition creators. Results from this Transformer implementation rivals the concatenated bot in terms of win rates and guess precision and outperforms it in terms of minimum and average turns taken to win the game and training data load time. Additionally, in an initial evaluation performed with 10 human players, the Transformer agent performed slightly better than the baseline as Codemaster, but worse as a Guesser.

Utilization of Simulation and Machine Learning to Analyze and Predict Win Rates of the Characters Battle

Utilization of Simulation and Machine Learning to Analyze and Predict Win Rates of the Characters Battle

Portfolio Management of High Growth Firms and Technical Buy Points

This paper shows that current earnings and prices at or near new highs can be combined with the Average Directional Index (DM) technical indicator to generate excess return in the market. Excess returns are statistically significant via difference in means tests (robust to recession only periods, and additional timeframes) and CAPM, Fama and French 3 and 5 factor models. The model is robust to alternate weight specification and has consistent results across multiple start dates. The system is replicable and has a win rate of 63% turning $100,000 into $33,000,000. It is constructed without lookahead bias.

Changes in perceptions of mental fatigue during a season in professional under-23 English Premier League soccer players

ABSTRACT The present study assessed changes in academy soccer players’ perception of mental fatigue (MF) across a competitive season, investigating the relationship between MF and other subjective measures of wellness. Ten players completed a modified Brief Assessment of Mood (BAM+) questionnaire that included the question: “How mentally fatigued do you feel”? on match-day (MD) and one (MD+1), two (MD+2) and three (MD+3) days post-match (35 matches). Players reported their MF, along with other subjective measures (sleep, muscle soreness, fatigue and motivation). Results found MF was elevated on MD+1 (43±1 mm) compared to all other days (all P≤0.001). Players reported lower MF on MD+1 in the late-season phase (34±2 mm) compared to both early- (50±2 mm, P≤0.001) and mid-season (46±2 mm, P≤0.001). This coincided with an 80%-win rate in the late-season phase versus the early- (33%) and mid-season (50%). There were very strong repeated-measures correlations between changes in MF and sleep (r=−0.77), muscle soreness (r=0.94), fatigue (r=0.92) and motivation (r=−0.89; all P ≤ 0.0005). In conclusion, MF was closely aligned to match success and other wellness variables. This data suggests a potential lack of sensitivity for identifying MF using a subjective questionnaire. Therefore, researchers and practitioners could work together to identify other ways of practically assessing MF.

Model Construction of Female Professional Volleyball V-league Season Win Rate

Model Construction of Female Professional Volleyball V-league Season Win Rate

Learning the Game of Go by Scalable Network Without Prior Knowledge of Komi

AlphaGo trains a value network to predict the win rate of the current state with 7.5 komi on a 19 × 19 board. The komi of most rectangular boards is unknown, so we do not know who the winner is at the end of the game. We need to use the human experience to guess a komi and then train the value network with this komi. Therefore, the accuracy of the value network is related to the accuracy of the guess. This article uses the board value network to calculate the score of the current state and tries to maximize the score. Then, we do not need to guess the komi. We also modify the network structure to support the board with arbitrary board size as input. Furthermore, we can transfer knowledge of the small board to the large board. We propose an algorithm that can adapt to the bonus rule. We have experimentally proved that our method is effective on a small board and has the ability to transfer knowledge to the large board. In order to better understand the learning process, we visualize the policy and score of some major branches. Finally, we show the solution that our program obtained on 6 × 6, 6 × 7, and 7 × 8 boards.

AI-based competition of autonomous vehicle fleets with application to fleet modularity

Because operational environments change over time and technology upgrades are common in fleets of ground vehicles, a large number of vehicles quickly become obsolete. A possible solution is to develop fleets of modular vehicles, which are built with interchangeable components, i.e. modules. This paper evaluates the performance of a future reconfigurable and autonomous vehicle fleet in a high fidelity military operation scenario. The military fleet operates in a hostile environment under a high risk of damage and needs to react to adversarial actions in real-time. The operation decisions are numerous including vehicle reconfiguration, relocation, damage recovery, and dispatch decisions. Given the limited resources and time delays in the operation, decision effectiveness and foresightedness are necessities, which require a good understanding of the adversary, close collaboration among commanders, and breaking of the equilibrium between adversaries. To capture these characteristics, we formulate an intelligent agent-based model for the decision-making process during fleet operations by combining real-time optimization with artificial intelligence. With continuous updating of learning models, the intelligent agents refine their decisions during interactions with the environment and other agents, and evolve their competition strategies according to adversarial historical behaviors. With the same level of resources, the conventional fleet wins when the dispatch decisions are stochastic. However, once each fleets start to learn from each other’s behavior, the modular fleet outperforms the conventional fleet. The strategic and operational benefits of fleet modularity are revealed and discussed in terms of win rate, adaptability, unpredictability and damage recovery.

Do the Affluent Override Average Americans? Measuring Policy Disagreement and Unequal Influence

AbstractObjectiveIn this article, I seek to adjudicate between contradictory findings in studies of unequal responsiveness that focus on policy issues over which affluent and average Americans disagree.MethodsI assess the impact of 22 definitions of policy disagreement and two methods of measuring policy influence—based on win rates and policy change rates—on analyses of unequal responsiveness.ResultsWin rates are an unreliable measure of influence and comparing policy change rates across cases of agreement and disagreement minimizes bias and allows for a more direct measure of relative independent policy influence. When income groups disagree over proposed policy changes, the preferences of the affluent—but not those of the middle—significantly impact the likelihood that the policies are implemented.ConclusionMy results support previous findings that the affluent have substantial influence over policy making while average Americans have little to no influence.

A Tale of Two Forums: Employment Discrimination Outcomes in Arbitration and Litigation

This article presents data from a novel survey of 1,256 employment plaintiff attorneys to test whether employee rights and remedies are affected by mandatory employment arbitration. By surveying attorneys directly about their most recent employment discrimination cases taken to verdict in arbitration and civil litigation, the author presents a systematic empirical comparison of outcomes between civil courts and arbitration with robust controls. The ability to control for the legal basis of the claim, defendant size, use of summary judgment, and attorney and plaintiff characteristics significantly improves on previous empirical research studies. Consistent with previous research, employee win rates in arbitration are lower than those found in state and federal court. In addition, monetary award amounts and percentage of claim amount awarded to employees who prevail in their cases are significantly lower in arbitration compared to outcomes in state and federal jury trials.

Market Timing with the Backward SADF Test

In this article, we explore the possibility of acquiring positive returns from the stock market during bubble periods. A structural break test is used to detect bubbles, and a technical statistic is applied to decide the direction of the bet. Unlike most of the previous articles on trading strategies, we present the out-of-sample return distributions of the strategy with respect to different trading rules based on three stock data sets, which helps investors understand the profitability and risk of the method. The result shows that our strategy is highly robust across different data sets, as it can acquire positive returns for more than 95% of the cases, with the highest annualized averaged return per trade at more than 36%. Additionally, the win rates are significantly above 0.5 for more than 95% of the cases. TOPICS:Quantitative methods, Statistical methods, Simulations Key Findings • The backward SADF test was designed to detect recently formed exponential patterns in time series. • The proposed strategy applies the backward SADF test and MACD statistics to exploit the statistical arbitrage from the exponential patterns. • We validate the strategy by comparing it with the buy-and-hold and daily-rebalance strategies and by running the Hendrickson–Merton based on three different data sets.

Accelerating and Improving AlphaZero Using Population Based Training

AlphaZero has been very successful in many games. Unfortunately, it still consumes a huge amount of computing resources, the majority of which is spent in self-play. Hyperparameter tuning exacerbates the training cost since each hyperparameter configuration requires its own time to train one run, during which it will generate its own self-play records. As a result, multiple runs are usually needed for different hyperparameter configurations. This paper proposes using population based training (PBT) to help tune hyperparameters dynamically and improve strength during training time. Another significant advantage is that this method requires a single run only, while incurring a small additional time cost, since the time for generating self-play records remains unchanged though the time for optimization is increased following the AlphaZero training algorithm. In our experiments for 9x9 Go, the PBT method is able to achieve a higher win rate for 9x9 Go than the baselines, each with its own hyperparameter configuration and trained individually. For 19x19 Go, with PBT, we are able to obtain improvements in playing strength. Specifically, the PBT agent can obtain up to 74% win rate against ELF OpenGo, an open-source state-of-the-art AlphaZero program using a neural network of a comparable capacity. This is compared to a saturated non-PBT agent, which achieves a win rate of 47% against ELF OpenGo under the same circumstances.

Analysis of Kelly betting on finite repeated games

The Kelly criterion can be used to maximize returns in a game with win rate p and odds b; however, optimization theoretically requires wagering over an infinite number of time steps. Despite the fact that Kelly's theory has been extended to most of the trading strategies used in financial markets, there is still a large gap between the theoretical determination of optimal bidding fractions and practical application of these methods. In this paper, we illustrate the difference between the theoretical and simulation results obtained from a gambling situation involving a finite number of bidding steps T( = W + L), where W and L respectively denote the numbers of wins and losses. The optimal bidding fraction based on the Kelly criterion should employ the win\\loss proportion W/T rather than the win rate p; however, it is not possible to obtain the value of a priori. W. Thus, profits under the Kelly formula are calculated by applying win rate p and the win\\lose proportion W/T. In this paper, we denote pt as the current win\\lose proportion before time step t as an alternative to win rate p. The proposed approach does away with the need to apply win rate p and produces profits that are nearly optimal under Kelly betting.

The three and six-substitution rules in football: A preliminary comparative analysis in quantitative replacing, game statistics, win rate and winning probability

Abstract Aims: The present study aimed to analyze the effects of the 3- and 6-substitution rule and qualification round on substitution pattern, game statistics, win rate, and winning probability in Brazil's largest U-20 football tournament. Methods: We analyzed 1,339 matches from the Sao Paulo Junior Football Cup implementing a before-after design considering 6 years (three seasons for 3-subs and three for 6-subs). Results: Approximately 31% of teams used the total number of replacements of 6-subs with the same substitution pattern between rounds. The win rate for both substitution rules was approximately 79% (OR = 3.10, 95%CI [2.18, 4.51]) with a small reduced magnitude (approximately 5%) in 6-subs knockout games. Logistic regression analysis showed that the 6-subs regulation tournament predicted a reduced probability (-12%) of a match ending with a win in the knockout round. There were 6-subs knockout effects (p < 0.05) on game statistics with “negligible” effect size (ES) on “effective playing time” (+1 min; ES = 0.15); decreased “goal scoring” (-0.3;ES = 0.09) and “red cards” per match (-0.1; ES = 0.06). An ordered probit model and partial effects showed that including a replacement unit on a team increased the winning probability by 8% in the group stage, and reduced the chance (-4%) for the opponents trying to do the same in the knockout stage. Conclusion: Overall, 6-subs caused minor effects on the game statistics, win rate, and winning probability. Understanding substitution strategies and motivation, individual and collective performance effects, and whether there will be changes in the game profile will become relevant research issues when a greater number of substitutions rules are universally instituted.

An Empirical Study of Gender Differences in Competitive High School Debate

Gender differences in competitive environments have been studied extensively; however, the majority of literature concerns activities with an objective winner. In this paper, we examine gender differences in high school Public Forum debate, a two vs. two activity where judges subjectively decide the winner. We constructed a dataset with 44 variables and 125,087 unique debate rounds during the 2014-2015 to 2019-2020 school years. Using logistic modeling, we document a large difference in win rates between teams of different gender compositions, with female-female teams 17.1% less likely and male-female teams 10.0% less likely to win a debate round against male-male teams. However, there is no gender gap in win rates for novice debaters, suggesting that the disparity does not occur from innate entry ability differences but rather appears alongside experience in debate. We also find a large difference in participation rates between female and male students, which begins at the 9th grade level and is exacerbated over time due to female debaters being 30.34% more likely to quit than male debaters. Finally, we find that a higher ratio of female to male debaters from the same school reduces attrition rates of female debaters.

Differential Evolution Optimization of the Broken Wing Butterfly Option Strategy

The Broken Wing Butterfly (BWB) has become a popular options strategy for traders. Profit is generated primarily by exploiting option value time decay. In this paper the selection of the option strikes to be used along with trade entry and exit parameters, such as time to expiration and profit and loss targets, are optimized using over a decade of historical option data of the S & P 500 exchange traded fund (symbol: SPY). The importance of selecting an optimal strike mapping method, by which strikes are assigned in any time period, is highlighted. Of the three methods considered, the normalized strike mapping method was found to be optimal. Optimization was performed using a differential evolution (DE) evolutionary algorithm. The objective function used for optimization considered final cumulative profit, volatility, and maximum equity drawdown while achieving a high trade win rate. A trade example is given to illustrate the use of the obtained results.

Analytics and Operations Research Increases Win Rates for IBM’s Information Technology Service Deals

IBM Research in partnership with IBM Services developed an innovative set of tools that automatically and cognitively reads a request for proposal, extracts client requirements, and maps them to IBM offerings. It also performs accurate costing and recosting, pricing and repricing, and market benchmarking of the bid, and it predicts the status over time of the various deals being pursued to effectively manage IBM’s sales pipeline and align salesforce resources.

Empirically Investigating the Source of the Repeat Player Effect in Consumer Arbitration

Policymakers, courts, and scholars have long been interested in whether repeat players enjoy an advantage in forced arbitration. Sophisticated empirical studies of consumer and employment awards reveal that there is indeed a repeat player effect: even controlling for other factors, companies that arbitrate more than once boast higher win rates than one-shot firms. However, researchers have not yet tried to determine whether this repeat player effect is a product of experience within the arbitral forum (the “experience” hypothesis) or characteristics of the repeat playing companies themselves (the “defendant-specific” hypothesis). This Article begins to address this issue by analyzing 4,570 consumer arbitration awards from the American Arbitration Association. Using a unique regression specification that includes both discrete and continuous random variables (a combination of functional forms not used in previous literature), it finds that the repeat player effect is more consistent with the defendant-specific hypothesis than it is with the experience hypothesis. Indeed, to the extent that repeat-playing businesses enjoy an advantage in arbitration, it likely emanates from company-specific characteristics.