Using ontologies to enhance human understandability of global post-hoc explanations of black-box models

Confalonieri, Roberto; Weyde, Tillman; Besold, Tarek R.; Martín, Fermín Moscoso del Prado

doi:10.1016/j.artint.2021.103471

Cited by 85 publications

(38 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…This hierarchy of concept could also be used to improve the interpretability and explicability of results. Indeed, understanding how Deep Learning combines information to effectively classify land use classes remains a challenging task, but recent research using ontologies could be useful to achieve this goal [54]. This idea could also highlight missing exogenous information (elevation, cadastre, etc.)…”

Section: Discussionmentioning

confidence: 99%

Assessment of Deep Learning Techniques for Land Use Land Cover Classification in Southern New Caledonia

et al. 2021

View full text Add to dashboard Cite

Land use (LU) and land cover (LC) are two complementary pieces of cartographic information used for urban planning and environmental monitoring. In the context of New Caledonia, a biodiversity hotspot, the availability of up-to-date LULC maps is essential to monitor the impact of extreme events such as cyclones and human activities on the environment. With the democratization of satellite data and the development of high-performance deep learning techniques, it is possible to create these data automatically. This work aims at determining the best current deep learning configuration (pixel-wise vs semantic labelling architectures, data augmentation, image prepossessing, …), to perform LULC mapping in a complex, subtropical environment. For this purpose, a specific data set based on SPOT6 satellite data was created and made available for the scientific community as an LULC benchmark in a tropical, complex environment using five representative areas of New Caledonia labelled by a human operator: four used as training sets, and the fifth as a test set. Several architectures were trained and the resulting classification was compared with a state-of-the-art machine learning technique: XGboost. We also assessed the relevance of popular neo-channels derived from the raw observations in the context of deep learning. The deep learning approach showed comparable results to XGboost for LC detection and over-performed it on the LU detection task (61.45% vs. 51.56% of overall accuracy). Finally, adding LC classification output of the dedicated deep learning architecture to the raw channels input significantly improved the overall accuracy of the deep learning LU classification task (63.61% of overall accuracy). All the data used in this study are available on line for the remote sensing community and for assessing other LULC detection techniques.

show abstract

Section: Discussionmentioning

confidence: 99%

Assessment of Deep Learning Techniques for Land Use Land Cover Classification in Southern New Caledonia

et al. 2021

View full text Add to dashboard Cite

show abstract

“…als Antwort darauf werden zu der Frage geführt, wie leistungsstark Random Forests sein können [7,26]. Eine weitere Alternative stellen Gradient Boosted Trees dar [1] [2,3,21]; die Blackbox-Modelle gehen mit den Entscheidungsbäumen eine symbiotische Beziehung ein, ein spannendes Forschungsfeld mit Potenzial für erklärbare und leistungsstarke Modelle. Die traditionellen Entscheidungsbäume werden ständig weiterentwickelt und stellen ein vielseitiges Werkzeug für moderne KI-Anwendungen dar.…”

Section: Fortgeschrittene Methodenunclassified

Lernende Entscheidungsbäume

Knuth

2021

Informatik Spektrum

View full text Add to dashboard Cite

ZusammenfassungEntscheidungsbäume bereichern moderne künstliche Intelligenz durch vielseitige Weiterentwicklungen. Lernende Entscheidungsbäume sind seit Jahrzehnten ein tradiertes Werkzeug zur Erstellung von erklärbaren Modellen zu Klassifikation und Regression. Während moderne Trends wie Deep Learning die Aufmerksamkeit der KI-Entwicklung auf sich ziehen, werden Entscheidungsbäume kontinuierlich weiterentwickelt, in neuen Bereichen eingesetzt und mit scheinbar im Wettstreit stehenden Methoden verschmolzen. Der Artikel stellt die Funktionsweise von Entscheidungsbäumen dar und zeigt Entwicklungen der aktuellen Forschung auf.

show abstract

“…It is not new for modeling experts to face questions about model acceptability and validation as we expose here, but they were trained in a different context and environment when comparing with the currently era of data scientists exposed to AI. Efforts are being made to promote greater transparency in algorithms with low levels of expert intervention [84][85][86][87][88]. Miller [89] names these efforts as "explainable artificial intelligence research" and considers that there will be an increasing need to integrate AI with other fields of knowledge, such as philosophy, cognitive psychology/science, and social psychology.…”

Section: Decision Makingmentioning

confidence: 99%

New dilemmas, old problems: advances in data analysis and its geoethical implications in groundwater management

Silva¹,

Matulovic

Manzione

2021

SN Appl. Sci.

View full text Add to dashboard Cite

Groundwater governance uses modeling to support decision making. Therefore, data science techniques are essential. Specific difficulties arise because variables must be used that cannot be directly measured, such as aquifer recharge and groundwater flow. However, such techniques involve dealing with (often not very explicitly stated) ethical questions. To support groundwater governance, these ethical questions cannot be solved straightforward. In this study, we propose an approach called “open-minded roadmap” to guide data analytics and modeling for groundwater governance decision making. To frame the ethical questions, we use the concept of geoethical thinking, a method to combine geoscience-expertise and societal responsibility of the geoscientist. We present a case study in groundwater monitoring modeling experiment using data analytics methods in southeast Brazil. A model based on fuzzy logic (with high expert intervention) and three data-driven models (with low expert intervention) are tested and evaluated for aquifer recharge in watersheds. The roadmap approach consists of three issues: (a) data acquisition, (b) modeling and (c) the open-minded (geo)ethical attitude. The level of expert intervention in the modeling stage and model validation are discussed. A search for gaps in the model use is made, anticipating issues through the development of application scenarios, to reach a final decision. When the model is validated in one watershed and then extrapolated to neighboring watersheds, we found large asymmetries in the recharge estimatives. Hence, we can show that more information (data, expertise etc.) is needed to improve the models’ predictability-skill. In the resulting iterative approach, new questions will arise (as new information comes available), and therefore, steady recourse to the open-minded roadmap is recommended. Graphic abstract

show abstract

Using ontologies to enhance human understandability of global post-hoc explanations of black-box models

Cited by 85 publications

References 37 publications

Assessment of Deep Learning Techniques for Land Use Land Cover Classification in Southern New Caledonia

Assessment of Deep Learning Techniques for Land Use Land Cover Classification in Southern New Caledonia

Lernende Entscheidungsbäume

New dilemmas, old problems: advances in data analysis and its geoethical implications in groundwater management

Contact Info

Product

Resources

About