A5020511097- Artificial Intelligence in Games
- Reinforcement Learning in Robotics
- Evolutionary Algorithms and Applications
- Digital Games and Media
- Modular Robots and Swarm Intelligence
- Video Analysis and Summarization
- Generative Adversarial Networks and Image Synthesis
- Advanced Memory and Neural Computing
- Music Technology and Sound Studies
- Neural Networks and Applications
- Cellular Automata and Applications
- Explainable Artificial Intelligence (XAI)
- Educational Games and Gamification
- Neural dynamics and brain function
- Machine Learning and Data Classification
- Advanced Materials and Mechanics
- Adversarial Robustness in Machine Learning
- Music and Audio Processing
- Human Motion and Animation
- Multimodal Machine Learning Applications
- Topic Modeling
- Sports Analytics and Performance
- Ferroelectric and Negative Capacitance Devices
- Domain Adaptation and Few-Shot Learning
- Micro and Nano Robotics
IT University of Copenhagen
2016-2025
Danish Ministry of Defence
2020-2023
University of Central Florida
2009-2021
Uber AI (United States)
2019
Olgahospital
2015
University of Copenhagen
2014
Cornell University
2013
Philipps University of Marburg
2008
Generative Adversarial Networks (GANs) are a machine learning approach capable of generating novel example outputs across space provided training examples. Procedural Content Generation (PCG) levels for video games could benefit from such models, especially where there is pre-existing corpus to emulate. This paper trains GAN generate Super Mario Bros using level the Video Game Level Corpus. The successfully generates variety similar one in original corpus, but further improved by application...
In this paper, we review recent deep learning advances in the context of how they have been applied to play different types video games such as first-person shooters, arcade games, and real-time strategy games. We analyze unique requirements that game genres pose a system highlight important open challenges applying these machine methods general playing, dealing with extremely large decision spaces sparse rewards.
Growing interest in eXplainable Artificial Intelligence (XAI) aims to make AI and machine learning more understandable human users. However, most existing work focuses on new algorithms, not usability, practical interpretability efficacy real In this vision paper, we propose a research area of for Designers (XAID), specifically game designers. By focusing specific user group, their needs tasks, human-centered approach facilitating designers co-create with AI/ML techniques through XAID. We...
This paper surveys research on applying neuroevolution (NE) to games. In neuroevolution, artificial neural networks are trained through evolutionary algorithms, taking inspiration from the way biological brains evolved. We analyze application of NE in games along five different axes, which role is chosen play a game, types used, these evolved, how fitness determined and what type input network receives. The also highlights important open challenges field.
Deep reinforcement learning (RL) has shown impressive results in a variety of domains, directly from high-dimensional sensory streams. However, when neural networks are trained fixed environment, such as single level video game, they will usually overfit and fail to generalize new levels. When RL models overfit, even slight modifications the environment can result poor agent performance. This paper explores how procedurally generated levels during training increase generality. We show that...
Generative Adversarial Networks (GANs) have shown impressive results for image generation. However, GANs face challenges in generating contents with certain types of constraints, such as game levels. Specifically, it is difficult to generate levels that aesthetic appeal and are playable at the same time. Additionally, because training data usually limited, challenging unique current GANs. In this paper, we propose a new GAN architecture named Conditional Embedding Self-Attention Net-work...
Search-based procedural content generation methods allow video games to introduce new continually, thereby engaging the player for a longer time while reducing burden on developers. However, so far have not explored potential economic value of unique evolved artifacts. Building this insight, paper presents first Facebook game called Petalz in which players can share flowers they breed themselves with other through global marketplace. In particular, market social allows set price their...
The real-time strategy game StarCraft has proven to be a challenging environment for artificial intelligence techniques, and as result, current state-of-the-art solutions consist of numerous hand-crafted modules. In this paper, we show how macromanagement decisions in can learned directly from replays using deep learning. Neural networks are trained on 789,571 state-action pairs extracted 2,005 highly skilled players, achieving top-1 top-3 error rates 54.6% 22.9% predicting the next build...
Generative Adversarial Networks (GANs) are an emerging form of indirect encoding. The GAN is trained to induce a latent space on training data, and real-valued evolutionary algorithm can search that space. Such Latent Variable Evolution (LVE) has recently been applied game levels. However, it hard for objective scores capture level features appealing players. Therefore, this paper introduces tool interactive LVE tile-based levels games. also allows direct exploration the dimensions, users...
A major goal for researchers in neuroevolution is to evolve artificial neural networks (ANNs) that can learn during their lifetime. Such adapt changes environment evolution on its own cannot anticipate. However, a profound problem with evolving adaptive systems if the impact of learning fitness agent only marginal, then likely produce individuals do not exhibit desired behavior. Instead, because it easier at first improve without ability learn, they are exploit domain-dependent static (i.e....
Biological brains can adapt and learn from past experience. Yet neuroevolution, that is, automatically creating artificial neural networks (ANNs) through evolutionary algorithms, has sometimes focused on static ANNs cannot change their weights during lifetime. A profound problem with evolving adaptive systems is learning to highly deceptive. Because it easier at first improve fitness without the ability learn, evolution likely exploit domain-dependent (i.e., nonadaptive) heuristics. This...
Intelligence in nature is the product of living brains, which are themselves natural evolution. Although researchers field neuroevolution (NE) attempt to recapitulate this process, artificial neural networks (ANNs) so far evolved through NE algorithms do not match distinctive capabilities biological brains. The recently introduced hypercube-based augmenting topologies (HyperNEAT) approach narrowed gap by demonstrating that pattern weights across connectivity an ANN can be generated as a...
Besides the life-as-it-could-be driver of artificial life research there is also concept extending natural by creating hybrids or mixed societies that are built from and components. In this paper we motivate present program project flora robotica. Our objective to develop investigate closely linked symbiotic relationships between robots plants explore potentials a plant-robot society able produce architectural artifacts living spaces. These robot-plant bio-hybrids create synergies allow for...
The advent of artificial intelligence (AI) and machine learning (ML) bring human-AI interaction to the forefront HCI research. This paper argues that games are an ideal domain for studying experimenting with how humans interact AI. Through a systematic survey neural network (n = 38), we identified dominant metaphors AI patterns in these games. In addition, applied existing guidelines further shed light on player-AI context AI-infused systems. Our core finding is as play can expand current...
Procedural Content Generation (PCG) is a technique to generate complex and diverse environments in an automated way. However, while generating content with PCG methods often straightforward, meaningful that reflects specific intentions constraints remains challenging. Furthermore, many algorithms lack the ability open-ended manner. Recently, Large Language Models (LLMs) have shown be incredibly effective domains. These trained LLMs can fine-tuned, re-using information accelerating training...
Controlling units in real-time strategy (RTS) games is a challenging problem Artificial Intelligence (AI) as these are fast-paced with simultaneous moves and massive branching factors. This paper presents two extensions to the algorithm UCT Considering Durations (UCTCD) for finding optimal sequences of actions engaged combat using RTS game StarCraft test bed. The first extension uses script-based approach inspired by Portfolio Greedy Search searches scripts instead actions. second...
The impact of game content on the player experience is potentially more critical in casual games than competitive because diminished role strategic or tactical diversions. Interestingly, until now procedural generation (PCG) has nevertheless been investigated almost exclusively context competitive, skills-based gaming. This paper therefore opens a new direction for PCG by placing it at center an entirely flower-breeding platform called Petalz. That way, behavior players and their reactions...