Nico Piatkowski

Towards Intelligent Team Composition and Maneuvering in Real-Time Strategy Games

Players of real-time strategy (RTS) games are often annoyed by the inability of the game AI to select and move teams of units in a natural way. Units travel and battle separately, resulting in huge losses and the AI looking unintelligent, as can the choice of units sent to counteract the opponents. Players are affected as well as computer commanded factions because they cannot micromanage all team related issues. We suggest improving AI behavior by combining well-known computational intelligence techniques applied in an original way. Team composition for battling spatially distributed opponent groups is supported by a learning self-organizing map (SOM) that relies on an evolutionary algorithm (EA) to adapt it to the game. Different abilities of unit types are thus employed in a near-optimal way, reminiscent of human ad hoc decisions. Team movement is greatly enhanced by flocking and influence map-based path finding, leading to a more natural behavior by preserving individual motion types. The team decision to either attack or avoid a group of enemy units is easily parametrizable, incorporating team characteristics from fearful to daredevil. We demonstrate that these two approaches work well separately, but also that they go together naturally, thereby leading to an improved and flexible group behavior.

Dynamic route planning with real-time traffic predictions

Situation aware route planning gathers increasing interest as cities become crowded and jammed. We present a system for individual trip planning that incorporates future traffic hazards in routing. Future traffic conditions are computed by a Spatio-Temporal Random Field based on a stream of sensor readings. In addition, our approach estimates traffic flow in areas with low sensor coverage using a Gaussian Process Regression. The conditioning of spatial regression on intermediate predictions of a discrete probabilistic graphical model allows us to incorporate historical data, streamed online data and a rich dependency structure at the same time. We demonstrate the system with a real-world use-case from Dublin city, Ireland. HighlightsDynamic traffic cost prediction.Situation dependent trip planner.Prediction-as-a-service with TUD streams framework.

Spatio-temporal random fields: compressible representation and distributed estimation

Modern sensing technology allows us enhanced monitoring of dynamic activities in business, traffic, and home, just to name a few. The increasing amount of sensor measurements, however, brings us the challenge for efficient data analysis. This is especially true when sensing targets can interoperate—in such cases we need learning models that can capture the relations of sensors, possibly without collecting or exchanging all data. Generative graphical models namely the Markov random fields (MRF) fit this purpose, which can represent complex spatial and temporal relations among sensors, producing interpretable answers in terms of probability. The only drawback will be the cost for inference, storing and optimizing a very large number of parameters—not uncommon when we apply them for real-world applications.In this paper, we investigate how we can make discrete probabilistic graphical models practical for predicting sensor states in a spatio-temporal setting. A set of new ideas allows keeping the advantages of such models while achieving scalability. We first introduce a novel alternative to represent model parameters, which enables us to compress the parameter storage by removing uninformative parameters in a systematic way. For finding the best parameters via maximum likelihood estimation, we provide a separable optimization algorithm that can be performed independently in parallel in each graph node. We illustrate that the prediction quality of our suggested method is comparable to those of the standard MRF and a spatio-temporal k-nearest neighbor method, while using much less computational resources.

Intelligent anti-grouping in real-time strategy games

Assembling suitable groups of fighting units to combat incoming enemy groups is a tactical necessity in real-time strategy (RTS) games. Furthermore it heavily influences future strategic decisions like unit building. Here, we demonstrate how to efficiently (offline) solve the problem of finding matches for the current enemy group(s) based on self-organizing maps (SOMs), powered by a simple evolutionary algorithm. The concept is implemented and thoroughly experimentally investigated in the RTS game Glest. We show that the offline learning is reliable and can be sped up considerably by employing a very simple substitute objective function instead of game simulations, making it a nearly universal, simple, and transparent technique.

Parallel Inference on Structured Data with CRFs on GPUs

Structured real world data can be represented with graphs whose structure encodes independence assumptions within the data. Due to statistical advantages over generative graphical models, Conditional Random Fields (CRFs) are used in a wide range of classification tasks on structured data sets. CRFs can be learned from both, fully or partially supervised data, and may be used to infer fully unlabeled or partially labelled data. However, performing inference in CRFs with an arbitrary graphical structure on a large amount of data is computational expensive and nearly intractable on a reseacher’s workstation. Hence, we take advantage of recent developments in computer hardware, namely generalpurpose Graphics Processing Units (GPUs). We not merely run given algorithms on GPUs, but present a novel framework of parallel algorithms at several levels for training general CRFs on very large data sets. We evaluate their performance in terms of runtime and F1-Score.

Heterogeneous Stream Processing and Crowdsourcing for Traffic Monitoring: Highlights

We give an overview of an intelligent urban traffic management system. Complex events related to congestions are detected from heterogeneous sources involving fixed sensors mounted on intersections and mobile sensors mounted on public transport vehicles. To deal with data veracity, sensor disagreements are resolved by crowdsourcing. To deal with data sparsity, a traffic model offers information in areas with low sensor coverage. We apply the system to a real-world use-case.

To model or not to model: Controlling Pac-Man ghosts without incorporating global knowledge

The creation of interesting opponents for human players in computer games is an interesting and challenging task. In contrast to up-to-date computer games, e.g. real time strategy games, learning of non-player-character strategies for older games seems to be easier and not that time-consuming. This way, older games, like the famous arcade game Pac-Man, serve as a test bed for the creation of strategies that are fun to play against. The paper at hand uses computational intelligence methods to accomplish this challenge, namely evolutionary algorithms (EA) and artificial neural networks (ANN). The latter are trained on a model of the game whereas the EA learn good behavior by playing. The performance of these two approaches is compared on the original Pac-Man level as well as on other maps with different properties to test the ability of generalizing the learned strategies.

Integer undirected graphical models for resource-constrained systems

Abstract Machine learning on resource-constrained ubiquitous devices suffers from high energy consumption and slow execution. The number of clock cycles that is consumed by arithmetic instructions has an immediate impact on both. In computer systems, the number of consumed cycles depends on particular operations and the types of their operands. We propose a new class of probabilistic graphical models that approximates the full joint probability distribution of discrete multivariate random variables by relying only on integer addition/multiplication and binary bit shift operations. This allows us to sample from high-dimensional generative models and to use structured discriminative classifiers even on computational devices with slow floating point units or in situations where energy has to be saved. While theory and experiments on random synthetic data suggest that hard instances (leading to a large approximation error) exist, experiments on benchmark and real-world data show that the integer models achieve qualitatively the same results as their double-precision counterparts. Moreover, clock cycle consumption on two hardware platforms is regarded, where our results show that resource savings due to integer approximation is even larger on low-end hardware. The integer models consume half of the clock cycles and a small fraction of memory compared to ordinary undirected graphical models.

Towards adjusting mobile devices to user's behaviour

Mobile devices are a special class of resource-constrained embedded devices. Computing power, memory, the available energy, and network bandwidth are often severely limited. These constrained resources require extensive optimization of a mobile system compared to larger systems. Any needless operation has to be avoided. Timeconsuming operations have to be started early on. For instance, loading files ideally starts before the user wants to access the file. So-called prefetching strategies optimize systems operation. Our goal is to adjust such strategies on the basis of logged system data. Optimization is then achieved by predicting an applications behavior based on facts learned from earlier runs on the same system. In this paper, we analyze system-calls on operating system level and compare two paradigms, namely server-based and device-based learning. The results could be used to optimize the runtime behaviour of mobile devices.

Parallel Algorithms for GPU accelerated Probabilistic Inference

Real world data is likely to contain an inherent structure. Those structures may be represented with graphs which encode independence assumptions within the data. Performing inference in those models is nearly intractable on mobile devices or casual workstations. This work introduces and compares two approaches for accelerating the inference in graphical models by using GPUs as parallel processing units. It is empirically showed, that in order to achieve a scaleable parallel algorithm, one has to distribute the workload equally among all processing units of a GPU. We accomplished this by introducing Thread-Cooperative message computations.