Mask-based explanation methods offer a powerful framework for interpreting deep learning model predictions across diverse data modalities, such as images and time series, in which the central idea is to identify an instance-dependent mask that minimizes the performance drop from the resulting masked input. Different objectives for learning such masks have been proposed, all of which, in our view, can be unified under an information-theoretic framework that balances performance degradation of the masked input with the complexity of the resulting masked representation. Typically, these methods initialize the masks either uniformly or as all-ones. In this paper, we argue that an effective mask initialization strategy is as important as the development of novel learning objectives, particularly in light of the significant computational costs associated with existing mask-based explanation methods. To this end, we introduce a new gradient-based initialization technique called StartGrad, which is the first initialization method specifically designed for mask-based post-hoc explainability methods. Compared to commonly used strategies, StartGrad is provably superior at initialization in striking the aforementioned trade-off. Despite its simplicity, our experiments demonstrate that StartGrad enhances the optimization process of various state-of-the-art mask-explanation methods by reaching target metrics faster and, in some cases, boosting their overall performance.

Machine learning models predicting neurodevelopmental outcome in preterm infants have great potential, but have often relied on inaccessible brain magnetic resonance imaging measurements. This study aimed to build models using readily available clinical predictors and investigate the potential of vital sign data in the prediction of neurodevelopmental outcome after preterm birth. Readily available predictors from the antenatal and neonatal period of preterm infants born <30 weeks gestation were combined with vital sign data from the first seven days after birth to predict motor and cognitive outcome at two and five years corrected age. A conventional logistic regression model was compared with a support vector machine and a neural network. Vital sign times series were investigated using two approaches; basic descriptives of vital sign data were compared to an advanced approach in which vital sign time series were processed in an auto-encoder and long-short-term-memory network. Best performing models reached moderate area under the receiver operating characteristic curves (0.592 to 0.703), yet reaching high negative predictive values (85% to 94% ). Vital sign data did modestly improve prediction of motor outcome, but not prediction of cognitive outcome. Advanced handling of vital sign time series did not improve prediction above basic descriptives of vital signs. Neurodevelopmental outcome prediction on routine clinical data remains challenging, but also shows potential in the identification of infants with low risk of adverse outcome. Future work may take advantage of higher resolution and a wider variety of vital signs.

Reinforcement Learning (RL) has recently found many applications in the healthcare domain thanks to its natural fit to clinical decision-making and ability to learn optimal decisions from observational data. A key challenge in adopting RL-based solution in clinical practice, however, is the inclusion of existing knowledge in learning a suitable solution. Existing knowledge from e.g. medical guidelines may improve the safety of solutions, produce a better balance between short- and long-term outcomes for patients and increase trust and adoption by clinicians. We present a framework for including knowledge available from medical guidelines in RL. The framework includes components for enforcing safety constraints and an approach that alters the learning signal to better balance short- and long-term outcomes based on these guidelines. We evaluate the framework by extending an existing RL-based mechanical ventilation (MV) approach with clinically established ventilation guidelines. Results from off-policy policy evaluation indicate that our approach has the potential to decrease 90-day mortality while ensuring lung protective ventilation. This framework provides an important stepping stone towards implementations of RL in clinical practice and opens up several avenues for further research.

Coping with distributional shifts is an important part of transfer learning methods in order to perform well in real-life tasks. However, most of the existing approaches in this area either focus on an ideal scenario in which the data does not contain noises or employ a complicated training paradigm or model design to deal with distributional shifts. In this paper, we revisit the robustness of the minimum error entropy (MEE) criterion, a widely used objective in statistical signal processing to deal with non-Gaussian noises, and investigate its feasibility and usefulness in real-life transfer learning regression tasks, where distributional shifts are common. Specifically, we put forward a new theoretical result showing the robustness of MEE against covariate shift. We also show that by simply replacing the mean squared error (MSE) loss with the MEE on basic transfer learning algorithms such as fine-tuning and linear probing, we can achieve competitive performance with respect to state-of-the-art transfer learning algorithms. We justify our arguments on both synthetic data and 5 real-world time-series data.

As machine and deep learning solutions are deployed on edge devices to tackle real-world classification problems, the approach of early classification during inference is becoming increasingly popular. This approach entails performing the classification after having observed only part of the input and it is motivated by goals such as faster results, computation and communication reduction, and energy conservation, especially in resource-constrained edge intelligence environments. Early classification in the time series analysis domain has been extensively studied. However, when it comes to multivariate time series problems, current solutions only consider early exiting across the temporal dimension, treating the multiple input channels as a singular entity. In this work, we propose a flexible early-exit framework, which can consider input channels in a more fine-grained manner and exit at different points for different channel groups. We implement this framework using reinforcement learning methods and we set heuristics to make our objective tractable while maintaining its practicality. We verify the expected behavior of our framework on synthetic data and evaluate its performance on 26 real datasets. Our extensive experiments demonstrate that, depending on the use case, it can add value to the early classification paradigm, achieving better accuracy for equal input savings.

Nowadays, with the rising number of sensor signals in sectors such as healthcare and industry, the problem of multivariate time series classification (MTSC) is getting increasingly relevant and is a prime target for machine and deep learning approaches. Their expanding adoption in real-world environments is causing a shift in focus from the pursuit of everhigher prediction accuracy with complex models towards practical, deployable solutions that balance accuracy and parameters such as prediction speed. An MTSC model that has attracted attention recently is ROCKET, based on random convolutional kernels, both because of its very fast training process and its state-of-the-art accuracy. However, the large number of features it utilizes may be detrimental to inference time. Examining its theoretical background and limitations enables us to address potential drawbacks and present LightWaveS: a framework for accurate MTSC, which is fast both during training and inference. We show that LightWaveS achieves accuracy comparable to recent MTSC models and speedup ranging from 9x to 53x compared to ROCKET during inference on an edge device, on datasets with comparable accuracy.

In current research, machine and deep learning solutions for the classification of temporal data are shifting from single-channel datasets (univariate) to problems with multiple channels of information (multivariate). The majority of these works are focused on the method novelty and architecture, and the format of the input data is often treated implicitly. Particularly, multivariate datasets are often treated as a stack of univariate time series in terms of input preprocessing, with scaling methods applied across each channel separately. In this evaluation, we aim to demonstrate that the additional channel dimension is far from trivial and different approaches to scaling can lead to significantly different results in the accuracy of a solution. To that end, we test seven different data transformation methods on four different temporal dimensions and study their effect on the classification accuracy of five recent methods. We show that, for the large majority of tested datasets, the best transformation-dimension configuration leads to an increase in the accuracy compared to the result of each model with the same hyperparameters and no scaling, ranging from 0.16 to 76.79 percentage points. We also show that if we keep the transformation method constant, there is a statistically significant difference in accuracy results when applying it across different dimensions, with accuracy differences ranging from 0.23 to 47.79 percentage points. Finally, we explore the relation of the transformation methods and dimensions to the classifiers, and we conclude that there is no prominent general trend, and the optimal configuration is dataset- and classifier-specific.

Reinforcement learning (RL) is a powerful framework for learning complex behaviors, but lacks adoption in many settings due to sample size requirements. We introduce a framework for increasing sample efficiency of RL algorithms. Our approach focuses on optimizing environment rewards with high-level instructions. These are modeled as a high-level controller over temporally extended actions known as options. These options can be looped, interleaved and partially ordered with a rich language for high-level instructions. Crucially, the instructions may be underspecified in the sense that following them does not guarantee high reward in the environment. We present an algorithm for control with these so-called option machines (OMs), discuss option selection for the partially ordered case and describe an algorithm for learning with OMs. We compare our approach in zero-shot, single- and multi-task settings in an environment with fully specified and underspecified instructions. We find that OMs perform significantly better than or comparable to the state-of-art in all environments and learning settings.

There are many e-Health mobile apps on the apps store, from apps to improve a user’s lifestyle to mental coaching. Whilst these apps might consider user context when they give their interventions, prompts, and encouragements, they still tend to be rigid e.g., not using user context and experience to tailor themselves to the user. To better engage and tailor to the user, we have previously proposed a Reference Architecture for enabling self-adaptation and AI personalization in e-Health mobile apps. In this work we evaluate the end users’ perception, usability, performance impact, and energy consumption contributed by this Reference Architecture. We do so by implementing a Reference Architecture compliant app and conducting two experiments: a user study and a measurement-based experiment. Although limited in the number of participants, the results of our user study show that usability of the Reference Architecture compliant app is similar to the control app. Users’ perception was found to be positively influenced by the compliant app when compared to the control group. Results of our measurement-based experiment showed some differences in performance and energy consumption measurements between the two apps. The differences are, however, deemed minimal. Our experiments show promising results for an app implemented following our proposed Reference Architecture. This is preliminary evidence that the use of personalization and self-adaptation techniques can be beneficial within the domain of e-Health apps.

Abstract Deep reinforcement learning (DRL) has shown remarkable success in artificial domains and in some real-world applications. However, substantial challenges remain such as learning efficiently under safety constraints. Adherence to safety constraints is a hard requirement in many high-impact application domains such as healthcare and finance. These constraints are preferably represented symbolically to ensure clear semantics at a suitable level of abstraction. Existing approaches to safe DRL assume that being unsafe leads to low rewards. We show that this is a special case of symbolically constrained RL and analyze a generic setting in which total reward and being safe may or may not be correlated. We analyze the impact of symbolic constraints and identify a connection between expected future reward and distance towards a goal in an automaton representation of the constraints. We use this connection in an algorithm for learning complex behaviors safely and efficiently. This algorithm relies on symbolic reasoning over safety constraints to improve the efficiency of a subsymbolic learner with a symbolically obtained measure of progress. We measure sample efficiency on a grid world and a conversational product recommender with real-world constraints. The so-called Planning for Potential algorithm converges quickly and significantly outperforms all baselines. Specifically, we find that symbolic reasoning is necessary for safety during and after learning and can be effectively used to guide a neural learner towards promising areas of the solution space. We conclude that RL can be applied both safely and efficiently when combined with symbolic reasoning.

INTRODUCTION: In recent years, reinforcement learning (RL) has gained traction in the healthcare domain. In particular, RL methods have been explored for haemodynamic optimization of septic patients in the Intensive Care Unit. Most hospitals however, lack the data and expertise for model development, necessitating transfer of models developed using external datasets. This approach assumes model generalizability across different patient populations, the validity of which has not previously been tested. In addition, there is limited knowledge on safety and reliability. These challenges need to be addressed to further facilitate implementation of RL models in clinical practice. METHOD: We developed and validated a new reinforcement learning model for hemodynamic optimization in sepsis on the MIMIC intensive care database from the USA using a dueling double deep Q network. We then transferred this model to the European AmsterdamUMCdb intensive care database. T-Distributed Stochastic Neighbor Embedding and Sequential Organ Failure Assessment scores were used to explore the differences between the patient populations. We apply off-policy policy evaluation methods to quantify model performance. In addition, we introduce and apply a novel deep policy inspection to analyse how the optimal policy relates to the different phases of sepsis and sepsis treatment to provide interpretable insight in order to assess model safety and reliability. RESULTS: The off-policy evaluation revealed that the optimal policy outperformed the physician policy on both datasets despite marked differences between the two patient populations and physician's policies. Our novel deep policy inspection method showed insightful results and unveiled that the model could initiate therapy adequately and adjust therapy intensity to illness severity and disease progression which indicated safe and reliable model behaviour. Compared to current physician behavior, the developed policy prefers a more liberal use of vasopressors with a more restrained use of fluid therapy in line with previous work. CONCLUSION: We created a reinforcement learning model for optimal bedside hemodynamic management and demonstrated model transferability between populations from the USA and Europe for the first time. We proposed new methods for deep policy inspection integrating expert domain knowledge. This is expected to facilitate progression to bedside clinical decision support for the treatment of critically ill patients.

Energy disaggregation, a.k.a. Non-Intrusive Load Monitoring, aims to separate the energy consumption of individual appliances from the readings of a mains power meter measuring the total energy consumption of, e.g., a whole house. Energy consumption of individual appliances can be useful in many applications, e.g., providing appliance-level feedback to the end users to help them understand their energy consumption and ultimately save energy. Recently, with the availability of large-scale energy consumption datasets, various neural network models such as convolutional neural networks and recurrent neural networks have been investigated to solve the energy disaggregation problem. Neural network models can learn complex patterns from large amounts of data and have been shown to outperform the traditional machine learning methods such as variants of hidden Markov models. However, current neural network methods for energy disaggregation are either computational expensive or are not capable of handling long-term dependencies. In this article, we investigate the application of the recently developed WaveNet models for the task of energy disaggregation. Based on a real-world energy dataset collected from 20 households over 2 years, we show that WaveNet models outperforms the state-of-the-art deep learning methods proposed in the literature for energy disaggregation in terms of both error measures and computational cost. On the basis of energy disaggregation, we then investigate the performance of two deep-learning based frameworks for the task of on/off detection which aims at estimating whether an appliance is in operation or not. The first framework obtains the on/off states of an appliance by binarising the predictions of a regression model trained for energy disaggregation, while the second framework obtains the on/off states of an appliance by directly training a binary classifier with binarised energy readings of the appliance serving as the target values. Based on the same dataset, we show that for the task of on/off detection the second framework, i.e., directly training a binary classifier, achieves better performance in terms of F1 score.

Although group convolutional networks are able to learn powerful representations based on symmetry patterns, they lack explicit means to learn meaningful relationships among them (e.g., relative positions and poses). In this paper, we present attentive group equivariant convolutions, a generalization of the group convolution, in which attention is applied during the course of convolution to accentuate meaningful symmetry combinations and suppress non-plausible, misleading ones. We indicate that prior work on visual attention can be described as special cases of our proposed framework and show empirically that our attentive group equivariant convolutional networks consistently outperform conventional group convolutional networks on benchmark image datasets. Simultaneously, we provide interpretability to the learned concepts through the visualization of equivariant attention maps.

Grouping patients meaningfully can give insights about the different types of patients, their needs, and the priorities. Finding groups that are meaningful is however very challenging as background knowledge is often required to determine what a useful grouping is. In this paper we propose an approach that is able to find groups of patients based on a small sample set of positive examples given by domain experts. Because of that, the approach relies on very limited efforts by the domain experts. The approach groups based on the activities and diagnostic/billing codes within health pathways of patients. To define such a grouping based on the sample of patients efficiently, frequent patterns of activities are discovered and used to measure the similarity between the care pathways of other patients to the patients in the sample group. This approach results in an insightful definition of the group. The proposed approach is evaluated using several datasets obtained from a large university medical center. The evaluation shows F1-scores of around 0.7 for grouping kidney injury and around 0.6 for diabetes.

Clustering of users underlies many of the personalisation algorithms that are in use nowadays. Such clustering is mostly performed in an offline fashion. For a health and wellbeing setting, offline clustering might however not be suitable, as limited data is often available and patient states can also quickly evolve over time. Existing online clustering algorithms are not suitable for the health domain due to the type of data that involves multiple time series evolving over time. In this paper we propose a new online clustering algorithm called CluStream-GT that is suitable for health applications. By using both artificial and real datasets, we show that the approach is far more efficient compared to regular clustering, with an average speedup of 93%, while only losing 12% in the accuracy of the clustering with artificial data and 3% with real data.

We introduce an end-to-end reinforcement learning (RL) solution for the problem of sending personalized digital health interventions. Previous work has shown that personalized interventions can be obtained through RL using simple, discrete state information such as the recent activity performed. In reality however, such features are often not observed, but instead could be inferred from noisy, low-level sensor information obtained from mobile devices (e.g. accelerometers in mobile phones). One could first transform such raw data into discrete activities, but that could throw away important details and would require training a classifier to infer these discrete activities which would need a labeled training set. Instead, we propose to directly learn intervention strategies for the low-level sensor data end-to-end using deep neural networks and RL. We test our novel approach in a self-developed simulation environment which models, and generates, realistic sensor data for daily human activities and show the short-and long-term efficacy of sending personalized physical workout interventions using RL policies. We compare several different input representations and show that learning using raw sensor data is nearly as effective and much more flexible.

Language systems have been of great interest to the research community and have recently reached the mass market through various assistant platforms on the web. Reinforcement Learning methods that optimize dialogue policies have seen successes in past years and have recently been extended into methods that personalize the dialogue, e.g. take the personal context of users into account. These works, however, are limited to personalization to a single user with whom they require multiple interactions and do not generalize the usage of context across users. This work introduces a problem where a generalized usage of context is relevant and proposes two Reinforcement Learning (RL)-based approaches to this problem. The first approach uses a single learner and extends the traditional POMDP formulation of dialogue state with features that describe the user context. The second approach segments users by context and then employs a learner per context. We compare these approaches in a benchmark of existing non-RL and RL-based methods in three established and one novel application domain of financial product recommendation. We compare the influence of context and training experiences on performance and find that learning approaches generally outperform a handcrafted gold standard.

The huge wealth of data in the health domain can be exploited to create models that predict development of health states over time. Temporal learning algorithms are well suited to learn relationships between health states and make predictions about their future developments. However, these algorithms: (1) either focus on learning one generic model for all patients, providing general insights but often with limited predictive performance, or (2) learn individualized models from which it is hard to derive generic concepts. In this paper, we present a middle ground, namely parameterized dynamical systems models that are generated from data using a Genetic Programming (GP) framework. A fitness function suitable for the health domain is exploited. An evaluation of the approach in the mental health domain shows that performance of the model generated by the GP is on par with a dynamical systems model developed based on domain knowledge, significantly outperforms a generic Long Term Short Term Memory (LSTM) model and in some cases also outperforms an individualized LSTM model.

This paper demonstrates how different machine learning techniques performed on a recent, partially labeled dataset (based on the Locked Shields 2017 exercise) and which features were deemed important. Moreover, a cybersecurity expert analyzed the results and validated that the models were able to classify the known intrusions as malicious and that they discovered new attacks. In a set of 500 detected anomalies, 50 previously unknown intrusions were found. Given that such observations are uncommon, this indicates how well an unlabeled dataset can be used to construct and to evaluate a network intrusion detection system.

Predicting therapeutic outcome in the mental health domain is of utmost importance to enable therapists to provide the most effective treatment to a patient. Using information from the writings of a patient can potentially be a valuable source of information, especially now that more and more treatments involve computer-based exercises or electronic conversations between patient and therapist. In this paper, we study predictive modeling using writings of patients under treatment for a social anxiety disorder. We extract a wealth of information from the text written by patients including their usage of words, the topics they talk about, the sentiment of the messages, and the style of writing. In addition, we study trends over time with respect to those measures. We then apply machine learning algorithms to generate the predictive models. Based on a dataset of 69 patients, we are able to show that we can predict therapy outcome with an area under the curve of 0.83 halfway through the therapy and with a precision of 0.78 when using the full data (i.e., the entire treatment period). Due to the limited number of participants, it is hard to generalize the results, but they do show great potential in this type of information.

More than a decade after the first extensive overview on parameter control, we revisit the field and present a survey of the state-of-the-art. We briefly summarize the development of the field and discuss existing work related to each major parameter or component of an evolutionary algorithm. Based on this overview, we observe trends in the area, identify some (methodological) shortcomings, and give recommendations for future research.

Parameter control in Evolutionary Computing stands for an approach to parameter setting that changes the parameters of an Evolutionary Algorithm (EA) on-the-fly during the run. In this paper we address the issue of a generic and parameter-independent controller that can be readily plugged into an existing EA and offer performance improvements by varying the EA parameters during the problem solution process. Our approach is based on a careful study of Reinforcement Learning (RL) theory and the use of existing RL techniques. We present experiments using various state-of-the-art EAs solving different difficult problems. Results show that our RL control method has very good potential in improving the quality of the solution found without requiring additional resources or time and with minimal effort from the designer of the application.

Parameter control mechanisms in evolutionary algorithms (EAs) dynamically change the values of the EA parameters during a run. Research over the last two decades has delivered ample examples where an EA using a parameter control mechanism outperforms its static version with fixed parameter values. However, very few have investigated why such parameter control approaches perform better. In principle, it could be the case that using different parameter values alone is already sufficient and EA performance can be improved without sophisticated control strategies raising an issue in the methodology of parameter control mechanisms' evaluation. This paper investigates whether very simple random variation in parameter values during an evolutionary run can already provide improvements over static values. Results suggest that random variation of parameters should be included in the benchmarks when evaluating a new parameter control mechanism.

Collective decision making involves on the one hand individual mental states such as beliefs, emotions and intentions, and on the other hand interaction with others with possibly different mental states. Achieving a satisfactory common group decision on which all agree requires that such mental states are adapted to each other by social interaction. Recent developments in social neuroscience have revealed neural mechanisms by which such mutual adaptation can be realised. These mechanisms not only enable intentions to converge to an emerging common decision, but at the same time enable to achieve shared underlying individual beliefs and emotions. This paper presents a computational model for such processes. As an application of the model, an agent-based analysis was made of patterns in crowd behaviour, in particular to simulate a real-life incident that took place on May 4, 2010 in Amsterdam. From available video material and witness reports, useful empirical data were extracted. Similar patterns were achieved in simulations, whereby some of the parameters of the model were tuned to the case addressed, and most parameters were assigned default values. The results show the inclusion of contagion of belief, emotion, and intention states of agents results in better reproduction of the incident than non-inclusion.

Within human trust related behaviour, according to the literature from the domains of Psychology and Social Sciences often non-rational behaviour can be observed. Current trust models that have been developed typically do not incorporate non-rational

To support human functioning, ambient intelligent agents require knowledge about the tasks executed by the human. This knowledge includes design-time information like: (i) the goal of a task and (ii) the alternative ways for a human to achieve that goal, as well as run-time information such as the choices made by a human during task execution. In order to provide effective support, the agent must know exactly what steps the human is following. However, if not all steps along the path can be observed, it is possible that the agent cannot uniquely derive which path the human is following. Furthermore, in order to provide timely support, the agent must observe, reason, conclude and support within a limited period of time. To deal with these problems, this paper presents a generic focused reasoning mechanism to enable a guided selection of the path which is most likely followed by the human. This mechanism is based upon knowledge about the human and the workflow to perform the task. In order to come to such an approach, a reasoning mechanism is adopted in combination with the introduction of a new workflow representation, which is utilized to focus the reasoning process in an appropriate manner. The approach is evaluated by means of an extensive case study.

In order for agents to be able to act intelligently in an environment, a first necessary step is to become aware of the current situation in the environment. Forming such awareness is not a trivial matter. Appropriate observations should be selected by the agent, and the observation results should be interpreted and combined into one coherent picture. Humans use dedicated mental models which represent the relationships between various observations and the formation of beliefs about the environment, which then again direct the further observations to be performed. In this paper, a generic agent model for situation awareness is proposed that is able to take a mental model as input, and utilize this model to create a picture of the current situation. In order to show the suitability of the approach, it has been applied within the domain of F-16 fighter pilot training for which a dedicated mental model has been specified, and simulations experiments have been conducted. 1

Central research questions addressed within Criminology are how the geographical displacement of crime can be understood, explained, and predicted. The process of crime displacement is usually explained by referring to the interaction of three types

Multi-agent organization modeling, norm violation, avoidance In contemporary society not adhering to norms is something which is unwanted. Currently approaches to prevent this from happening are often taken whereby the deviation of a norm is punished and possible repair actions are performed. However, this is a reactive approach and can only happen once the norm has already been violated. This paper presents a proactive approach which allows intervention before the deviation actually occurs. In order to do this, an approach is specified for agents that enforce norms and can influence the input states of agents. This approach includes learning of input/output correlations of these agents by constructing decision trees, and utilizing this decision tree to intervene such that norm violation can be avoided. The approach is evaluated in the domain of Criminology.

Within Criminology, the process of crime displacement is usually explained by referring to the interaction of three types of agents: criminals, passers-by, and guardians. Most existing simulation models of this process are agent-based. However, when the number of agents considered becomes large, population-based simulation has computational advantages over agent-based simulation. This paper presents both an agent-based and a population-based simulation model of crime displacement, and reports a comparative evaluation of the two models. In addition, an approach is put forward to analyse the behaviour of both models by means of formal techniques.

Both for developing human-like virtual agents and for developing intelligent systems that make use of knowledge about the emotional state of the user, it is important to model the mood of a person. In this paper, a model for simulating the dynamics of mood is presented, based on psychological theories about a uni-polar clinical depression. The model was analyzed mathematically and by means of simulations, and it was shown that the model exhibits the most important characteristics of the theories. It shows how stress factors under some conditions can lead to a depression, while it will not lead to a depression under other conditions.

The ability to express preferences for specific tasks in multi-agent auctions is an important element for potential users who are considering to use such auctioning systems. This paper presents an approach to make such preferences explicit and to use these preferences in bids for reverse combinatorial auctions. Three different types of preference are considered: (1) preferences for particular durations of tasks, (2) preferences for certain time points, and (3) preferences for specific types of tasks. We study empirically the tradeoffs between the quality of the solutions obtained and the use of preferences in the bidding process, focusing on effects such as increased execution time. We use both synthetic data as well as real data from a logistics company.

This paper presents a decentralized task allocation method that can handle allocation of tasks with time and precedence constraints in a multi-agent setting where not all information needed for a centralized approach is shared.