Robustness of Generalized Learning Vector Quantization Models Against Adversarial Attacks

Saralajew, Sascha; Holdijk, Lars; Rees, Maike; Villmann, Thomas

doi:10.1007/978-3-030-19642-4_19

Cited by 17 publications

(16 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…According to the SCE-decision, 59 data points were rejected from classification by the 341 learned GMLVQ classifier. The result is given in S7 Fig using the 2), which underlies its 359 well-known robustness [47]. Thus, we observe an overall precise agreement supporting 360 the findings in [12].…”

supporting

confidence: 81%

See 1 more Smart Citation

Analysis of SARS-CoV-2 RNA-Sequences by Interpretable Machine Learning Models

Kaden

Bohnsack

Weber

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

We present an approach to investigate SARS-CoV-2 virus sequences based on alignment-free methods for RNA sequence comparison. In particular, we verify a given clustering result for the GISAID data set, which was obtained analyzing the molecular differences in coronavirus populations by phylogenetic trees. For this purpose, we use alignment-free dissimilarity measures for sequences and combine them with learning vector quantization classifiers for virus type discriminant analysis and classification. Those vector quantizers belong to the class of interpretable machine learning methods, which, on the one hand side provide additional knowledge about the classification decisions like discriminant feature correlations, and on the other hand can be equipped with a reject option. This option gives the model the property of self controlled evidence if applied to new data, i.e. the models refuses to make a classification decision, if the model evidence for the presented data is not given. After training such a classifier for the GISAID data set, we apply the obtained classifier model to another but unlabeled SARS-CoV-2 virus data set. On the one hand side, this allows us to assign new sequences to already known virus types and, on the other hand, the rejected sequences allow speculations about new virus types with respect to nucleotide base mutations in the viral sequences. Author summaryThe currently emerging global disease COVID-19 caused by novel SARS-CoV-2 viruses requires all scientific effort to investigate the development of the viral epidemy, the 1/24 properties of the virus and its types. Investigations of the virus sequence are of special interest. Frequently, those are based on mathematical/statistical analysis. However, machine learning methods represent a promising alternative, if one focuses on interpretable models, i.e. those that do not act as black-boxes. Doing so, we apply variants of Learning Vector Quantizers to analyze the SARS-CoV-2 sequences. We encoded the sequences and compared them in their numerical representations to avoid the computationally costly comparison based on sequence alignments. Our resulting model is interpretable, robust, efficient, and has a self-controlling mechanism regarding the applicability to data. This framework was applied to two data sets concerning SARS-CoV-2. We were able to verify previously published virus type findings for one of the data sets by training our model to accurately identify the virus type of sequences. For sequences without virus type information (second data set), our trained model can predict them. Thereby, we observe a new scattered spreading of the sequences in the data space which probably is caused by mutations in the viral sequences. 25Frequent mutations in SARS-CoV-2 genomes are in the genes encoding the S-protein 26 and RNA polymerase, RNA primase, and nucleoprotein. Applying a sequence alignment 27 and similarity comparison using the Jaccard index, a method for monitoring and tracing 28 SARS-CoV-2 mutations was established in [17]. However, a g...

show abstract

supporting

confidence: 81%

“…Using such methods for the SARS-CoV-2 sequence data, first we 60 verify the classification results for the GISAID-data. In particular, we classify the 61 sequences by a learning vector quantizer, which is proven to be robust and 62 interpretable [45,47]. Thereafter, we use this model to classify the new data from the 63 NCBI.…”

mentioning

confidence: 99%

Analysis of SARS-CoV-2 RNA-Sequences by Interpretable Machine Learning Models

Kaden

Bohnsack

Weber

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

show abstract

“…In particular, a classification correlation matrix is delivered, which describes correlations between data [5,56]. GMLVQ is known to be robust and easy to interpret [46,58]. In biomedical context it was successfully applied to analyse flow cytometry data and to detect early folding residues during protein folding [4,7].…”

Section: /14mentioning

confidence: 99%

Detection of native and mirror protein structures based on Ramachandran plot analysis by interpretable machine learning models

Villmann

Abel

Bohnsack

et al. 2020

Preprint

Self Cite

View full text Add to dashboard Cite

In this contribution the discrimination between native and mirror models of proteins according to their chirality is tackled based on the structural protein information. This information is contained in the Ramachandran plots of the protein models. We provide an approach to classify those plots by means of an interpretable machine learning classifier - the Generalized Matrix Learning Vector Quantizer. Applying this tool, we are able to distinguish with high accuracy between mirror and native structures just evaluating the Ramachandran plots. The classifier model provides additional information regarding the importance of regions, e.g. α-helices and β-strands, to discriminate the structures precisely. This importance weighting differs for several considered protein classes.

show abstract

“…Of particular interest for potential users is the advantage of easy interpretability of LVQ networks according to the prototype principle [63]. Further, LVQ networks belong to the class of classification margin optimizers like SVM [10] and are proven to be robust against adversarial attacks [41].…”

Section: Introductionmentioning

confidence: 99%

“…As mentioned before, this standard GLVQ constitutes a margin classifier and is robust against adversarial attacks [10,41]. If the expression…”

Section: Introductionmentioning

confidence: 99%

Quantum-inspired learning vector quantizers for prototype-based classification

Villmann

Engelsberger

Ravichandran

et al. 2020

Neural Comput & Applic

Self Cite

View full text Add to dashboard Cite

Prototype-based models like the Generalized Learning Vector Quantization (GLVQ) belong to the class of interpretable classifiers. Moreover, quantum-inspired methods get more and more into focus in machine learning due to its potential efficient computing. Further, its interesting mathematical perspectives offer new ideas for alternative learning scenarios. This paper proposes a quantum computing-inspired variant of the prototype-based GLVQ for classification learning. We start considering kernelized GLVQ with real- and complex-valued kernels and their respective feature mapping. Thereafter, we explain how quantum space ideas could be integrated into a GLVQ using quantum bit vector space in the quantum state space $${\mathcal {H}}^{n}$$ H n and show the relations to kernelized GLVQ. In particular, we explain the related feature mapping of data into the quantum state space $${\mathcal {H}}^{n}$$ H n . A key feature for this approach is that $${\mathcal {H}}^{n}$$ H n is an Hilbert space with particular inner product properties, which finally restrict the prototype adaptations to be unitary transformations. The resulting approach is denoted as Qu-GLVQ. We provide the mathematical framework and give exemplary numerical results.

show abstract

Robustness of Generalized Learning Vector Quantization Models Against Adversarial Attacks

Cited by 17 publications

References 11 publications

Analysis of SARS-CoV-2 RNA-Sequences by Interpretable Machine Learning Models

Analysis of SARS-CoV-2 RNA-Sequences by Interpretable Machine Learning Models

Detection of native and mirror protein structures based on Ramachandran plot analysis by interpretable machine learning models

Quantum-inspired learning vector quantizers for prototype-based classification

Contact Info

Product

Resources

About