Refine
Departments, institutes and facilities
- Fachbereich Informatik (45)
- Institut für Technik, Ressourcenschonung und Energieeffizienz (TREE) (37)
- Fachbereich Ingenieurwissenschaften und Kommunikation (6)
- Präsidium (4)
- Stabsstelle Kommunikation und Marketing (4)
- Graduierteninstitut (2)
- Internationales Zentrum für Nachhaltige Entwicklung (IZNE) (2)
- Institut für KI und Autonome Systeme (A2S) (1)
- Institut für Sicherheitsforschung (ISF) (1)
- Institute of Visual Computing (IVC) (1)
Document Type
- Conference Object (36)
- Article (11)
- Preprint (5)
- Part of Periodical (4)
- Doctoral Thesis (2)
- Report (2)
- Part of a Book (1)
- Master's Thesis (1)
Year of publication
Language
- English (62) (remove)
Keywords
- Quality diversity (6)
- Quality Diversity (4)
- Aerodynamics (3)
- Bayesian optimization (3)
- MAP-Elites (3)
- Surrogate Modeling (3)
- Autoencoder (2)
- Evolutionary Computation (2)
- Evolutionary computation (2)
- Evolutionary optimization (2)
Machine learning-based solutions are frequently adapted in several applications that require big data in operations. The performance of a model that is deployed into operations is subject to degradation due to unanticipated changes in the flow of input data. Hence, monitoring data drift becomes essential to maintain the model’s desired performance. Based on the conducted review of the literature on drift detection, statistical hypothesis testing enables to investigate whether incoming data is drifting from training data. Because Maximum Mean Discrepancy (MMD) and Kolmogorov-Smirnov (KS) have shown to be reliable distance measures between multivariate distributions in the literature review, both were selected from several existing techniques for experimentation. For the scope of this work, the image classification use case was experimented with using the Stream-51 dataset. Based on the results from different drift experiments, both MMD and KS showed high Area Under Curve values. However, KS exhibited faster performance than MMD with fewer false positives. Furthermore, the results showed that using the pre-trained ResNet-18 for feature extraction maintained the high performance of the experimented drift detectors. Furthermore, the results showed that the performance of the drift detectors highly depends on the sample sizes of the reference (training) data and the test data that flow into the pipeline’s monitor. Finally, the results also showed that if the test data is a mixture of drifting and non-drifting data, the performance of the drift detectors does not depend on how the drifting data are scattered with the non-drifting ones, but rather their amount in the test set
Quality diversity algorithms can be used to efficiently create a diverse set of solutions to inform engineers' intuition. But quality diversity is not efficient in very expensive problems, needing 100.000s of evaluations. Even with the assistance of surrogate models, quality diversity needs 100s or even 1000s of evaluations, which can make it use infeasible. In this study we try to tackle this problem by using a pre-optimization strategy on a lower-dimensional optimization problem and then map the solutions to a higher-dimensional case. For a use case to design buildings that minimize wind nuisance, we show that we can predict flow features around 3D buildings from 2D flow features around building footprints. For a diverse set of building designs, by sampling the space of 2D footprints with a quality diversity algorithm, a predictive model can be trained that is more accurate than when trained on a set of footprints that were selected with a space-filling algorithm like the Sobol sequence. Simulating only 16 buildings in 3D, a set of 1024 building designs with low predicted wind nuisance is created. We show that we can produce better machine learning models by producing training data with quality diversity instead of using common sampling techniques. The method can bootstrap generative design in a computationally expensive 3D domain and allow engineers to sweep the design space, understanding wind nuisance in early design phases.
This paper explores the role of artificial intelligence (AI) in elite sports. We approach the topic from two perspectives. Firstly, we provide a literature based overview of AI success stories in areas other than sports. We identified multiple approaches in the area of Machine Perception, Machine Learning and Modeling, Planning and Optimization as well as Interaction and Intervention, holding a potential for improving training and competition. Secondly, we discover the present status of AI use in elite sports. Therefore, in addition to another literature review, we interviewed leading sports scientist, which are closely connected to the main national service institute for elite sports in their countries. The analysis of this literature review and the interviews show that the most activity is carried out in the methodical categories of signal and image processing. However, projects in the field of modeling & planning have become increasingly popular within the last years. Based on these two perspectives, we extract deficits, issues and opportunities and summarize them in six key challenges faced by the sports analytics community. These challenges include data collection, controllability of an AI by the practitioners and explainability of AI results.
We consider multi-solution optimization and generative models for the generation of diverse artifacts and the discovery of novel solutions. In cases where the domain's factors of variation are unknown or too complex to encode manually, generative models can provide a learned latent space to approximate these factors. When used as a search space, however, the range and diversity of possible outputs are limited to the expressivity and generative capabilities of the learned model. We compare the output diversity of a quality diversity evolutionary search performed in two different search spaces: 1) a predefined parameterized space and 2) the latent space of a variational autoencoder model. We find that the search on an explicit parametric encoding creates more diverse artifact sets than searching the latent space. A learned model is better at interpolating between known data points than at extrapolating or expanding towards unseen examples. We recommend using a generative model's latent space primarily to measure similarity between artifacts rather than for search and generation. Whenever a parametric encoding is obtainable, it should be preferred over a learned representation as it produces a higher diversity of solutions.
In this thesis it is posed that the central object of preference discovery is a co-creative process in which the Other can be represented by a machine. It explores efficient methods to enhance introverted intuition using extraverted intuition's communication lines. Possible implementations of such processes are presented using novel algorithms that perform divergent search to feed the users' intuition with many examples of high quality solutions, allowing them to take influence interactively. The machine feeds and reflects upon human intuition, combining both what is possible and preferred. The machine model and the divergent optimization algorithms are the motor behind this co-creative process, in which machine and users co-create and interactively choose branches of an ad hoc hierarchical decomposition of the solution space.
The proposed co-creative process consists of several elements: a formal model for interactive co-creative processes, evolutionary divergent search, diversity and similarity, data-driven methods to discover diversity, limitations of artificial creative agents, matters of efficiency in behavioral and morphological modeling, visualization, a connection to prototype theory, and methods to allow users to influence artificial creative agents. This thesis helps putting the human back into the design loop in generative AI and optimization.
Computers can help us to trigger our intuition about how to solve a problem. But how does a computer take into account what a user wants and update these triggers? User preferences are hard to model as they are by nature vague, depend on the user’s background and are not always deterministic, changing depending on the context and process under which they were established. We pose that the process of preference discovery should be the object of interest in computer aided design or ideation. The process should be transparent, informative, interactive and intuitive. We formulate Hyper-Pref, a cyclic co-creative process between human and computer, which triggers the user’s intuition about what is possible and is updated according to what the user wants based on their decisions. We combine quality diversity algorithms, a divergent optimization method that can produce many, diverse solutions, with variational autoencoders to both model that diversity as well as the user’s preferences, discovering the preference hypervolume within large search spaces.
In complex, expensive optimization domains we often narrowly focus on finding high performing solutions, instead of expanding our understanding of the domain itself. But what if we could quickly understand the complex behaviors that can emerge in said domains instead? We introduce surrogate-assisted phenotypic niching, a quality diversity algorithm which allows to discover a large, diverse set of behaviors by using computationally expensive phenotypic features. In this work we discover the types of air flow in a 2D fluid dynamics optimization problem. A fast GPU-based fluid dynamics solver is used in conjunction with surrogate models to accurately predict fluid characteristics from the shapes that produce the air flow. We show that these features can be modeled in a data-driven way while sampling to improve performance, rather than explicitly sampling to improve feature models. Our method can reduce the need to run an infeasibly large set of simulations while still being able to design a large diversity of air flows and the shapes that cause them. Discovering diversity of behaviors helps engineers to better understand expensive domains and their solutions.
The encoding of solutions in black-box optimization is a delicate, handcrafted balance between expressiveness and domain knowledge between exploring a wide variety of solutions, and ensuring that those solutions are useful. Our main insight is that this process can be automated by generating a dataset of high-performing solutions with a quality diversity algorithm (here, MAP-Elites), then learning a representation with a generative model (here, a Varia-tional Autoencoder) from that dataset. Our second insight is that this representation can be used to scale quality diversity optimization to higher dimensions-but only if we carefully mix solutions generated with the learned representation and those generated with traditional variation operators. We demonstrate these capabilities by learning an low-dimensional encoding for the inverse kinemat-ics of a thousand joint planar arm. The results show that learned representations make it possible to solve high-dimensional problems with orders of magnitude fewer evaluations than the standard MAP-Elites, and that, once solved, the produced encoding can be used for rapid optimization of novel, but similar, tasks. The presented techniques not only scale up quality diversity algorithms to high dimensions, but show that black-box optimization encodings can be automatically learned, rather than hand designed.
Optimization plays an essential role in industrial design, but is not limited to minimization of a simple function, such as cost or strength. These tools are also used in conceptual phases, to better understand what is possible. To support this exploration we focus on Quality Diversity (QD) algorithms, which produce sets of varied, high performing solutions. These techniques often require the evaluation of millions of solutions -- making them impractical in design cases. In this thesis we propose methods to radically improve the data-efficiency of QD with machine learning, enabling its application to design. In our first contribution, we develop a method of modeling the performance of evolved neural networks used for control and design. The structures of these networks grow and change, making them difficult to model -- but with a new method we are able to estimate their performance based on their heredity, improving data-efficiency by several times. In our second contribution we combine model-based optimization with MAP-Elites, a QD algorithm. A model of performance is created from known designs, and MAP-Elites creates a new set of designs using this approximation. A subset of these designs are the evaluated to improve the model, and the process repeats. We show that this approach improves the efficiency of MAP-Elites by orders of magnitude. Our third contribution integrates generative models into MAP-Elites to learn domain specific encodings. A variational autoencoder is trained on the solutions produced by MAP-Elites, capturing the common “recipe” for high performance. This learned encoding can then be reused by other algorithms for rapid optimization, including MAP-Elites. Throughout this thesis, though the focus of our vision is design, we examine applications in other fields, such as robotics. These advances are not exclusive to design, but serve as foundational work on the integration of QD and machine learning.
The way solutions are represented, or encoded, is usually the result of domain knowledge and experience. In this work, we combine MAP-Elites with Variational Autoencoders to learn a Data-Driven Encoding (DDE) that captures the essence of the highest-performing solutions while still able to encode a wide array of solutions. Our approach learns this data-driven encoding during optimization by balancing between exploiting the DDE to generalize the knowledge contained in the current archive of elites and exploring new representations that are not yet captured by the DDE. Learning representation during optimization allows the algorithm to solve high-dimensional problems, and provides a low-dimensional representation which can be then be re-used. We evaluate the DDE approach by evolving solutions for inverse kinematics of a planar arm (200 joint angles) and for gaits of a 6-legged robot in action space (a sequence of 60 positions for each of the 12 joints). We show that the DDE approach not only accelerates and improves optimization, but produces a powerful encoding that captures a bias for high performance while expressing a variety of solutions.
In optimization methods that return diverse solution sets, three interpretations of diversity can be distinguished: multi-objective optimization which searches diversity in objective space, multimodal optimization which tries spreading out the solutions in genetic space, and quality diversity which performs diversity maintenance in phenotypic space. We introduce niching methods that provide more flexibility to the analysis of diversity and a simple domain to compare and provide insights about the paradigms. We show that multiobjective optimization does not always produce much diversity, quality diversity is not sensitive to genetic neutrality and creates the most diverse set of solutions, and multimodal optimization produces higher fitness solutions. An autoencoder is used to discover phenotypic features automatically, producing an even more diverse solution set. Finally, we make recommendations about when to use which approach.
Change - shaping reality
(2019)
In mathematical modeling by means of performance models, the Fitness-Fatigue Model (FF-Model) is a common approach in sport and exercise science to study the training performance relationship. The FF-Model uses an initial basic level of performance and two antagonistic terms (for fitness and fatigue). By model calibration, parameters are adapted to the subject’s individual physical response to training load. Although the simulation of the recorded training data in most cases shows useful results when the model is calibrated and all parameters are adjusted, this method has two major difficulties. First, a fitted value as basic performance will usually be too high. Second, without modification, the model cannot be simply used for prediction. By rewriting the FF-Model such that effects of former training history can be analyzed separately – we call those terms preload – it is possible to close the gap between a more realistic initial performance level and an athlete's actual performance level without distorting other model parameters and increase model accuracy substantially. Fitting error of the preload-extended FF-Model is less than 32% compared to the error of the FF-Model without preloads. Prediction error of the preload-extended FF-Model is around 54% of the error of the FF-Model without preloads.
Are quality diversity algorithms better at generating stepping stones than objective-based search?
(2019)
The route to the solution of complex design problems often lies through intermediate "stepping stones" which bear little resemblance to the final solution. By greedily following the path of greatest fitness improvement, objective-based search overlooks and discards stepping stones which might be critical to solving the problem. Here, we hypothesize that Quality Diversity (QD) algorithms are a better way to generate stepping stones than objective-based search: by maintaining a large set of solutions which are of high-quality, but phenotypically different, these algorithms collect promising stepping stones while protecting them in their own "ecological niche". To demonstrate the capabilities of QD we revisit the challenge of recreating images produced by user-driven evolution, a classic challenge which spurred work in novelty search and illustrated the limits of objective-based search. We show that QD far outperforms objective-based search in matching user-evolved images. Further, our results suggest some intriguing possibilities for leveraging the diversity of solutions created by QD.
The initial phase in real world engineering optimization and design is a process of discovery in which not all requirements can be made in advance, or are hard to formalize. Quality diversity algorithms, which produce a variety of high performing solutions, provide a unique chance to support engineers and designers in the search for what is possible and high performing. In this work we begin to answer the question how a user can interact with quality diversity and turn it into an interactive innovation aid. By modeling a user's selection it can be determined whether the optimization is drifting away from the user's preferences. The optimization is then constrained by adding a penalty to the objective function. We present an interactive quality diversity algorithm that can take into account the user's selection. The approach is evaluated in a new multimodal optimization benchmark that allows various optimization tasks to be performed. The user selection drift of the approach is compared to a state of the art alternative on both a planning and a neuroevolution control task, thereby showing its limits and possibilities.
This work addresses the issue of finding an optimal flight zone for a side-by-side tracking and following Unmanned Aerial Vehicle(UAV) adhering to space-restricting factors brought upon by a dynamic Vector Field Extraction (VFE) algorithm. The VFE algorithm demands a relatively perpendicular field of view of the UAV to the tracked vehicle, thereby enforcing the space-restricting factors which are distance, angle and altitude. The objective of the UAV is to perform side-by-side tracking and following of a lightweight ground vehicle while acquiring high quality video of tufts attached to the side of the tracked vehicle. The recorded video is supplied to the VFE algorithm that produces the positions and deformations of the tufts over time as they interact with the surrounding air, resulting in an airflow model of the tracked vehicle. The present limitations of wind tunnel tests and computational fluid dynamics simulation suggest the use of a UAV for real world evaluation of the aerodynamic properties of the vehicle’s exterior. The novelty of the proposed approach is alluded to defining the specific flight zone restricting factors while adhering to the VFE algorithm, where as a result we were capable of formalizing a locally-static and a globally-dynamic geofence attached to the tracked vehicle and enclosing the UAV.