Methods for interpreting and understanding deep neural networks

Montavon, Gré,goire,Samek, Wojciech,Mü,ller, Klaus-Robert Elsevier 2018 Digital signal processing Vol.73 No.-

<P><B>Abstract</B></P> <P>This paper provides an entry point to the problem of interpreting a deep neural network model and explaining its predictions. It is based on a tutorial given at ICASSP 2017. As a tutorial paper, the set of methods covered here is not exhaustive, but sufficiently representative to discuss a number of questions in interpretability, technical challenges, and possible applications. The second part of the tutorial focuses on the recently proposed layer-wise relevance propagation (LRP) technique, for which we provide theory, recommendations, and tricks, to make most efficient use of it on real data.</P>

Explaining nonlinear classification decisions with deep Taylor decomposition

Montavon, Gré,goire,Lapuschkin, Sebastian,Binder, Alexander,Samek, Wojciech,Mü,ller, Klaus-Robert Elsevier 2017 Pattern recognition Vol.65 No.-

<P><B>Abstract</B></P> <P>Nonlinear methods such as Deep Neural Networks (DNNs) are the gold standard for various challenging machine learning problems such as image recognition. Although these methods perform impressively well, they have a significant disadvantage, the lack of transparency, limiting the interpretability of the solution and thus the scope of application in practice. Especially DNNs act as black boxes due to their multilayer nonlinear structure. In this paper we introduce a novel methodology for interpreting generic multilayer neural networks by decomposing the network classification decision into contributions of its input elements. Although our focus is on image classification, the method is applicable to a broad set of input data, learning tasks and network architectures. Our method called deep Taylor decomposition efficiently utilizes the structure of the network by backpropagating the explanations from the output to the input layer. We evaluate the proposed method empirically on the MNIST and ILSVRC data sets.</P> <P><B>Highlights</B></P> <P> <UL> <LI> A novel method to explain nonlinear classification decisions in terms of input variables is introduced. </LI> <LI> The method is based on Taylor expansions and decomposes the output of a deep neural network in terms of input variables. </LI> <LI> The resulting deep Taylor decomposition can be applied directly to existing neural networks without retraining. </LI> <LI> The method is tested on two large-scale neural networks for image classification: BVLC CaffeNet and GoogleNet. </LI> </UL> </P>

An additive manufacturing oriented design approach to mechanical assemblies

Sossou, Germain,Demoly, Frederic,Montavon, Ghislain,Gomes, Samuel Society for Computational Design and Engineering 2018 Journal of computational design and engineering Vol.5 No.1

Firstly introduced as a prototyping process, additive manufacturing (AM) is being more and more considered as a fully-edged manufacturing process. The number of AM processes, along with the range of processed materials are expanding. AM has made manufacturable shapes that were too difficult (or even impossible) to manufacture with conventional technologies. This has promoted a shift in engineering design, from conventional design for manufacturing and assembly to design for additive manufacturing (DFAM). Research efforts into the DFAM field have been mostly dedicated to part's design, which is actually a requirement for a better industrial adoption. This has given rise to topologically optimized and/or latticed designs. However, since AM is also capable of manufacturing fully functional assemblies requiring a few or no assembly operations, there is a need for DFAM methodologies tackling product's development more holistically, and which are, therefore, dedicated to assembly design. Considering all the manufacturing issues related to AM of assembly-free mechanisms and available post-processing capabilities, this paper proposes a top-down assembly design methodology for AM in a proactive manner. Such an approach, can be seen as the beginning of a shift from conventional design for assembly (DFA) to a new paradigm. From a product's concept and a selected AM technology, the approach first provides assistance in the definition of the product architecture so that both functionality and successful manufacturing (including post-processing) are ensured. Particularly, build-orientation and downstream processes' characteristics are taken into account early in the design process. Secondly, for the functional flow (energy, material, signal) to be appropriately conveyed by the right amount of matter, the methodology provides guidance into how the components can be designed in a minimalism fashion leveraging the shape complexity afforded by AM. A mechanical assembly as case study is presented to illustrate the DFAM methodology. It is found that clearances and material (be it raw unprocessed material or support structures) within them plays a pivotal role in a successful assembly's design to be additively manufactured. In addition, the methodology for components' design proves to be an efficient alternative to topology optimization. Though, the approach can be extended by considering a strategy for part consolidation and the possibility to manufacture the assemblies with more than one AM process. As regards components' design, considering anisotropy can also improved the approach.

Evaluating the Visualization of What a Deep Neural Network Has Learned

Samek, Wojciech,Binder, Alexander,Montavon, Gregoire,Lapuschkin, Sebastian,Muller, Klaus-Robert IEEE 2017 IEEE transactions on neural networks and learning Vol.28 No.11

<P>Deep neural networks (DNNs) have demonstrated impressive performance in complex machine learning tasks such as image classification or speech recognition. However, due to their multilayer nonlinear structure, they are not transparent, i.e., it is hard to grasp what makes them arrive at a particular classification or recognition decision, given a new unseen data sample. Recently, several approaches have been proposed enabling one to understand and interpret the reasoning embodied in a DNN for a single test image. These methods quantify the 'importance' of individual pixels with respect to the classification decision and allow a visualization in terms of a heatmap in pixel/input space. While the usefulness of heatmaps can be judged subjectively by a human, an objective quality measure is missing. In this paper, we present a general methodology based on region perturbation for evaluating ordered collections of pixels such as heatmaps. We compare heatmaps computed by three different methods on the SUN397, ILSVRC2012, and MIT Places data sets. Our main result is that the recently proposed layer-wise relevance propagation algorithm qualitatively and quantitatively provides a better explanation of what made a DNN arrive at a particular classification decision than the sensitivity-based approach or the deconvolution method. We provide theoretical arguments to explain this result and discuss its practical implications. Finally, we investigate the use of heatmaps for unsupervised assessment of the neural network performance.</P>

An additive manufacturing oriented design approach to mechanical assemblies

Germain Sossou,Frédéric Demoly,Ghislain Montavon,Samuel Gomes 한국CDE학회 2018 Journal of computational design and engineering Vol.5 No.1

Firstly introduced as a prototyping process, additive manufacturing (AM) is being more and more considered as a fully-edged manufacturing process. The number of AM processes, along with the range of processed materials are expanding. AM has made manufacturable shapes that were too difficult (or even impossible) to manufacture with conventional technologies. This has promoted a shift in engineering design, from conventional design for manufacturing and assembly to design for additive manufacturing (DFAM). Research efforts into the DFAM field have been mostly dedicated to part’s design, which is actually a requirement for a better industrial adoption. This has given rise to topologically optimized and/or latticed designs. However, since AM is also capable of manufacturing fully functional assemblies requiring a few or no assembly operations, there is a need for DFAM methodologies tackling product’s development more holistically, and which are, therefore, dedicated to assembly design. Considering all the manufacturing issues related to AM of assembly-free mechanisms and available post-processing capabilities, this paper proposes a top-down assembly design methodology for AM in a proactive manner. Such an approach, can be seen as the beginning of a shift from conventional design for assembly (DFA) to a new paradigm. From a product’s concept and a selected AM technology, the approach first provides assistance in the definition of the product architecture so that both functionality and successful manufacturing (including post-processing) are ensured. Particularly, build-orientation and downstream processes’ characteristics are taken into account early in the design process. Secondly, for the functional flow (energy, material, signal) to be appropriately conveyed by the right amount of matter, the methodology provides guidance into how the components can be designed in a minimalism fashion leveraging the shape complexity afforded by AM. A mechanical assembly as case study is presented to illustrate the DFAM methodology. It is found that clearances and material (be it raw unprocessed material or support structures) within them plays a pivotal role in a successful assembly’s design to be additively manufactured. In addition, the methodology for components’ design proves to be an efficient alternative to topology optimization. Though, the approach can be extended by considering a strategy for part consolidation and the possibility to manufacture the assemblies with more than one AM process. As regards components’ design, considering anisotropy can also improved the approach.

Machine learning analysis of DNA methylation profiles distinguishes primary lung squamous cell carcinomas from head and neck metastases

Jurmeister, Philipp,Bockmayr, Michael,Seegerer, Philipp,Bockmayr, Teresa,Treue, Denise,Montavon, Gré,goire,Vollbrecht, Claudia,Arnold, Alexander,Teichmann, Daniel,Bressem, Keno,Schu¨ller, Ulrich American Association for the Advancement of Scienc 2019 Science translational medicine Vol.11 No.509

<P>Head and neck squamous cell carcinoma (HNSC) patients are at risk of suffering from both pulmonary metastases or a second squamous cell carcinoma of the lung (LUSC). Differentiating pulmonary metastases from primary lung cancers is of high clinical importance, but not possible in most cases with current diagnostics. To address this, we performed DNA methylation profiling of primary tumors and trained three different machine learning methods to distinguish metastatic HNSC from primary LUSC. We developed an artificial neural network that correctly classified 96.4% of the cases in a validation cohort of 279 patients with HNSC and LUSC as well as normal lung controls, outperforming support vector machines (95.7%) and random forests (87.8%). Prediction accuracies of more than 99% were achieved for 92.1% (neural network), 90% (support vector machine), and 43% (random forest) of these cases by applying thresholds to the resulting probability scores and excluding samples with low confidence. As independent clinical validation of the approach, we analyzed a series of 51 patients with a history of HNSC and a second lung tumor, demonstrating the correct classifications based on clinicopathological properties. In summary, our approach may facilitate the reliable diagnostic differentiation of pulmonary metastases of HNSC from primary LUSC to guide therapeutic decisions.</P>