Training machine learning models on large datasets, such as those derived from single-cell RNA sequencing (scRNA-seq), poses significant challenges due to high computational and memory demands. Additionally, integrating data from diverse sources introduces complexities stemming from experimental variability, technological differences, and data privacy concerns. Continual learning (CL) algorithms offer a promising solution by incrementally training models on data batches while addressing issues like catastrophic forgetting. However, the sequence in which these batches are presented has been largely overlooked despite its potential to influence learning efficiency and model performance. This study introduces a confidence-based batch ordering strategy for CL algorithms, leveraging confidence estimation to prioritize training samples. By structuring batches in ascending order of confidence, we observed consistent improvements in classification performance across multiple scRNA-seq datasets. Specifically, intra-dataset experiments revealed that ascending confidence ordering consistently outperformed random or descending orderings in terms of median F1 scores, highlighting its efficacy in enhancing model generalization. Similarly, inter-dataset analyses demonstrated that confidence-based ordering improved robustness when training on heterogeneous datasets generated using different sequencing protocols. Our findings highlight the critical role of batch sequencing in optimizing CL workflows for data-intensive tasks. Future research may explore the extension of this strategy to other domains and investigate adaptive confidence metrics tailored to dynamic datasets.

Expert video game players exhibit unique behaviors compared to their less experienced counterparts. Such behaviours may also influence physiological aspects such as blinks and eyelid movements. In this study, we used the Eye Aspect Ratio (EAR) signal from a webcam to investigate the complex dynamics of eyelid movements among players with different levels of expertise in Tetris. We measured complex dynamics using recurrence quantification analysis (RQA) based measures (Determinism, Laminarity, Average Diagonal Line, and Trapping Time). Our results show that expert Tetris players display more complex patterns in their eyelid behaviour, but also that some of the measures obtained using RQA correlate directly with player actions (keys pressed) and events in Tetris (numbers of lines cleared). This study provides the first example of a direct connection between RQA measures extracted from the EAR signal and behavior displayed in a game. Our results also demonstrate the potential of using RQA measures extracted from the EAR in analysing human behavior during other screen-presented tasks.

Serious games are an effective method of reproducing aspects of the complex interplay between environments and stakeholders in business situations. In the game, we describe here,The Sustainable Port, players experience what it is like to make decisions in such a complex environment. Their aim in the game is to grow the Port of Rotterdam while keeping economic growth in balance with sustainability goals. In this study, we assessed whether experienced Port of Rotterdam employees (PoR employees) show different psychophysiological patterns, and more specifically eye aspect ratio (EAR)-derived features, compared to students. We did this on the assumption that physiological patterns will tell us something about how people who are familiar with the environment of the Port of Rotterdam, more specifically PoR employees, make decisions compared to those lacking such familiarity. Our sample consisted of 28 PoR employees and 65 students, all of whom playedThe Sustainable Portgame and had their faces recorded with a camera. The EAR was extracted from these recordings, and then from those, we extracted EAR-derived features. Our results show that PoR employees perform better than students and that the two groups are characterized by different physiological variations in their EAR-derived features. A logistic regression model used to identify PoR employees and students obtained an F1 score of 0.62, an area under the precision–recall curve score of 0.64, and an ROC AUC score of 0.70. Such a performance significantly above baseline suggests the effectiveness of using EAR-derived features for this task. Our interpretation was further confirmed by a pseudo-R2 score used to evaluate the goodness of fit of a logistic regression model on the entire dataset. We found that PoR employees had a lower variation in blink rate per minute and higher variation in the root mean square of the successive differences in blinks (RMSSD), the consecutive difference between two continuous blinks. Moreover, this study shows that our methods were robust enough to negate the effects of confounders, such as biological sex and age, that affect some other studies that analyze blinks.

This study aimed to evaluate if eye blinks can be used to discriminate players with different performance in a session of Nintendo Entertainment System (NES) Tetris. To that end, we developed a state-of-the-art method for blink extraction from EAR measures, which is robust enough to be used with data collected by a low-grade webcam such as the ones widely available on laptop computers. Our results show a significant decrease in blink rate per minute (blinks/m) during the first minute of playing Tetris. After having defined 3 groups of proficiency based on in-game performance (Novices, Intermediates, and Experts) we found out that expert players display a significantly lower decrease in blinks/m compared to novices during the first minute of gameplay, which shows that Tetris players' proficiency can be detected by looking at eye blinks/m variations during the early phase of a game session. This difference in blinks/m is observed throughout the entire game session, which supports the general conclusion that proficient Tetris players have a lower decrease in blinks/m, even when playing more difficult levels. Finally, we offer some interpretations of this effect and the relationship that our results may have with the visual cognitive workload experienced during the gameplay

Virtual reality (VR) research has provided overviews of locomotion techniques, how they work, their strengths and overall user experience. Considerable research has investigated new methodologies, particularly machine learning to develop redirection algorithms. To best support the development of redirection algorithms through machine learning, we must understand how best to replicate human navigation and behaviour in VR, which can be supported by the accumulation of results produced through live-user experiments. However, it can be difficult to identify, select and compare relevant research without a pre-existing framework in an ever-growing research field. Therefore, this work aimed to facilitate the ongoing structuring and comparison of the VR-based natural walking literature by providing a standardised framework for researchers to utilise. We applied thematic analysis to study methodology descriptions from 140 VR-based papers that contained live-user experiments. From this analysis, we developed the LoCoMoTe framework with three themes: navigational decisions, technique implementation, and modalities. The LoCoMoTe framework provides a standardised approach to structuring and comparing experimental conditions. The framework should be continually updated to categorise and systematise knowledge and aid in identifying research gaps and discussions.

Traditionally, the relative strength of a chess player within a competitive pool is identified by a rating number. In order to reach a fair rating that best represents their level of play, chess players are required to play numerous games against various opponents within that pool. However, intuitively, experienced chess players are capable of extracting a rough estimate of a player’s strength by looking at the moves they made in a single game. How accurately could a machine learning model based on a large dataset of chess games predict player ratings from a single game, and what would these predictions depend on? This paper presents an attempt to identify, encode and model chess gameplay features in order to predict a player’s rating from a single game played. If successful, such a model could be employed to attach a fair initial rating to a new player within a pool before any games are played. We use an extensive dataset of chess games downloaded from a popular online chess platform, from which we extract a set of 30 features which are used to model and ultimately predict players’ ratings. Our findings show that we are capable of predicting the rating bracket of a player with 79.3% accuracy when considering the extreme ends of the dataset (lowest vs. highest rated players), while the accuracy consistently drops as we increase the respective bracket width. We discovered that the most important features of our predictive models are both theory-and engine-related; most importantly, the features that we have extracted lead to explainable, quantifiable predictions of chess player strength.

While quality estimation (QE) can play an important role in the translation process, its effectiveness relies on the availability and quality of training data. For QE in particular, high-quality labeled data is often lacking due to the high-cost and effort associated with labeling such data. Aside from the data scarcity challenge, QE models should also be generalizabile, i.e., they should be able to handle data from different domains, both generic and specific. To alleviate these two main issues — data scarcity and domain mismatch — this paper combines domain adaptation and data augmentation within a robust QE system. Our method is to first train a generic QE model and then fine-tune it on a specific domain while retaining generic knowledge. Our results show a significant improvement for all the language pairs investigated, better cross-lingual inference, and a superior performance in zero-shot learning scenarios as compared to state-of-the-art baselines.

Games are designed to elicit strong emotions during game play, especially when players are competing against each other. Artificial Intelligence applied to predict a player’s emotions has mainly been tested on single-player experiences in low-stakes settings and short-term interactions. How do players experience and manifest affect in high-stakes competitions, and which modalities can capture this? This paper reports a first experiment in this line of research, using a competition of the video game Hearthstone where both competing players’ game play and facial expressions were recorded over the course of the entire match which could span up to 41 minutes. Using two experts’ annotations of tension using a continuous video affect annotation tool, we attempt to predict tension from the webcam footage of the players alone. Treating both the input and the tension output in a relative fashion, our best models reach 66.3% average accuracy (up to 79.2% at the best fold) in the challenging leave-one-participant out cross-validation task. This initial experiment shows a way forward for affect annotation in games “in the wild” in high-stakes, real-world competitive settings.

In this study, we predict the different levels of performance in a Nintendo Entertainment System (NES) Tetris session based on the score and the number of matches played by the players. Using the first 45 seconds of gameplay, a Random Forest Classifier was trained on the five keys used in the game obtaining a ROC_AUC score of 0.80. Further analysis revealed that the number of down keys (forced drop) and the number of left keys (left translation) are the most relevant keys in this task, showing that by merely including the data from these two keys our Random Forest Classifier reached a ROC_AUC score of 0.83. We conclude that the keylogger data during the early phases of a game session can be successfully used to predict performance in longer sessions of Tetris.

In this study, we extracted facial action units (AUs) data during a Hearthstone tournament to investigate behavioural differences between expert, intermediate, and novice players. Our aim was to obtain insights into the nature of expertise and how it may be tracked using non-invasive methods such as AUs. These insights may shed light on the endogenous responses in the player and at the same time may provide information to the opponents during a competition. Our results show that player expertise may be characterised by specific patterns in facial expressions. More specifically, AU17 (chin raiser), AU25 (lips apart), and AU26 (jaws drop) intensity responses during gameplay may vary according to players’ expertise. Such results were obtained by training a random forest classifier to test whether we can use these three AUs alone to accurately detect players’ expertise. The classifier reached 0.75 accuracy on 5-fold cross-validation, after balancing the class weights, and 0.85 after having applied the Synthetic Minority Over-sampling Technique (SMOTE) function. These results suggest that AUs can be effectively used to discriminate different levels of expertise in competitive video game players.

In this study, we analyzed the blinking behavior of players in a video game tournament. We aimed to test whether spontaneous blink patterns differ across levels of expertise. We used blink rate (blinks/m), blink duration, and general eyelid movements represented by features extracted from the Eye Aspect Ratio (EAR) to train a machine learning classifier to discriminate between different levels of expertise. Classifier performance was highly influenced by features such as the mean, standard deviation, and median EAR. Moreover, further analysis suggests that the blink rate is likely to increase with the level of expertise. We speculate this may be indicative of a reduction in cognitive load and lowered stress of expert players. In general, our results suggest that EAR and blink patterns can be used to identify different levels of expertise of video game players.

Board gaming as a leisure activity is becoming increasingly popular. The research on modern commercial board gaming is also gaining momentum. In this study, we aim to investigate the factors that may play a role in board game purchase intentions. We conducted an online survey and collected data from habitual board gamers. Multiple regression analyses showed that enjoyment, positive word of mouth, age and gender were positively associated with purchase intentions whereas income, play frequency, prior board gaming experience and feelings of presence were not. We discuss the results and present potential future research.

Esports have evolved into a major form of entertainment, drawing hundreds of millions of viewers to its online competitive broadcasts. Using Esports telemetry data to predict the outcome of a match is a well-researched topic, but micropredictions of specific in-game events are explored only sparingly. How accurately can we predict specific in-game events within a limited time window, and how can these predictions be used in a live broadcast? This research aims at predicting in-game deaths using telemetry data in Counter-Strike: Global offensive (CS:GO). We establish a data processing pipeline to acquire and re-structure raw in-game data and propose a set 36 features which will ultimately be used to predict in-game deaths within a three second window. Three neural network models are compared, namely convolutional (CNN), recurrent (RNN) and long short-term memory (LSTM). Our results show that the LSTM network has the best predictive accuracy (F1 0.38) when prompted, for all 10 players of a competitive game of CS:GO. The predictions are most influenced by features related to a player’s average in-game death count, health points, enemies in range and equipment value. Our model enables real-time micropredictions of deaths in CS:GO, and may be leveraged by Esports commentators and game observers to direct their focus on critical in-game events during a live competitive broadcast.

This paper investigates the development of an Artificial Intelligence (AI) agent which plays the voting phase of the board game Dixit. Given a set of open cards and a lexical “hint” provided by a player, our algorithm aims to predict which card the hint originally refers to. The AI agent is developed using Machine Learning (ML) algorithms for Natural Language Processing (NLP). The AI agent is equipped with models that explore data of human-played games and retrieves information from the internet to deal with any shortage of information. We show that the Dixit AI agent we developed is more accurate than the average human player in finding the card which corresponds to a hint.

In this article we examine playing styles in four popular massively multiplayer online games, namely,Battlefield 4,Counter-Strike,Dota 2, andDestiny. We investigate to what extent national culture influences these playing styles, and whether players from countries with similar cultures exhibit similar playing styles as well. We gathered playing style information from hundreds of thousands of players of these games, and applied correlation and clustering algorithms to relate playing styles to nationalities and to Hofstede cultural dimensions. We found that playing styles are influenced by nationality and cultural dimensions, and that there are clear similarities between the playing styles of similar cultures. In particular, the Hofstede dimension “Individualism” explained most of the variance in playing styles between national cultures for the games that we examined.

In this study, we used recordings of players’ facial expressions that are captured during competitive Hearthstone games to analyse the correlation between in-game player affective responses and subjective post-game self-reports. With this, we aimed to examine whether eye gaze, head pose and emotions gathered as objective data from face recordings would be associated with subjective experiences of players which were collected in the form of a post-game survey. Data was collected during a live offline Hearthstone competition, which involved a total of 17 players and 31 matches played. Correlation analyses between in-game and post-game variables show that players’ facial expressions and eye gaze measurements are associated with both players’ attention to the opponent and their mood influenced by the opponent. In future research, these results may be used to implement predictive player models.

Live Action Role-Playing (LARP) games and similar experiences are becoming a popular game genre. Here, we discuss how artificial intelligence techniques, particularly those commonly used in AI for Games, could be applied to LARP. We discuss the specific properties of LARP that make it a surprisingly suitable application field, and provide a brief overview of some existing approaches. We then outline several directions where utilizing AI seems beneficial, by both making LARPs easier to organize, and by enhancing the player experience with elements not possible without AI.

In this paper, we study the effect of time pressure on player behaviour during a dilemma-based crisis management game. We employ in-game action tracking, physiological sensor data and self-reporting in order to create multi-modal predictive models of player stress responses during a crisis management scenario. We were able to predict the experimental condition (time pressure vs. no time pressure) with 84.5% accuracy, using a game-only feature set. However, lower accuracy was observed when physiological sensor data was used for the same task. The method presented in this paper can be employed in crisis management training, aiming at assessing players’ responses to stressful conditions and manipulating player stress levels to provide personalised training scenarios.

Increasingly more scholars have regarded the virtual worlds of massively multiplayer online games (MMOGs) as a social laboratory, and have paid research attention to the online interactions between its large number of players. In the present study, we examine a widely observed real-life phenomenon gender homophily (i.e., people of the same gender flocking together) in this virtual context. Specifically, we investigate how collaboration networks in an MMOG (Destiny) are shaped by the adopted gender of the avatar characters. This focus is interesting, as avatar gender in video games is generally a choice that is less constrained than it is in real life. To investigate the effect of gender in online video games, while controlling for the effects of several confounding factors, we employed a technique called exponential random graph modeling. Mirroring how interpersonal relationships are often gendered in real life, and despite common phenomenon such as gender swapping, we found evidence supporting gender homophily in the MMOG environment.

With the constant rise in popularity of competitive video gaming (also known as Esports), Esports analytics has been a field of growing scientific interest in the recent years. Studies discussing player behaviour, skill learning and team performance have been conducted through Multiplayer Online Battle Arena games such as League of Legends. In this paper, we propose a multi-modal approach towards stress response modeling in competitive LoL games. We collect wearable physiological sensor data, mouse & keyboard logs and in-game data in order to study the relationship between player stress responses and in-game behaviour. We discuss the design criteria and propose future studies using the collected dataset.

In this paper, we propose a framework for personalised crisis management training through the use of an applied game. The framework particularly focuses on ubiquitously assessing and manipulating player stress levels during training, and evaluating player performance by providing personalised feedback. To achieve these goals, the framework leverages techniques for multi-modal player modeling through physiological sensors, in-game events and self-report data. Specifically, the present paper (1) discusses design decisions for the personalised crisis management training framework, and (2) presents the game prototype with which user-studies will be performed. Presently, the game prototype is being developed in close collaboration with actual crisis management experts.

In recent years, multiple tabletop games have been released which harness the power of smart phones and tablets to enhance their gameplay. We call these games "augmented tabletop games." Such games have met with a variety of reactions from traditional tabletop gamers, ranging from highly negative to highly positive. The main purpose of this study is to understand the opinions and attitudes of players on the emerging augmented tabletop games. The study also aims to come up with suggestions for game developers by revealing player expectations. A qualitative content analysis was carried out on prevalent tabletop gaming forums. In total, 928 posts on 15 threads were analyzed. From these we derived typologies of negative and positive attitudes towards augmented tabletop games. The overall findings are summarized in a conceptual model. We use this model to discuss design implications for developers of augmented tabletop games.

Domain-Specific Languages (DSL's) offer language-level abstractions that General-Purpose Languages do not offer, thus speeding up the implementation of the solution of problems within a specific domain. Developers have the choice of developing a DSL by building an interpreter/compiler for it, which is a hard and time-consuming task, or embedding it in a host language, thus speeding up the development process but losing several advantages that having a dedicated compiler might bring. In this work we present a meta-compiler called Metacasanova, whose meta-language is based on operational semantics. Then, we propose a language extension with functors and modules that allows to embed the type system of a language definition inside the meta-type system of Metacasanova and improves the performance of manipulating data structures at run-time. Our results show that Metacasanova dramatically reduces the code lines required to develop a compiler, and that the running time of the Meta-program is improved by embedding the host language type system in the meta-type system with the use of functors in the meta-language.

Adaptive behaviour for computer generated forces enriches training simulations with appropriate challenge levels. For adequate insight into the range of possible behaviour, the adaptation has to take place in a rapid fashion. Ideally, each new behaviour model should remain readable by (and thereby under the control of) human experts. Although various attempts have been made at creating adaptive behaviour, current solutions require large numbers of simulations. Moreover, usability by end users has been of subordinate interest, as is compliance with doctrine and ethics. In this work, we present a machine learning method that enables fast behaviour adaptation, while keeping the behaviour models in a human-readable format. We demonstrate the effectiveness of the proposed method in beyond-visual-range air combat simulations.

Machine learning techniques can help to automatically generate behavior for computer generated forces inhabiting air combat training simulations. However, as the complexity of scenarios increases, so does the time to learn optimal behavior. Transfer learning has the potential to significantly shorten the learning time between domains that are sufficiently similar. In this paper, we transfer air combat agents with experience fighting in 2-versus-1 scenarios to various 2-versus-2 scenarios. The performance of the transferred agents is compared to that of agents that learn from scratch in the 2v2 scenarios. The experiments show that the experience gained in the 2v1 scenarios is very beneficial in the plain 2v2 scenarios, where further learning is minimal. In difficult 2v2 scenarios transfer also occurs, and further learning ensues. The results pave the way for fast generation of behavior rules for air combat agents for new, complex scenarios using existing behavior models.

Computer generated forces (CGFs) inhabiting air combat training simulations must show realistic and adaptive behavior to effectively perform their roles as allies and adversaries. In earlier work, behavior for these CGFs was successfully generated using reinforcement learning. However, due to missile hits being subject to chance (a.k.a. The probability of-kill), the CGFs have in certain cases been improperly rewarded and punished. We surmise that taking this probability of-kill into account in the reward function will improve performance. To remedy the false rewards and punishments, a new reward function is proposed that rewards agents based on the expected outcome of their actions. Tests show that the use of this function significantly increases the performance of the CGFs in various scenarios, compared to the previous reward function and a naïve baseline. Based on the results, the new reward function allows the CGFs to generate more intelligent behavior, which enables better training simulations.

Abstract For patients with mental health problems, various treatments exist. Before a treatment is assigned to a patient, a team of clinicians must decide which of the available treatments has the best chance of succeeding. This is a difficult decision to make, as the effectiveness of a treatment might depend on various factors, such as the patient's diagnosis, background and social environment. Which factors are the predictors for successful treatment is mostly unknown. In this article, we present a case‐based reasoning approach for predicting the effect of treatments for patients with anxiety disorders. We investigated which techniques are suitable for implementing such a system to achieve a high level of accuracy. For our evaluation, we used data from a professional mental healthcare centre. Our application correctly predicted the success factor of 65% of the cases, which is significantly higher than the prediction of the baseline of 55%. Under the condition that the prediction was based on only cases with a similarity of at least 0.62, the success rate of 80% of the cases was predicted correctly. These results warrant further development of the system.

In recent decades, video games have come to appeal to people of all ages. The effect of age on how people play games is not fully understood. In this paper, we delve into the question how age relates to an individual's play style. “Play style” is defined as any (set of) patterns in game actions performed by a player. Based on data from 10 416 Battlefield 3 players, we found that age strongly correlates to how people start out playing a game (initial play style), and to how they change their play style over time (play style development). Our data shows three major trends: 1) correlations between age and initial play style peak around the age of 20; 2) performance decreases with age; and 3) speed of play decreases with age. The relationship between age and play style may be explained by the neurocognitive effects of aging: as people grow older, their cognitive performance decays, their personalities shift to a more conscientious style, and their gaming motivations become less achievement-oriented.

Current approaches to adaptive game AI typically require numerous trials to learn effective behavior (i.e., game adaptation is not rapid). In addition, game developers are concerned that applying adaptive game AI may result in uncontrollable and unpredictable behavior (i.e., game adaptation is not reliable). These characteristics hamper the incorporation of adaptive game AI in commercially available video games. In this paper, we discuss an alternative to these current approaches. Our alternative approach to adaptive game AI has as its goal adapting rapidly and reliably to game circumstances. Our approach can be classified in the area of case-based adaptive game AI. In the approach, domain knowledge required to adapt to game circumstances is gathered automatically by the game AI, and is exploited immediately (i.e., without trials and without resource-intensive learning) to evoke effective behavior in a controlled manner in online play. We performed experiments that test case-based adaptive game AI on three different maps in a commercial real-time strategy (RTS) game. From our results, we may conclude that case-based adaptive game AI provides a strong basis for effectively adapting game AI in video games.

Adaptive techniques tend to converge to a single optimum. For adaptive game AI, such convergence is often undesirable, as repetitive game AI is considered to be uninteresting for players. In this paper, we propose a method for automatically learning diverse but effective macros that can be used as components of adaptive game AI scripts. Macros are learned by a cross-entropy method (CEM). This is a selection-based optimization method that, in our experiments, maximizes an interestingness measure. We demonstrate the approach in a computer role-playing game (CRPG) simulation with two duelling wizards, one of which is controlled by an adaptive game AI technique called “dynamic scripting.” Our results show that the macros that we learned manage to increase both adaptivity and diversity of the scripts generated by dynamic scripting, while retaining playing strength.