Barbara Hammer - Selected Publications#

According to Google Scholar, the Barbara Hammer's publications have received a total of over 11,000 citations and have an h-index of 49 as of January 2024. Below are 10 major publications. The list is a combination of works that are responsible for Barbara Hammer's international recognition, as well as recent high-impact publications that appeared in top journals / A* conferences. [Number of citations is reported according to Google Scholar as of January 2024]

[1] Physics-Informed Graph Neural Networks for Water Distribution Systems; I Ashraf, J Strotherm, L Hermes, B Hammer, accepted for AAAI 2024 (AISI Track) as an oral contribution, https://aaai.org/wp-content/uploads/2024/01/AI-for-Social-Impact-Schedule.pdf)

(This publication introduces a novel physics-informed machine learning technology to support time-critical decisions in critical infrastructure, accompanied by mathematical guarantees. The contribution has been accepted in the A* conference AAAI-2024 as an oral talk of the AI for Social Impact (AISI) Track, only 6.2% of papers submitted to the AISI track have been scheduled for oral presentation.)

[2] Incremental permutation feature importance (iPFI): towards online explanations on data streams; F Fumagalli, M Muschalik, E Hüllermeier, B Hammer, Machine Learning 112 (12), 4863-4903, 2023 [4 citations]

(This is one of the first efficient explainable artificial intelligence technologies for streaming data, such as required for adaptive human machine interaction. The contribution has been selected as one of the best papers of the journal track of ECML'23)

[3] Let's go to the Alien Zoo: Introducing an experimental framework to study usability of counterfactual explanations for machine learning; U Kuhl, A Artelt, B Hammer; Frontiers in Computer Science 5, 20, 2023 [10 citations]

(The contribution introduces a novel experimental framework to evaluate the effectiveness of explainable AI methods via gamification - it constitutes an example of the high degree of interdisciplinary work conducted in my group, which is required to account for cognitive, ethical and social aspects of machine learning.)

[4] Reservoir memory machines as neural computers; B Paaßen, A Schulz, TC Stewart, B Hammer; IEEE Transactions on Neural Networks and Learning Systems 33 (6), 2575-2585, 2021 [4 citations]

(This article links efficiently trainable neural computation models to classical computation paradigms from the Chomsky hierarchy, thereby using novel and innovative combinations of proof techniques from theoretical computer science and machine learning. The work displays the diversity of mathematical techniques which are required to substantiate neural methods by rigorous theoretical insights. )

[5] Counteracting Electrode Shifts in Upper-Limb Prosthesis Control via Transfer Learning, C Prahm, A Schulz, B Paaßen, J Schoisswohl, E. Kaniusas, G. Dorffner, B. Hammer, O. Aszmann, in IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 27, no. 5, pp. 956-962, 2019 [46 citations]

(This contribution introduces an efficient machine learning scheme which enables robust individualized prosthesis control. It has been developed together with medical experts and successfully evaluated in the clinical context.)

[6] Benjamin Paaßen, Barbara Hammer, Thomas William Price, Tiffany Barnes, Sebastian Gross, Niels Pinkwart: The Continuous Hint Factory - Providing Hints in Vast and Sparsely Populated Edit Distance Spaces. Journal of Educational Data Mining, 10(1): 1–35 (2018) [60 citations]

(The work displays the capability of neuro-symbolic methods to support persons when learning how to program - a topic of increasing relevance in the light of AI tools such as large language models; our model has been used e.g. in the first semester programming courses in Computer Science at FU Berlin and Bielefeld University. )

[7] Incremental on-line learning: A review and comparison of state of the art algorithms, V Losing, B Hammer, H Wersing, Neurocomputing 275, 1261-1274, 2018 [359 citations]

(Unlike classical batch models, incremental learning systems adapt instantaneously when data arrive; such technologies are crucial for model personalization and learning in possibly changing environments. The article is the first to experimentally compare the main approaches. Based on the insights, new models have been developed in my group.)

[8] KNN Classifier with Self Adjusting Memory for Heterogeneous Concept Drift, V Losing, B Hammer, H Wersing, IEEE International Conference on Data Mining (ICDM), 2016 [256]

(The model SAM-kNN constitutes one specific model for on-line learning in the presence of non-stationary environments which has been developed in from my group. It has been awarded the best paper award at the A* conference IEEE ICDM'2016, it has been integrated into the popular toolbox Scikit-Multiflow, and it has been used by companies for personalization in assistive driving, among other applications)

[9] Parametric nonlinear dimensionality reduction using kernel t-SNE, A Gisbrecht, A Schulz, B Hammer, Neurocomputing 147, 71-82, 2015 [254 citations]

(Modern nonlinear dimensionality reduction methods offer powerful tools to visually inspect large volumes of data. Many popular methods, however, are non-parametric and require novel training whenever new data arrives. The proposal offers an efficient method to avoid this issue and which can be used on top of every dimensionality reduction method. It has been used as starting point for visualization tools for deep networks, among others.)

[10] Adaptive relevance matrices in learning vector quantization, P Schneider, M Biehl, B Hammer, Neural computation 21 (12), 3532-3561, 2009 [438 citations]

(The work proposes GMLVQ, a natively interpretably prototype-based machine learning method which can be accompanied by very strong formal guarantees as concerns its generalization ability. It has been used in diverse context, including medicine, autonomous driving, industrial quality control, etc including academic partners and several major companies. As an example, the relevant features identified by the model have been the starting point of a patent filed for the non-invasive identification of adreno-cortical cancer.)