Combining Symbolic Reasoning and Deep Learning for Human Activity Recognition

Rueda, Fernando Moya; Lüdtke, Stefan; Schröder, Max; Yordanova, Kristina; Kirste, Thomas; Fink, Gernot A.

doi:10.1109/percomw.2019.8730792

Cited by 8 publications

(3 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Data cleaning techniques use one or more filters to identify noisy data and to correct or delete them [22] [23] [24]. They can also process missing values and detect outliers [18] [25].…”

Section: Using Machine Learning For Context Miningmentioning

confidence: 99%

Machine Learning for a Context Mining Facility

Rabah

Pinheiro²,

Grand³

et al. 2020

2020 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops)

View full text Add to dashboard Cite

This paper considers generalizing context reasoning capabilities through a context mining facility offered to all Information System applications. This facility requires mining context data at the system scale, which raises several challenges for Machine Learning approaches used for such mining. Through a detailed literature review, we analyze these approaches with regard to the requirements of such a context mining facility at the Information System level, pointing to the potential and to the challenges raised by this perspective.

show abstract

“…Data cleaning techniques use one or more filters to identify noisy data and to correct or delete them [22] [23] [24]. They can also process missing values and detect outliers [18] [25].…”

Section: Using Machine Learning For Context Miningmentioning

confidence: 99%

Machine Learning for a Context Mining Facility

Rabah

Pinheiro²,

Grand³

et al. 2020

2020 IEEE International Conference on Pervasive Computing and Communications Workshops (PerCom Workshops)

View full text Add to dashboard Cite

show abstract

“…However, deep learning approaches can be overconfident in their classification, even for incorrect classifications, and information regarding the user’s context is not easily extracted from the learned network. To address these issues, Rueda et al [32] proposed a hybrid activity recognition architecture, called Hybrid Computational Causal Behaviour Model (HCCBM), which combines deep learning with symbolic models, namely, CSSMs. However, deep learning has not been applied to our graph-based representation of sensor data.…”

Section: Related Workmentioning

confidence: 99%

Improving IoT Predictions through the Identification of Graphical Features

Akter

Holder

2019

Sensors

View full text Add to dashboard Cite

IoT sensor networks have an inherent graph structure that can be used to extract graphical features for improving performance in a variety of prediction tasks. We propose a framework that represents IoT sensor network data as a graph, extracts graphical features, and applies feature selection methods to identify the most useful features that are to be used by a classifier for prediction tasks. We show that a set of generic graph-based features can improve performance of sensor network predictions without the need for application-specific and task-specific feature engineering. We apply this approach to three different prediction tasks: activity recognition from motion sensors in a smart home, demographic prediction from GPS sensor data in a smart phone, and activity recognition from GPS sensor data in a smart phone. Our approach produced comparable results with most of the state-of-the-art methods, while maintaining the additional advantage of general applicability to IoT sensor networks without using sophisticated and application-specific feature generation techniques or background knowledge. We further investigate the impact of using edge-transition times, categorical features, different sensor window sizes, and normalization in the smart home domain. We also consider deep learning approaches, including the Graph Convolutional Network (GCN), for the elimination of feature engineering in the smart home domain, but our approach provided better performance in most cases. We conclude that the graphical feature-based framework that is based on IoT sensor categorization, nodes and edges as features, and feature selection techniques provides superior results when compared to the non-graph-based features.

show abstract

“…Technological advancements grow the range of wearable sensors from smartwatches to textiles and smart glasses. Advancements in Artificial Intelligence (AI) and data analytics [ 8 , 9 ] allow detection and prediction of patterns and risk indicators over wearable sensor data and can enable more timely and efficient decision making. Still, interoperability and universal knowledge representation and management are needed to address the vast heterogeneity of data sources, devices and vendors to allow such knowledge extraction [ 10 ].…”

Section: Introductionmentioning

confidence: 99%

Detection of Health-Related Events and Behaviours from Wearable Sensor Lifestyle Data Using Symbolic Intelligence: A Proof-of-Concept Application in the Care of Multiple Sclerosis

Stavropoulos

Meditskos

Lazarou

et al. 2021

Sensors

View full text Add to dashboard Cite

In this paper, we demonstrate the potential of a knowledge-driven framework to improve the efficiency and effectiveness of care through remote and intelligent assessment. More specifically, we present a rule-based approach to detect health related problems from wearable lifestyle sensor data that add clinical value to take informed decisions on follow-up and intervention. We use OWL 2 ontologies as the underlying knowledge representation formalism for modelling contextual information and high-level concepts and relations among them. The conceptual model of our framework is defined on top of existing modelling standards, such as SOSA and WADM, promoting the creation of interoperable knowledge graphs. On top of the symbolic knowledge graphs, we define a rule-based framework for infusing expert knowledge in the form of SHACL constraints and rules to recognise patterns, anomalies and situations of interest based on the predefined and stored rules and conditions. A dashboard visualizes both sensor data and detected events to facilitate clinical supervision and decision making. Preliminary results on the performance and scalability are presented, while a focus group of clinicians involved in an exploratory research study revealed their preferences and perspectives to shape future clinical research using the framework.

show abstract

Combining Symbolic Reasoning and Deep Learning for Human Activity Recognition

Cited by 8 publications

References 15 publications

Machine Learning for a Context Mining Facility

Machine Learning for a Context Mining Facility

Improving IoT Predictions through the Identification of Graphical Features

Detection of Health-Related Events and Behaviours from Wearable Sensor Lifestyle Data Using Symbolic Intelligence: A Proof-of-Concept Application in the Care of Multiple Sclerosis

Contact Info

Product

Resources

About