KiMeKo (KI-Med-Kollaborationsplattform) is a publically funded collaborative research project that develops a sustainable AI-Med ecosystem for AI-based medical device development. The project runs from July 2024 to December 2027 and joins seven Northern German research institutions. KiMeKo addresses the complete development trajectory, from concept and data acquisition to validation, regulatory evidence generation, and approval-oriented documentation. The project contributes a practical toolchain and platform capabilities for non-experts and experts, including structured innovation support, uncertaintyaware sensor-data fusion, hybrid expert-system modeling, and workflow-guided data acquisition and anonymization. This paper summarizes project objectives, expected outputs, relevance to IEEE COMPSAC 2026 themes, and current progress. In particular, KiMeKo aligns with Applied AI and Smart & Connected Health by combining AI engineering, privacy-conscious data processing, and regulation-aware medical software development.

Robust analysis of biosignals hinges on well-crafted processing pipelines, yet assembling them is still a slow, errorprone exercise, requiring both domain expertise and programming skills. While Large Language Models (LLMs) offer promising assistance, they fundamentally lack the combinatorial reasoning capabilities needed for designing reliable and reproducible processing pipelines. We present an innovative hybrid system that harnesses LLMs' natural language understanding while leveraging classical AI planning for logical reasoning and validation. A retrieval-augmented model parses natural language pipeline descriptions into implementation-independent atomic biosignal processing operations, which follow to a Hierarchical Task Network (HTN) domain that can plan and validate processing workflows, as well as flag violations of biosignal processing best practices. Evaluation on a preliminary corpus of scenarios demonstrates$\mathbf{0. 8 9 - 0. 9 8}$precision in mapping natural language to processing blocks and 0.97-0.98 F1-score in generating valid 2-15 step pipelines using SHOP3 planner. A drag-and-drop pipeline design interface allows users to build and formally validate their pipeline ideas from scratch, showing potential to democratize and expedite biosignal analysis for scientists of different backgrounds.

This paper examines the impact of neurosymbolic learning on sequence analysis in Structured Probability Spaces (SPS), comparing its effectiveness against a purely neural approach. Sequence analysis in SPS is challenging due to the combinatorial explosion of states and the difficulty of obtaining sufficient annotated training samples. Additionally, in SPS, the set of realizations with non-zero support is often a scattered, non-trivial subset of the Cartesian product of variables, adding complexity to learning and inference. The problem of sequence analysis in SPS emerges, for example, in reconstructing the activities of goal-directed agents from noisy and ambiguous sensor data. We explore the potential of neurosymbolic methods, which integrate symbolic background knowledge with neural learning, to constrain the hypothesis space and improve learning efficiency. Specifically, we conduct a simulation study in human activity recognition using DeepProbLog as a representative for neurosymbolic learning. Our results demonstrate that incorporating symbolic knowledge improves sample efficiency, generalization, and zero-shot learning, compared to a purely neural approach. Furthermore, we show that neurosymbolic models maintain robust performance under data scarcity while offering enhanced interpretability and stability. These findings suggest that neurosymbolic learning provides a promising foundation for sequence analysis in complex, structured domains, where purely neural approaches struggle with insufficient training data and limited generalization ability.

This document provides guidelines to using the data in the kitchen task assessment dataset. This dataset contains normal behaviour as well as erroneous behaviour due to dementia. The dataset including annotation, sensor data and synchronisation scripts is publicly available and can be downloaded from the Library of the University of Rostock [4] , download link https://doi.org/10.18453/rosdok_id00002605 . The video data is accessible on request. If you are interested in the video data, please, contact Peter Eschholz at peter.eschholz( at )uni-rostock.de.

We present a model for Bayesian filtering (BF) in discrete dynamic systems where multiple entities (inter)-act, i.e. where the system dynamics is naturally described by a Multiset rewriting system (MRS). Typically, BF in such situations is computationally expensive due to the high number of discrete states that need to be maintained explicitly. We devise a lifted state representation, based on a suitable decomposition of multiset states, such that some factors of the distribution are exchangeable and thus afford an efficient representation. Intuitively, this representation groups together similar entities whose properties follow an exchangeable joint distribution. Subsequently, we introduce a BF algorithm that works directly on lifted states, without resorting to the original, much larger ground representation. This algorithm directly lends itself to approximate versions by limiting the number of explicitly represented lifted states in the posterior. We show empirically that the lifted representation can lead to a factorial reduction in the representational complexity of the distribution, and in the approximate cases can lead to a lower variance of the estimate and a lower estimation error compared to the original, ground representation.

We present a model for exact recursive Bayesian filtering based on lifted multiset states. Combining multisets with lifting makes it possible to simultaneously exploit multiple strategies for reducing inference complexity when compared to list-based grounded state representations. The core idea is to borrow the concept of Maximally Parallel Multiset Rewriting Systems and to enhance it by concepts from Rao-Blackwellization and Lifted Inference, giving a representation of state distributions that enables efficient inference. In worlds where the random variables that define the system state are exchangeable -- where the identity of entities does not matter -- it automatically uses a representation that abstracts from ordering (achieving an exponential reduction in complexity) -- and it automatically adapts when observations or system dynamics destroy exchangeability by breaking symmetry.

Tasks such as social network analysis, human behavior recognition, or modeling biochemical reactions, can be solved elegantly by using the probabilistic inference framework. However, standard probabilistic inference algorithms work at a propositional level, and thus cannot capture the symmetries and redundancies that are present in these tasks. Algorithms that exploit those symmetries have been devised in different research fields, for example by the lifted inference-, multiple object tracking-, and modeling and simulation-communities. The common idea, that we call state space abstraction, is to perform inference over compact representations of sets of symmetric states. Although they are concerned with a similar topic, the relationship between these approaches has not been investigated systematically. This survey provides the following contributions. We perform a systematic literature review to outline the state of the art in probabilistic inference methods exploiting symmetries. From an initial set of more than 4,000 papers, we identify 116 relevant papers. Furthermore, we provide new high-level categories that classify the approaches, based on common properties of the approaches. The research areas underlying each of the categories are introduced concisely. Researchers from different fields that are confronted with a state space explosion problem in a probabilistic system can use this classification to identify possible solutions. Finally, based on this conceptualization, we identify potentials for future research, as some relevant application domains are not addressed by current approaches.

In sensor-based activity recognition often huge amounts of data have to be acquired from multiple sensors, which need to be communicated for further processing. When using wireless sensor nodes, energy efficiency is of outstanding importance, since it directly influences the time until the battery needs to be recharged. However, communicating a huge amount of sensor data over wireless interfaces causes high energy consumption as well. Furthermore, the host controller receiving the sensor data is hindered of using its low power modes, as it needs to be woken up more frequently as well. In the paper at hand, we introduce our approach of reducing energy consumption of sensor nodes in online activity recognition scenarios by calculating the feature extraction on the sensor subsystem itself. By doing so, we can reduce the output rate of the sensor which enables the host controller to stay in its low power modes for longer periods. Additionally, this approach drastically reduces the amount of data to be transmitted over wireless interfaces, which further improves energy consumption. In our experiments, the proposed approach reduces the energy consumption of a sensor node by up to 33 %.

The reconstruction of human activities is an important prerequisite to provide assistance. In this paper, we present an activity and plan recognition approach which is based on causal models of human activities. We show, that it is possible to estimate current activities, the underlying goal of the user, and context information about the state of the environment from noisy sensor data. Therefore we use real world data obtained from a smart home system while observing unrestricted activities of daily living in an inhabited flat. We evaluate the accuracy of the recognition for simulated data of different granularity and data obtained from the smart home system. We furthermore show that performance measures solely based on action sequences are not sufficient to evaluate a recognition system.

There are various activity recognition approaches that rely on manual definition of precondition-effect rules to describe human behaviour. These rules are later used to generate computational models of human behaviour that are able to reason about the user behaviour based on sensor observations. One problem with these approaches is that the manual rule definition is time consuming and error prone process. To address this problem, in this paper we propose an approach that learns the rules from textual instructions. In difference to existing approaches, it is able to learn the causal relations between the actions without initial training phase. Furthermore, it learns the domain ontology that is used for the model generalisation and specialisation. To evaluate the approach, a model describing cooking task was learned and later applied for explaining seven plans of actual human behaviour. It was then compared to a hand-crafted model describing the same problem. The results showed that the learned model was able to recognise the plans with higher overall probability compared to the hand-crafted model. It also learned a more complex domain ontology and was more general than the hand-crafted model. In general, the results showed that it is possible to learn models of human behaviour from textual instructions which are able to explain actual human behaviour.

Several emerging approaches to activity recognition (AR) combine symbolic representation of user actions with probabilistic elements for reasoning under uncertainty. These approaches provide promising results in terms of recognition performance, coping with the uncertainty of observations, and model size explosion when complex problems are modelled. But experience has shown that it is not always intuitive to model even seemingly simple problems. To date, there are no guidelines for developing such models. To address this problem, in this work we present a development process for building symbolic models that is based on experience acquired so far as well as on existing engineering and data analysis workflows. The proposed process is a first attempt at providing structured guidelines and practices for designing, modelling, and evaluating human behaviour in the form of symbolic models for AR. As an illustration of the process, a simple example from the office domain was developed. The process was evaluated in a comparative study of an intuitive process and the proposed process. The results showed a significant improvement over the intuitive process. Furthermore, the study participants reported greater ease of use and perceived effectiveness when following the proposed process. To evaluate the applicability of the process to more complex AR problems, it was applied to a problem from the kitchen domain. The results showed that following the proposed process yielded an average accuracy of 78%. The developed model outperformed state-of-the-art methods applied to the same dataset in previous work, and it performed comparably to a symbolic model developed by a model expert without following the proposed development process.

Bayesian filtering is one of the most important frameworks for applications such as activity recognition and situation recognition. Current applications involving human behaviour models of large (possibly infinite) discrete state spaces imposes challenges to current inference algorithms. In these complex models approximate solutions are inevitable, in particular Sequential Monte Carlo methods are of great interest. We investigate a key component of the particle filter - the resampling step - for semi-Markov models with discrete states. Particle filters and resampling strategies have only been investigated in detail for continuous models. However, efficient inference for models of human behaviour with discrete states requires methods tailored for discrete state spaces.

Imagine an office environment equipped with infrared presence sensors for light control as modern buildings offer. Several persons might be moving in this environment independently from each other. Potential questions arising here are: Which rooms are occupied?", "How many persons are in room 123?", or "Where is John?". In this paper we present an information fusion approach based on causal models. Prior knowledge is used to solve the simultaneous identification and multi target tracking problem. Based on an experimental setting, we show that the proposed approach enables tracking up to five persons. Furthermore, we discuss potential applications in the domain of fire service or monitoring of elderly people.

Activity recognition is a challenging research problem in ubiquitous computing domain and has to tackle omnipresent uncertainties, e.g., resulting from ambiguous or intermittent sensor readings. In this paper, we introduce an activity recognition approach based on causal modeling and probabilistic plan recognition. To evaluate the performance of our approach systematically, we generated sensor data with different error rates using a simulation. This data served as input for the activity recognition in a series of experiments. In these experiments we stepwise introduced and combined additional sources of uncertainty, i.e., different duration models and ignoring certain sensors, to demonstrate the robustness of our approach. Our evaluation shows that Computational Causal Behavior Models provide a basis for a robust activity recognition system.

Smart Environments are designed to proactively assist their users. One way to achieve this is by composing sequences of potential user actions and of devices' actions at run time. This requires the actions to be specified in a declarative manner. A favorable formalism for declarative descriptions are semantic web services. In this paper, we describe our approach to proactive assistance in smart environments and explain how semantic web services can be employed as the basic building blocks for such an environment. We provide a state-of-the-art review of existing semantic web service technologies and discuss how they should be enhanced to suit our needs.

Among the scenarios for pervasive learning are teleteaching and project work. The former tries to deliver a lecture to a distributed audience, while the latter tries to enhance the learning effect by concentrated collaboration. In this paper, we want to show how an ambient environment, designed to support teams in a meeting, can be used to support pervasive learning scenarios as well as meetings. The developed applications help the users in both types of scenarios with pro-active assistance, enabling the users to focus on their task, rather than fiddling with the controls of the devices.

Ubiquitous computing aims for the realization of environments that assist users autonomously and proactively. Therefore smart environment infrastructures need to be able to identify users needs (intention recognition) and to plan an appropriate assisting strategy. Both is matter for research. In our approach we address inferring the intention of a team within a smart meeting environment. This becomes a central challenge, especially if multiple users are observed by noisy heterogeneous sensors. We propose a team behavior model based on hierarchical dynamic Bayesian network (DBN) for inferring the current task and activity of a team of users online. Given (noisy and intermittent) sensor readings of the team members' positions in a meeting room, we are interested in inferring the team's current objective. We implemented the model using particle filters for inference and demonstrate that by adding knowledge about the meeting agenda prediction accuracy and speed is improved. Evaluation of simulation data answers the question, how precise agenda knowledge must be to predict team behavior optimally.

Many context-aware projects try to develop the next step in human computer interaction, systems that adapt to a users need and help him to focus on his specific task. Probabilistic models are used to infer the current activity of a user. These techniques for predicting the actions of a user are often custom-tailored to a fixed location and scenario. We developed a method to generate probabilistic models for different context, therefore broaden their use in different domains of ubiquitous computing. This makes our intention analysis more generic.