diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index f5fbd306f5abd134ae236ee86f74f8272c676315..1a4b0f780473838fd8f821e6d94646a1048f1c73 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -21,11 +21,11 @@ build_results:
build_talk:
image: armagetron/ubuntu-mpi:22.10
script:
- - make -C src/slides
+ - make -C src/thesis_presentation
stage: build
artifacts:
paths:
- - src/slides/main.pdf
+ - src/thesis_presentation/main.pdf
expire_in: 1 week
allow_failure: false
build_thesis:
diff --git a/src/slides/GUM.bib b/src/slides/GUM.bib
deleted file mode 100644
index cede686fd57c286f36f67c4877f78acfb4856ce3..0000000000000000000000000000000000000000
--- a/src/slides/GUM.bib
+++ /dev/null
@@ -1,351 +0,0 @@
-
-@misc{jcgm_guide_2020,
- title = {Guide to the expression of uncertainty in measurement - {Part} 6: {Developing} and using measurement models},
- url = {https://www.bipm.org/documents/20126/2071204/JCGM_GUM_6_2020.pdf/d4e77d99-3870-0908-ff37-c1b6a230a337?version=1.3&download=true},
- language = {english},
- author = {{JCGM}},
- year = {2020},
- file = {Gum - Guide to the expression of uncertainty in measurem.pdf:/home/bjorn/Zotero/storage/BHWWFVP2/Gum - Guide to the expression of uncertainty in measurem.pdf:application/pdf},
-}
-
-@article{ji_uncertainty_2020,
- title = {Uncertainty {Propagation} in {Deep} {Neural} {Network} {Using} {Active} {Subspace}},
- url = {http://arxiv.org/abs/1903.03989},
- abstract = {The inputs of deep neural network (DNN) from real-world data usually come with uncertainties. Yet, it is challenging to propagate the uncertainty in the input features to the DNN predictions at a low computational cost. This work employs a gradient-based subspace method and response surface technique to accelerate the uncertainty propagation in DNN. Specifically, the active subspace method is employed to identify the most important subspace in the input features using the gradient of the DNN output to the inputs. Then the response surface within that low-dimensional subspace can be efficiently built, and the uncertainty of the prediction can be acquired by evaluating the computationally cheap response surface instead of the DNN models. In addition, the subspace can help explain the adversarial examples. The approach is demonstrated in MNIST datasets with a convolutional neural network. Code is available at: https://github.com/jiweiqi/nnsubspace.},
- urldate = {2021-11-14},
- journal = {arXiv:1903.03989 [cs, stat]},
- author = {Ji, Weiqi and Ren, Zhuyin and Law, Chung K.},
- month = jan,
- year = {2020},
- note = {arXiv: 1903.03989},
- keywords = {Computer Science - Machine Learning, Statistics - Machine Learning, Computer Science - Computer Vision and Pattern Recognition},
- file = {Ji et al. - 2020 - Uncertainty Propagation in Deep Neural Network Usi.pdf:/home/bjorn/Zotero/storage/RMPDWPYQ/Ji et al. - 2020 - Uncertainty Propagation in Deep Neural Network Usi.pdf:application/pdf},
-}
-
-@inproceedings{abdelaziz_uncertainty_2015,
- title = {Uncertainty propagation through deep neural networks},
- url = {https://www.isca-speech.org/archive/interspeech_2015/abdelaziz15_interspeech.html},
- doi = {10.21437/Interspeech.2015-706},
- abstract = {In order to improve the ASR performance in noisy environments, distorted speech is typically pre-processed by a speech enhancement algorithm, which usually results in a speech estimate containing residual noise and distortion. We may also have some measures of uncertainty or variance of the estimate. Uncertainty decoding is a framework that utilizes this knowledge of uncertainty in the input features during acoustic model scoring. Such frameworks have been well explored for traditional probabilistic models, but their optimal use for deep neural network (DNN)-based ASR systems is not yet clear. In this paper, we study the propagation of observation uncertainties through the layers of a DNN-based acoustic model. Since this is intractable due to the nonlinearities of the DNN, we employ approximate propagation methods, including Monte Carlo sampling, the unscented transform, and the piecewise exponential approximation of the activation function, to estimate the distribution of acoustic scores. Finally, the expected value of the acoustic score distribution is used for decoding, which is shown to further improve the ASR accuracy on the CHiME database, relative to a highly optimized DNN baseline.},
- language = {en},
- urldate = {2021-11-14},
- booktitle = {Interspeech 2015},
- publisher = {ISCA},
- author = {Abdelaziz, Ahmed Hussen and Watanabe, Shinji and Hershey, John R. and Vincent, Emmanuel and Kolossa, Dorothea},
- month = sep,
- year = {2015},
- pages = {3561--3565},
- file = {Abdelaziz et al. - 2015 - Uncertainty propagation through deep neural networ.pdf:/home/bjorn/Zotero/storage/K6WK4DP2/Abdelaziz et al. - 2015 - Uncertainty propagation through deep neural networ.pdf:application/pdf},
-}
-
-@article{jcgm_evaluation_2013,
- title = {Evaluation of {Measurement} {Data}: {The} {Role} of {Measurement} {Uncertainty} in {Conformity} {Assessment}},
- volume = {35},
- issn = {1365-2192, 0193-6484},
- shorttitle = {Evaluation of {Measurement} {Data}},
- url = {https://www.degruyter.com/document/doi/10.1515/ci.2013.35.2.22/html},
- doi = {10.1515/ci.2013.35.2.22},
- language = {english},
- number = {2},
- urldate = {2021-11-14},
- journal = {Chemistry International -- Newsmagazine for IUPAC},
- author = {{JCGM}},
- month = jan,
- year = {2013},
- file = {2013 - Evaluation of Measurement Data The Role of Measur.pdf:/home/bjorn/Zotero/storage/ZJFPAXMH/2013 - Evaluation of Measurement Data The Role of Measur.pdf:application/pdf},
-}
-
-@misc{jcgm_evaluation_2008,
- title = {Evaluation of measurement data — {Guide} to the expression of uncertainty in measurement},
- url = {https://www.bipm.org/documents/20126/2071204/JCGM_100_2008_E.pdf/cb0ef43f-baa5-11cf-3f85-4dcd86f77bd6?version=1.7&download=true},
- language = {english},
- publisher = {JCGM},
- author = {{JCGM}},
- year = {2008},
- file = {JCGM_100_2008_E.pdf:/home/bjorn/Zotero/storage/TYSRP7RL/JCGM_100_2008_E.pdf:application/pdf},
-}
-
-@misc{jcgm_evaluation_2008-1,
- title = {Evaluation of measurement data — {Supplement} 1 to the “{Guide} to the expression of uncertainty in measurement†— {Propagation} of distributions using a {Monte} {Carlo} method},
- url = {https://www.bipm.org/documents/20126/2071204/JCGM_101_2008_E.pdf/325dcaad-c15a-407c-1105-8b7f322d651c?version=1.5&download=true},
- language = {english},
- publisher = {JCGM},
- author = {{JCGM}},
- year = {2008},
- file = {JCGM_101_2008_E.pdf:/home/bjorn/Zotero/storage/P7BHF3LG/JCGM_101_2008_E.pdf:application/pdf},
-}
-
-@misc{jcgm_evaluation_2011,
- title = {Evaluation of measurement data – {Supplement} 2 to the “{Guide} to the expression of uncertainty in measurement†– {Extension} to any number of output quantities},
- url = {https://www.bipm.org/documents/20126/2071204/JCGM_102_2011_E.pdf/6a3281aa-1397-d703-d7a1-a8d58c9bf2a5?version=1.4&download=true},
- language = {english},
- publisher = {JCGM},
- author = {JCGM},
- year = {2011},
- file = {JCGM_102_2011_E.pdf:/home/bjorn/Zotero/storage/7XIHF4MP/JCGM_102_2011_E.pdf:application/pdf},
-}
-
-@misc{jcgm_evaluation_2009,
- title = {Evaluation of measurement data — {An} introduction to the “{Guide} to the expression of uncertainty in measurement†and related documents},
- url = {https://www.bipm.org/documents/20126/2071204/JCGM_104_2009.pdf/19e0a96c-6cf3-a056-4634-4465c576e513?version=1.8&download=true},
- language = {english},
- publisher = {JCGM},
- author = {{JCGM}},
- year = {2009},
- file = {JCGM_104_2009.pdf:/home/bjorn/Zotero/storage/XP4N9UKD/JCGM_104_2009.pdf:application/pdf},
-}
-
-@misc{amh_van_der_veen_compendium_2020,
- title = {Compendium of examples: good practice in evaluating measurement uncertainty},
- copyright = {Creative Commons Attribution 4.0 International, Open Access},
- shorttitle = {Compendium of examples},
- url = {https://zenodo.org/record/5142180},
- abstract = {This document illustrates good practice in the evaluation of measurement uncertainty. It contains examples from a variety of areas in calibration and testing, and illustrates the use of the methods from the “Guide to the expression of Uncertainty in Measurement†and its supplements, as well as Bayesian approaches. Compendium\_M36.pdf: document},
- language = {en},
- urldate = {2021-12-06},
- publisher = {Zenodo},
- author = {A.M.H Van Der Veen and M.G. Cox and J. Greenwood and A. Bošnjakovic and V. Karahodžic and S. Martens and K. Klauenberg and C. Elster and S. Demeyer and N. Fischer and J.A. Sousa and O. Pellegrino and L.L. Martins and A.S. Ribeiro and D. Loureiro and M.C. Almeida and M.A. Silva and R. Brito and A.C. Soares and K. Shirono and F. Pennecchi and P.M. Harris and S.L.R. Ellison and F. Rolle and A. Alard and T. Caebergs and B. De Boeck and J. Pétry and N. Sebaïhi and P. Pedone and F. Manta and M. Sega and P.G. Spazzini and I. De Krom and M. Singh and T. Gardiner and R. Robinson and T. Smith and T. Arnold and M. Reader-Harris and C. Forsyth and T. Boussouara and B. Mickan and C. Yardin and M.ˇCauševic and A. Arduino and L. Zilberti and U. Katscher and J. Neukammer and S. Cowen and A. Furtado and J. Pereira and E. Batista and J. Dawkins and J. Gillespie and T. Lowe and W. Ng and J. Roberts and M. Griepentrog and A. Germak and O. Barroso and A. Danion and B. Garrido and S. Westwood and A. Carullo and S. Corbellini and A. Vallan},
- month = nov,
- year = {2020},
- doi = {10.5281/ZENODO.5142180},
- note = {Version Number: 2
-Type: dataset},
- keywords = {Measurement uncertainty, calibration, testing, GUM, conformity assessment, Bayesian inference},
- file = {Compendium_M36.pdf:/home/bjorn/Zotero/storage/2STVQQ62/Compendium_M36.pdf:application/pdf},
-}
-
-@article{farrance_uncertainty_nodate,
- title = {Uncertainty of {Measurement}: {A} {Review} of the {Rules} for {Calculating} {Uncertainty} {Components} through {Functional} {Relationships}},
- abstract = {The Evaluation of Measurement Data - Guide to the Expression of Uncertainty in Measurement (usually referred to as the GUM) provides general rules for evaluating and expressing uncertainty in measurement. When a measurand, y, is calculated from other measurements through a functional relationship, uncertainties in the input variables will propagate through the calculation to an uncertainty in the output y. The manner in which such uncertainties are propagated through a functional relationship provides much of the mathematical challenge to fully understanding the GUM.},
- language = {en},
- author = {Farrance, Ian and Frenkel, Robert},
- pages = {28},
- file = {Farrance and Frenkel - Uncertainty of Measurement A Review of the Rules .pdf:/home/bjorn/Zotero/storage/SQ8LDUDN/Farrance and Frenkel - Uncertainty of Measurement A Review of the Rules .pdf:application/pdf},
-}
-
-@incollection{szewczyk_uncertainty_2021,
- address = {Cham},
- title = {Uncertainty {Bands} of the {Regression} {Line} for {Data} with {Type} {A} and {Type} {B} {Uncertainties} of {Dependent} {Variable} {Y}},
- volume = {1390},
- isbn = {978-3-030-74892-0 978-3-030-74893-7},
- url = {https://link.springer.com/10.1007/978-3-030-74893-7_32},
- abstract = {This work concerns on the estimation of the accuracy of function determined by the linear regression method for the description of noncorrelated measured data of Y. Recommendations of the international Guide to the Expression of Uncertainty in Measurement (GUM) are used. The impact of Type B measurement uncertainties is included, which is omitted in the statistical literature on the accuracy of regression method. The introduction presents the essence of the uncertainty calculations used in GUM. The case of random changes of variable Y and the criteria used in linear regression are discussed in detail. For known values of X variable, the parameters and uncertainty bands of regression line are determined for measurements of uncorrelated values of Y with Type A and Type B uncertainties. Considerations are illustrated with four numerical examples of the measurement points with the same coordinates, but different absolute and relative uncertainties.},
- language = {en},
- urldate = {2021-12-06},
- booktitle = {Automation 2021: {Recent} {Achievements} in {Automation}, {Robotics} and {Measurement} {Techniques}},
- publisher = {Springer International Publishing},
- author = {Warsza, Zygmunt Lech and Puchalski, Jacek},
- editor = {Szewczyk, Roman and Zieliński, Cezary and Kaliczyńska, Małgorzata},
- year = {2021},
- doi = {10.1007/978-3-030-74893-7_32},
- note = {Series Title: Advances in Intelligent Systems and Computing},
- pages = {342--363},
- file = {Warsza and Puchalski - 2021 - Uncertainty Bands of the Regression Line for Data .pdf:/home/bjorn/Zotero/storage/SIFTTC7S/Warsza and Puchalski - 2021 - Uncertainty Bands of the Regression Line for Data .pdf:application/pdf},
-}
-
-@article{shekhovtsov_feed-forward_2019,
- title = {{FEED}-{FORWARD} {PROPAGATION} {IN} {PROBABILISTIC} {NEURAL} {NETWORKS} {WITH} {CATEGORICAL} {AND} {MAX} {LAYERS}},
- abstract = {Probabilistic Neural Networks deal with various sources of stochasticity: input noise, dropout, stochastic neurons, parameter uncertainties modeled as random variables, etc. In this paper we revisit a feed-forward propagation approach that allows one to estimate for each neuron its mean and variance w.r.t. all mentioned sources of stochasticity. In contrast, standard NNs propagate only point estimates, discarding the uncertainty. Methods propagating also the variance have been proposed by several authors in different context. The view presented here attempts to clarify the assumptions and derivation behind such methods, relate them to classical NNs and broaden their scope of applicability. The main technical contributions are new approximations for the distributions of argmax and max-related transforms, which allow for fully analytic uncertainty propagation in networks with softmax and max-pooling layers as well as leaky ReLU activations. We evaluate the accuracy of the approximation and suggest a simple calibration. Applying the method to networks with dropout allows for faster training and gives improved test likelihoods without the need of sampling.},
- language = {en},
- author = {Shekhovtsov, Alexander and Flach, Boris},
- year = {2019},
- pages = {21},
- file = {Shekhovtsov and Flach - 2019 - FEED-FORWARD PROPAGATION IN PROBABILISTIC NEURAL N.pdf:/home/bjorn/Zotero/storage/U7D5HXVC/Shekhovtsov and Flach - 2019 - FEED-FORWARD PROPAGATION IN PROBABILISTIC NEURAL N.pdf:application/pdf},
-}
-
-@inproceedings{postels_sampling-free_2019,
- address = {Seoul, Korea (South)},
- title = {Sampling-{Free} {Epistemic} {Uncertainty} {Estimation} {Using} {Approximated} {Variance} {Propagation}},
- isbn = {978-1-72814-803-8},
- url = {https://ieeexplore.ieee.org/document/9010684/},
- doi = {10.1109/ICCV.2019.00302},
- abstract = {We present a sampling-free approach for computing the epistemic uncertainty of a neural network. Epistemic uncertainty is an important quantity for the deployment of deep neural networks in safety-critical applications, since it represents how much one can trust predictions on new data. Recently promising works were proposed using noise injection combined with Monte-Carlo (MC) sampling at inference time to estimate this quantity (e.g. MC dropout). Our main contribution is an approximation of the epistemic uncertainty estimated by these methods that does not require sampling, thus notably reducing the computational overhead. We apply our approach to large-scale visual tasks (i.e., semantic segmentation and depth regression) to demonstrate the advantages of our method compared to sampling-based approaches in terms of quality of the uncertainty estimates as well as of computational overhead.},
- language = {en},
- urldate = {2021-12-12},
- booktitle = {2019 {IEEE}/{CVF} {International} {Conference} on {Computer} {Vision} ({ICCV})},
- publisher = {IEEE},
- author = {Postels, Janis and Ferroni, Francesco and Coskun, Huseyin and Navab, Nassir and Tombari, Federico},
- month = oct,
- year = {2019},
- pages = {2931--2940},
- file = {Postels et al. - 2019 - Sampling-Free Epistemic Uncertainty Estimation Usi.pdf:/home/bjorn/Zotero/storage/7SGGGG5Z/Postels et al. - 2019 - Sampling-Free Epistemic Uncertainty Estimation Usi.pdf:application/pdf},
-}
-
-@inproceedings{astudillo_propagation_2011,
- title = {Propagation of uncertainty through multilayer perceptrons for robust automatic speech recognition},
- url = {https://www.isca-speech.org/archive/interspeech_2011/astudillo11_interspeech.html},
- doi = {10.21437/Interspeech.2011-196},
- abstract = {Observation uncertainty techniques offer a way to dynamically compensate automatic speech recognizers to account for the information missing in real world scenarios. These techniques have been demonstrated to effectively be able to compensate multiple environment distortions and improve the integration of ASR systems with speech enhancement pre-processing through uncertainty propagation. Unfortunately observation uncertainty techniques rely on statistical methods and as such are limited to GMM-HMM architectures. In this paper we explore the application of observation uncertainty and uncertainty propagation techniques to multi-layer perceptrons (MLPs). We develop solutions for propagation through a generic MLP and exemplify potential gains with an large vocabulary robust ASR experiment on the AURORA4 database using an Hybrid MLP-HMM recognizer.},
- language = {en},
- urldate = {2021-12-12},
- booktitle = {Interspeech 2011},
- publisher = {ISCA},
- author = {Astudillo, Ramón Fernandez and Neto, João Paulo da Silva},
- month = aug,
- year = {2011},
- pages = {461--464},
- file = {Astudillo and Neto - 2011 - Propagation of uncertainty through multilayer perc.pdf:/home/bjorn/Zotero/storage/UF3CLDAK/Astudillo and Neto - 2011 - Propagation of uncertainty through multilayer perc.pdf:application/pdf},
-}
-
-@article{jimenez_experimental_2017,
- title = {Experimental {Approach} for the {Uncertainty} {Assessment} of {3D} {Complex} {Geometry} {Dimensional} {Measurements} {Using} {Computed} {Tomography} at the mm and {Sub}-mm {Scales}},
- volume = {17},
- issn = {1424-8220},
- url = {http://www.mdpi.com/1424-8220/17/5/1137},
- doi = {10.3390/s17051137},
- abstract = {The dimensional veriï¬cation of miniaturized components with 3D complex geometries is particularly challenging. Computed Tomography (CT) can represent a suitable alternative solution to micro metrology tools based on optical and tactile techniques. However, the establishment of CT systems’ traceability when measuring 3D complex geometries is still an open issue. In this work, an alternative method for the measurement uncertainty assessment of 3D complex geometries by using CT is presented. The method is based on the micro-CT system Maximum Permissible Error (MPE) estimation, determined experimentally by using several calibrated reference artefacts. The main advantage of the presented method is that a previous calibration of the component by a more accurate Coordinate Measuring System (CMS) is not needed. In fact, such CMS would still hold all the typical limitations of optical and tactile techniques, particularly when measuring miniaturized components with complex 3D geometries and their inability to measure inner parts. To validate the presented method, the most accepted standard currently available for CT sensors, the Verein Deutscher Ingenieure/Verband Deutscher Elektrotechniker (VDI/VDE) guideline 2630-2.1 is applied. Considering the high number of influence factors in CT and their impact on the measuring result, two different techniques for surface extraction are also considered to obtain a realistic determination of the influence of data processing on uncertainty. The uncertainty assessment of a workpiece used for micro mechanical material testing is ï¬rstly used to conï¬rm the method, due to its feasible calibration by an optical CMS. Secondly, the measurement of a miniaturized dental ï¬le with 3D complex geometry is carried out. The estimated uncertainties are eventually compared with the component’s calibration and the micro manufacturing tolerances to demonstrate the suitability of the presented CT calibration procedure. The 2U/T ratios resulting from the validation workpiece are, respectively, 0.27 (VDI) and 0.35 (MPE), by assuring tolerances in the range of ± 20–30 µm. For the dental ï¬le, the EN {\textless} 1 value analysis is favorable in the majority of the cases (70.4\%) and 2U/T is equal to 0.31 for sub-mm measurands (L {\textless} 1 mm and tolerance intervals of ± 40–80 µm).},
- language = {en},
- number = {5},
- urldate = {2022-01-09},
- journal = {Sensors},
- author = {Jiménez, Roberto and Torralba, Marta and Yagüe-Fabra, José and Ontiveros, Sinué and Tosello, Guido},
- month = may,
- year = {2017},
- pages = {1137},
- file = {Jiménez et al. - 2017 - Experimental Approach for the Uncertainty Assessme.pdf:/home/bjorn/Zotero/storage/67KSNL94/Jiménez et al. - 2017 - Experimental Approach for the Uncertainty Assessme.pdf:application/pdf},
-}
-
-@article{dey_regularizing_2018,
- title = {Regularizing {Multilayer} {Perceptron} for {Robustness}},
- volume = {48},
- issn = {2168-2216, 2168-2232},
- url = {https://ieeexplore.ieee.org/document/7862272/},
- doi = {10.1109/TSMC.2017.2664143},
- abstract = {The weights of a multi-layer perceptron (MLP) may be altered by multiplicative and/or additive noises if it is implemented in hardware. Moreover, if an MLP is implemented using analog circuits, it is prone to stuck-at 0 faults, i.e., link failures. In this paper, we have proposed a methodology for making an MLP robust with respect to link failures, multiplicative noise, and additive noise. This is achieved by penalizing the system error with three regularizing terms. To train the system we use a weighted sum of the following four terms: (i) mean squared error (MSE), (ii) l 2 norm of the weight vector, (iii) sum of squares of the ï¬rst order derivatives of MSE with respect to weights, and (iv) sum of squares of the second order derivatives of MSE with respect to weights. The proposed approach has been tested on ten regression and ten classiï¬cation tasks with link failure, multiplicative noise, and additive noise scenarios. Our experimental results demonstrate the effectiveness of the proposed regularization to achieve robust training of an MLP.},
- language = {en},
- number = {8},
- urldate = {2022-01-09},
- journal = {IEEE Trans. Syst. Man Cybern, Syst.},
- author = {Dey, Prasenjit and Nag, Kaustuv and Pal, Tandra and Pal, Nikhil R.},
- month = aug,
- year = {2018},
- pages = {1255--1266},
- file = {Dey et al. - 2018 - Regularizing Multilayer Perceptron for Robustness.pdf:/home/bjorn/Zotero/storage/NB97RM8W/Dey et al. - 2018 - Regularizing Multilayer Perceptron for Robustness.pdf:application/pdf},
-}
-
-@article{brown_methodology_1998,
- title = {A {Methodology} for {Determining} {Experimental} {Uncertainties} in {Regressions}},
- volume = {120},
- issn = {0098-2202, 1528-901X},
- url = {https://asmedigitalcollection.asme.org/fluidsengineering/article/120/3/445/431506/A-Methodology-for-Determining-Experimental},
- doi = {10.1115/1.2820683},
- abstract = {A methodology to determine the experimental uncertainties associated with regressions is presented. When a regression model is used to represent experimental information, the uncertainty associated with the model is affected by random, systematic, and correlated systematic uncertainties associated with the experimental data. The key to the proper estimation of the uncertainty associated with a regression is a careful, comprehensive accounting of systematic and correlated systematic uncertainties. The methodology presented in this article is developed by applying uncertainty propagation techniques to the linear regression analysis equations. The effectiveness of this approach was investigated and proven using Monte Carlo simulations. The application of that methodology to the calibration of a venturi flowmeter and its subsequent use to determine flowrate in a test is demonstrated. It is shown that the previously accepted way of accounting for the contribution of discharge coefficient uncertainty to the overall flowrate uncertainty does not correctly account for all uncertainty sources, and the appropriate approach is developed, discussed, and demonstrated.},
- language = {en},
- number = {3},
- urldate = {2022-05-01},
- journal = {Journal of Fluids Engineering},
- author = {Brown, K. K. and Coleman, H. W. and Steele, W. Glenn},
- month = sep,
- year = {1998},
- pages = {445--456},
- file = {Scanauftrag_A_Methodology_for_Determining_Experimental_Uncertainties_in_Regressions.pdf:/home/bjorn/Zotero/storage/T6W4KJZR/Scanauftrag_A_Methodology_for_Determining_Experimental_Uncertainties_in_Regressions.pdf:application/pdf},
-}
-
-@misc{xie_smooth_2021,
- title = {Smooth {Adversarial} {Training}},
- url = {http://arxiv.org/abs/2006.14536},
- abstract = {It is commonly believed that networks cannot be both accurate and robust, that gaining robustness means losing accuracy. It is also generally believed that, unless making networks larger, network architectural elements would otherwise matter little in improving adversarial robustness. Here we present evidence to challenge these common beliefs by a careful study about adversarial training. Our key observation is that the widely-used ReLU activation function significantly weakens adversarial training due to its non-smooth nature. Hence we propose smooth adversarial training (SAT), in which we replace ReLU with its smooth approximations to strengthen adversarial training. The purpose of smooth activation functions in SAT is to allow it to find harder adversarial examples and compute better gradient updates during adversarial training. Compared to standard adversarial training, SAT improves adversarial robustness for "free", i.e., no drop in accuracy and no increase in computational cost. For example, without introducing additional computations, SAT significantly enhances ResNet-50's robustness from 33.0\% to 42.3\%, while also improving accuracy by 0.9\% on ImageNet. SAT also works well with larger networks: it helps EfficientNet-L1 to achieve 82.2\% accuracy and 58.6\% robustness on ImageNet, outperforming the previous state-of-the-art defense by 9.5\% for accuracy and 11.6\% for robustness. Models are available at https://github.com/cihangxie/SmoothAdversarialTraining.},
- urldate = {2022-06-11},
- publisher = {arXiv},
- author = {Xie, Cihang and Tan, Mingxing and Gong, Boqing and Yuille, Alan and Le, Quoc V.},
- month = jul,
- year = {2021},
- note = {Number: arXiv:2006.14536
-arXiv:2006.14536 [cs]},
- keywords = {Computer Science - Machine Learning, Computer Science - Neural and Evolutionary Computing, Computer Science - Computer Vision and Pattern Recognition},
- file = {Xie et al. - 2021 - Smooth Adversarial Training.pdf:/home/bjorn/Zotero/storage/4T9WURDH/Xie et al. - 2021 - Smooth Adversarial Training.pdf:application/pdf},
-}
-
-@inproceedings{hao_zheng_improving_2015,
- address = {Killarney, Ireland},
- title = {Improving deep neural networks using softplus units},
- isbn = {978-1-4799-1960-4},
- url = {http://ieeexplore.ieee.org/document/7280459/},
- doi = {10.1109/IJCNN.2015.7280459},
- abstract = {Recently, DNNs have achieved great improvement for acoustic modeling in speech recognition tasks. However, it is difï¬cult to train the models well when the depth grows. One main reason is that when training DNNs with traditional sigmoid units, the derivatives damp sharply while back-propagating between layers, which restrict the depth of model especially with insufï¬cient training data. To deal with this problem, some unbounded activation functions have been proposed to preserve sufï¬cient gradients, including ReLU and softplus. Compared with ReLU, the smoothing and nonzero properties of the in gradient makes softplus-based DNNs perform better in both stabilization and performance. However, softplus-based DNNs have been rarely exploited for the phoneme recognition task. In this paper, we explore the use of softplus units for DNNs in acoustic modeling for context-independent phoneme recognition tasks.The revised RBM pre-training and dropout strategy are also applied to improve the performance of softplus units. Experiments show that, the DNNs with softplus units get signiï¬cantly performance improvement and uses less epochs to get convergence compared to the DNNs trained with standard sigmoid units and ReLUs.},
- language = {en},
- urldate = {2022-06-11},
- booktitle = {2015 {International} {Joint} {Conference} on {Neural} {Networks} ({IJCNN})},
- publisher = {IEEE},
- author = {{Hao Zheng} and {Zhanlei Yang} and {Wenju Liu} and {Jizhong Liang} and {Yanpeng Li}},
- month = jul,
- year = {2015},
- pages = {1--4},
- file = {Hao Zheng et al. - 2015 - Improving deep neural networks using softplus unit.pdf:/home/bjorn/Zotero/storage/96ZTA3RV/Hao Zheng et al. - 2015 - Improving deep neural networks using softplus unit.pdf:application/pdf},
-}
-
-@article{goebbels_training_nodate,
- title = {Training of {ReLU} {Activated} {Multilayer} {Neural} {Networks} with {Mixed} {Integer} {Linear} {Programs}},
- abstract = {In this paper, it is demonstrated through a case study that multilayer feedforward neural networks activated by ReLU functions can in principle be trained iteratively with Mixed Integer Linear Programs (MILPs) as follows. Weights are determined with batch learning. Multiple iterations are used per batch of training data. In each iteration, the algorithm starts at the output layer and propagates information back to the ï¬rst hidden layer to adjust the weights using MILPs or Linear Programs. For each layer, the goal is to minimize the difference between its output and the corresponding target output. The target output of the last (output) layer is equal to the ground truth. The target output of a previous layer is deï¬ned as the adjusted input of the following layer. For a given layer, weights are computed by solving a MILP. Then, except for the ï¬rst hidden layer, the input values are also modiï¬ed with a MILP to better match the layer outputs to their corresponding target outputs. The method was tested and compared with Tensorflow/Keras (Adam optimizer) using two simple networks on the MNIST dataset containing handwritten digits. Accuracies of the same magnitude as with Tensorflow/Keras were achieved.},
- language = {en},
- journal = {Technical Report},
- author = {Goebbels, Steffen},
- pages = {8},
- file = {Goebbels - Training of ReLU Activated Multilayer Neural Netwo.pdf:/home/bjorn/Zotero/storage/7F87GG63/Goebbels - Training of ReLU Activated Multilayer Neural Netwo.pdf:application/pdf},
-}
-
-@misc{tjeng_evaluating_2019,
- title = {Evaluating {Robustness} of {Neural} {Networks} with {Mixed} {Integer} {Programming}},
- url = {http://arxiv.org/abs/1711.07356},
- abstract = {Neural networks trained only to optimize for training accuracy can often be fooled by adversarial examples — slightly perturbed inputs misclassiï¬ed with high conï¬dence. Veriï¬cation of networks enables us to gauge their vulnerability to such adversarial examples. We formulate veriï¬cation of piecewise-linear neural networks as a mixed integer program. On a representative task of ï¬nding minimum adversarial distortions, our veriï¬er is two to three orders of magnitude quicker than the state-of-the-art. We achieve this computational speedup via tight formulations for non-linearities, as well as a novel presolve algorithm that makes full use of all information available. The computational speedup allows us to verify properties on convolutional and residual networks with over 100,000 ReLUs — several orders of magnitude more than networks previously veriï¬ed by any complete veriï¬er. In particular, we determine for the ï¬rst time the exact adversarial accuracy of an MNIST classiï¬er to perturbations with bounded l∞ norm = 0.1: for this classiï¬er, we ï¬nd an adversarial example for 4.38\% of samples, and a certiï¬cate of robustness to norm-bounded perturbations for the remainder. Across all robust training procedures and network architectures considered, and for both the MNIST and CIFAR-10 datasets, we are able to certify more samples than the state-of-the-art and ï¬nd more adversarial examples than a strong ï¬rst-order attack.},
- language = {en},
- urldate = {2022-06-13},
- publisher = {arXiv},
- author = {Tjeng, Vincent and Xiao, Kai and Tedrake, Russ},
- month = feb,
- year = {2019},
- note = {Number: arXiv:1711.07356
-arXiv:1711.07356 [cs]},
- keywords = {Computer Science - Machine Learning, Computer Science - Computer Vision and Pattern Recognition, Computer Science - Cryptography and Security},
- file = {Tjeng et al. - 2019 - Evaluating Robustness of Neural Networks with Mixe.pdf:/home/bjorn/Zotero/storage/GJCQR6LN/Tjeng et al. - 2019 - Evaluating Robustness of Neural Networks with Mixe.pdf:application/pdf},
-}
-
-@article{kuvshinov_robustness_2022,
- title = {Robustness verification of {ReLU} networks via quadratic programming},
- issn = {0885-6125, 1573-0565},
- url = {https://link.springer.com/10.1007/s10994-022-06132-9},
- doi = {10.1007/s10994-022-06132-9},
- abstract = {Abstract
- Neural networks are known to be sensitive to adversarial perturbations. To investigate this undesired behavior we consider the problem of computing the distance to the decision boundary (DtDB) from a given sample for a deep neural net classifier. In this work we present a procedure where we solve a convex quadratic programming (QP) task to obtain a lower bound on the DtDB. This bound is used as a robustness certificate of the classifier around a given sample. We show that our approach provides better or competitive results in comparison with a wide range of existing techniques.},
- language = {en},
- urldate = {2022-06-13},
- journal = {Mach Learn},
- author = {Kuvshinov, Aleksei and Günnemann, Stephan},
- month = mar,
- year = {2022},
- file = {Full Text:/home/bjorn/Zotero/storage/3GZPWDXG/Kuvshinov and Günnemann - 2022 - Robustness verification of ReLU networks via quadr.pdf:application/pdf},
-}
-
-@article{bingham_discovering_2022,
- title = {Discovering {Parametric} {Activation} {Functions}},
- volume = {148},
- issn = {08936080},
- url = {http://arxiv.org/abs/2006.03179},
- doi = {10.1016/j.neunet.2022.01.001},
- abstract = {Recent studies have shown that the choice of activation function can significantly affect the performance of deep learning networks. However, the benefits of novel activation functions have been inconsistent and task dependent, and therefore the rectified linear unit (ReLU) is still the most commonly used. This paper proposes a technique for customizing activation functions automatically, resulting in reliable improvements in performance. Evolutionary search is used to discover the general form of the function, and gradient descent to optimize its parameters for different parts of the network and over the learning process. Experiments with four different neural network architectures on the CIFAR-10 and CIFAR-100 image classification datasets show that this approach is effective. It discovers both general activation functions and specialized functions for different architectures, consistently improving accuracy over ReLU and other activation functions by significant margins. The approach can therefore be used as an automated optimization step in applying deep learning to new tasks.},
- urldate = {2022-06-13},
- journal = {Neural Networks},
- author = {Bingham, Garrett and Miikkulainen, Risto},
- month = apr,
- year = {2022},
- note = {arXiv:2006.03179 [cs, stat]},
- keywords = {Computer Science - Machine Learning, Statistics - Machine Learning, Computer Science - Neural and Evolutionary Computing, Computer Science - Computer Vision and Pattern Recognition},
- pages = {48--65},
- file = {arXiv Fulltext PDF:/home/bjorn/Zotero/storage/FZN2ZSVD/Bingham and Miikkulainen - 2022 - Discovering Parametric Activation Functions.pdf:application/pdf;arXiv.org Snapshot:/home/bjorn/Zotero/storage/NH2RIFW5/2006.html:text/html},
-}
-
-@misc{nwankpa_activation_2018,
- title = {Activation {Functions}: {Comparison} of trends in {Practice} and {Research} for {Deep} {Learning}},
- shorttitle = {Activation {Functions}},
- url = {http://arxiv.org/abs/1811.03378},
- abstract = {Deep neural networks have been successfully used in diverse emerging domains to solve real world complex problems with may more deep learning(DL) architectures, being developed to date. To achieve these state-of-the-art performances, the DL architectures use activation functions (AFs), to perform diverse computations between the hidden layers and the output layers of any given DL architecture. This paper presents a survey on the existing AFs used in deep learning applications and highlights the recent trends in the use of the activation functions for deep learning applications. The novelty of this paper is that it compiles majority of the AFs used in DL and outlines the current trends in the applications and usage of these functions in practical deep learning deployments against the state-of-the-art research results. This compilation will aid in making effective decisions in the choice of the most suitable and appropriate activation function for any given application, ready for deployment. This paper is timely because most research papers on AF highlights similar works and results while this paper will be the first, to compile the trends in AF applications in practice against the research results from literature, found in deep learning research to date.},
- urldate = {2022-06-13},
- publisher = {arXiv},
- author = {Nwankpa, Chigozie and Ijomah, Winifred and Gachagan, Anthony and Marshall, Stephen},
- month = nov,
- year = {2018},
- note = {Number: arXiv:1811.03378
-arXiv:1811.03378 [cs]},
- keywords = {Computer Science - Machine Learning, Computer Science - Computer Vision and Pattern Recognition},
- file = {arXiv Fulltext PDF:/home/bjorn/Zotero/storage/IIU9YVLX/Nwankpa et al. - 2018 - Activation Functions Comparison of trends in Prac.pdf:application/pdf;arXiv.org Snapshot:/home/bjorn/Zotero/storage/PP6HN5AA/1811.html:text/html},
-}
diff --git a/src/slides/Makefile b/src/slides/Makefile
deleted file mode 100644
index a9e41770052193ab6ca76062fb262a00c06e038e..0000000000000000000000000000000000000000
--- a/src/slides/Makefile
+++ /dev/null
@@ -1,13 +0,0 @@
-TARGET?=main
-SUBFOLDER?=src/slides
-
-default: build
-
-clean:
- cd $(SUBFOLDER); rm -f *.aux *.log *.out *.pdf *.toc *.synctex.gz *.bbl *.blg
-build:
- cd $(SUBFOLDER); \
- xelatex -shell-escape $(TARGET).tex; \
- biber $(TARGET); \
- xelatex -shell-escape $(TARGET).tex; \
- xelatex -shell-escape $(TARGET).tex
diff --git a/src/slides/images/hidden_layers_with_inputs.pdf b/src/slides/images/hidden_layers_with_inputs.pdf
deleted file mode 100644
index 69b7b3de031f33d2d2e92e12230215303929d726..0000000000000000000000000000000000000000
Binary files a/src/slides/images/hidden_layers_with_inputs.pdf and /dev/null differ
diff --git a/src/slides/images/hidden_layers_with_inputs.pdf_tex b/src/slides/images/hidden_layers_with_inputs.pdf_tex
deleted file mode 100644
index 2eebefa471ca456418fb5c7f35186392e58f345a..0000000000000000000000000000000000000000
--- a/src/slides/images/hidden_layers_with_inputs.pdf_tex
+++ /dev/null
@@ -1,62 +0,0 @@
-%% Creator: Inkscape 1.1.2 (0a00cf5339, 2022-02-04), www.inkscape.org
-%% PDF/EPS/PS + LaTeX output extension by Johan Engelen, 2010
-%% Accompanies image file 'hidden_layers_with_inputs.pdf' (pdf, eps, ps)
-%%
-%% To include the image in your LaTeX document, write
-%% \input{<filename>.pdf_tex}
-%% instead of
-%% \includegraphics{<filename>.pdf}
-%% To scale the image, write
-%% \def\svgwidth{<desired width>}
-%% \input{<filename>.pdf_tex}
-%% instead of
-%% \includegraphics[width=<desired width>]{<filename>.pdf}
-%%
-%% Images with a different path to the parent latex file can
-%% be accessed with the `import' package (which may need to be
-%% installed) using
-%% \usepackage{import}
-%% in the preamble, and then including the image with
-%% \import{<path to file>}{<filename>.pdf_tex}
-%% Alternatively, one can specify
-%% \graphicspath{{<path to file>/}}
-%%
-%% For more information, please see info/svg-inkscape on CTAN:
-%% http://tug.ctan.org/tex-archive/info/svg-inkscape
-%%
-\begingroup%
- \makeatletter%
- \providecommand\color[2][]{%
- \errmessage{(Inkscape) Color is used for the text in Inkscape, but the package 'color.sty' is not loaded}%
- \renewcommand\color[2][]{}%
- }%
- \providecommand\transparent[1]{%
- \errmessage{(Inkscape) Transparency is used (non-zero) for the text in Inkscape, but the package 'transparent.sty' is not loaded}%
- \renewcommand\transparent[1]{}%
- }%
- \providecommand\rotatebox[2]{#2}%
- \newcommand*\fsize{\dimexpr\f@size pt\relax}%
- \newcommand*\lineheight[1]{\fontsize{\fsize}{#1\fsize}\selectfont}%
- \ifx\svgwidth\undefined%
- \setlength{\unitlength}{1417.5bp}%
- \ifx\svgscale\undefined%
- \relax%
- \else%
- \setlength{\unitlength}{\unitlength * \real{\svgscale}}%
- \fi%
- \else%
- \setlength{\unitlength}{\svgwidth}%
- \fi%
- \global\let\svgwidth\undefined%
- \global\let\svgscale\undefined%
- \makeatother%
- \begin{picture}(1,0.5010582)%
- \lineheight{1}%
- \setlength\tabcolsep{0pt}%
- \put(0,0){\includegraphics[width=\unitlength,page=1]{hidden_layers_with_inputs.pdf}}%
- \put(0.04729358,0.29944813){\color[rgb]{0,0,0}\makebox(0,0)[lt]{\lineheight{1.25}\smash{\begin{tabular}[t]{l}$x_1 \pm u(x_1) $\end{tabular}}}}%
- \put(0.0473347,0.18931853){\color[rgb]{0,0,0}\makebox(0,0)[lt]{\lineheight{1.25}\smash{\begin{tabular}[t]{l}$x_2 \pm u(x_2) $\end{tabular}}}}%
- \put(0.81423301,0.30023344){\color[rgb]{0,0,0}\makebox(0,0)[lt]{\lineheight{1.25}\smash{\begin{tabular}[t]{l}$y_1 \pm u(y_1) $\end{tabular}}}}%
- \put(0.81427418,0.19010386){\color[rgb]{0,0,0}\makebox(0,0)[lt]{\lineheight{1.25}\smash{\begin{tabular}[t]{l}$y_2 \pm u(y_2) $\end{tabular}}}}%
- \end{picture}%
-\endgroup%
diff --git a/src/slides/images/hidden_layers_with_inputs.svg b/src/slides/images/hidden_layers_with_inputs.svg
deleted file mode 100644
index 79c11d459c1a6f50e7f81ddaa348722afdce738f..0000000000000000000000000000000000000000
--- a/src/slides/images/hidden_layers_with_inputs.svg
+++ /dev/null
@@ -1,550 +0,0 @@
-<?xml version="1.0" encoding="UTF-8" standalone="no"?>
-<svg
- width="1890"
- height="947"
- style="cursor: move; touch-action: none; -webkit-tap-highlight-color: rgba(0, 0, 0, 0);"
- version="1.1"
- id="svg605"
- sodipodi:docname="hidden_layers_with_inputs.svg"
- inkscape:version="1.1.2 (0a00cf5339, 2022-02-04)"
- xmlns:inkscape="http://www.inkscape.org/namespaces/inkscape"
- xmlns:sodipodi="http://sodipodi.sourceforge.net/DTD/sodipodi-0.dtd"
- xmlns="http://www.w3.org/2000/svg"
- xmlns:svg="http://www.w3.org/2000/svg"
- xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
- xmlns:cc="http://creativecommons.org/ns#"
- xmlns:dc="http://purl.org/dc/elements/1.1/">
- <metadata
- id="metadata609">
- <rdf:RDF>
- <cc:Work
- rdf:about="">
- <dc:format>image/svg+xml</dc:format>
- <dc:type
- rdf:resource="http://purl.org/dc/dcmitype/StillImage" />
- </cc:Work>
- </rdf:RDF>
- </metadata>
- <sodipodi:namedview
- pagecolor="#ffffff"
- bordercolor="#666666"
- borderopacity="1"
- objecttolerance="10"
- gridtolerance="10"
- guidetolerance="10"
- inkscape:pageopacity="0"
- inkscape:pageshadow="2"
- inkscape:window-width="1039"
- inkscape:window-height="1051"
- id="namedview607"
- showgrid="false"
- inkscape:zoom="1.4696522"
- inkscape:cx="298.02971"
- inkscape:cy="262.30697"
- inkscape:window-x="1950"
- inkscape:window-y="0"
- inkscape:window-maximized="0"
- inkscape:current-layer="svg605"
- inkscape:document-rotation="0"
- inkscape:pagecheckerboard="1" />
- <defs
- id="defs603">
- <marker
- id="arrow"
- viewBox="0 -5 10 10"
- markerWidth="7"
- markerHeight="7"
- orient="auto"
- refX="40">
- <path
- d="M0,-5L10,0L0,5"
- style="stroke: rgb(80, 80, 80); fill: rgb(80, 80, 80);"
- id="path600" />
- </marker>
- </defs>
- <text
- class="text"
- dy="17.033411"
- x="1009.1931"
- y="856.7785"
- style="font-size:48.6669px;stroke-width:4.05557"
- id="text589"
- transform="scale(1.0217749,0.97868915)" />
- <text
- class="text"
- dy="17.033411"
- x="1293.0834"
- y="856.7785"
- style="font-size:48.6669px;stroke-width:4.05557"
- id="text591"
- transform="scale(1.0217749,0.97868915)" />
- <text
- class="text"
- dy="17.033411"
- x="1576.9735"
- y="856.7785"
- style="font-size:48.6669px;stroke-width:4.05557"
- id="text593"
- transform="scale(1.0217749,0.97868915)" />
- <text
- class="text"
- dy="17.033411"
- x="1860.864"
- y="856.7785"
- style="font-size:48.6669px;stroke-width:4.05557"
- id="text595"
- transform="scale(1.0217749,0.97868915)" />
- <text
- class="text"
- dy="23.269838"
- x="873.88049"
- y="1075.0447"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text314" />
- <text
- class="text"
- dy="23.269838"
- x="1261.7112"
- y="1075.0447"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text316" />
- <text
- class="text"
- dy="23.269838"
- x="1649.5419"
- y="1075.0447"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text318" />
- <text
- class="text"
- dy="23.269838"
- x="2037.3728"
- y="1075.0447"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text320" />
- <text
- class="text"
- dy="23.269838"
- x="858.85999"
- y="1067.8901"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text764" />
- <text
- class="text"
- dy="23.269838"
- x="1246.6907"
- y="1067.8901"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text766" />
- <text
- class="text"
- dy="23.269838"
- x="1634.5214"
- y="1067.8901"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text768" />
- <text
- class="text"
- dy="23.269838"
- x="2022.3523"
- y="1067.8901"
- style="font-size:66.4853px;stroke-width:5.54044"
- id="text770" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,369.92279 751.82903,265.76382"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path661" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,369.92279 751.82903,474.08178"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path663" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,369.92279 751.82903,682.39974"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path665" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,578.24075 751.82903,265.76382"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path667" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,578.24075 751.82903,474.08178"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path669" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,578.24075 751.82903,682.39974"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path671" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,57.445853 1114.4089,265.76382"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path673" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,57.445853 1114.4089,474.08178"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path675" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,265.76382 H 1114.4089"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path677" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,265.76382 1114.4089,474.08178"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path679" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,474.08178 1114.4089,265.76382"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path681" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,474.08178 H 1114.4089"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path683" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,682.39974 1114.4089,265.76382"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path685" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,682.39974 1114.4089,474.08178"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path687" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,57.445853 1476.989,369.92279"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path689" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,265.76382 1476.989,369.92279"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path691" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,474.08178 1476.989,369.92279"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path693" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,57.445853 1476.989,578.24075"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path695" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,265.76382 1476.989,578.24075"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path697" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,474.08178 1476.989,578.24075"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path699" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,57.445853 1114.4089,682.39974"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path701" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,265.76382 1114.4089,682.39974"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path703" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,474.08178 1114.4089,682.39974"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path705" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,682.39974 H 1114.4089"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path707" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,682.39974 1476.989,369.92279"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path709" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,682.39974 1476.989,578.24075"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path711" />
- <path
- class="link"
- marker-end=""
- d="m 389.24897,369.92279 362.58006,520.7949"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path713" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,578.24075 751.82903,890.71769"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path715" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,890.71769 1114.4089,265.76382"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path717" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,890.71769 1114.4089,474.08178"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path719" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,890.71769 1114.4089,682.39974"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path721" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,57.445853 1114.4089,890.71769"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path723" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,265.76382 1114.4089,890.71769"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path725" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,474.08178 1114.4089,890.71769"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path727" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,682.39974 1114.4089,890.71769"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path729" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,890.71769 H 1114.4089"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path731" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,890.71769 1476.989,369.92279"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path733" />
- <path
- class="link"
- marker-end=""
- d="M 1114.4089,890.71769 1476.989,578.24075"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path735" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,369.92279 751.82903,57.445853"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path737" />
- <path
- class="link"
- marker-end=""
- d="M 389.24897,578.24075 751.82903,57.445853"
- style="stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path739" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,57.445853 H 1114.4089"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path741" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,265.76382 1114.4089,57.445853"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path743" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,474.08178 1114.4089,57.445853"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path745" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,682.39974 1114.4089,57.445853"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path747" />
- <path
- class="link"
- marker-end=""
- d="M 751.82903,890.71769 1114.4089,57.445853"
- style="fill:#c8beb7;stroke:#505050;stroke-width:2.07753;stroke-opacity:1"
- id="path749" />
- <ellipse
- class="node"
- id="1_0"
- cx="751.82898"
- cy="57.445866"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="1_1"
- cx="751.82898"
- cy="265.76382"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="1_2"
- cx="751.82898"
- cy="474.08179"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="1_3"
- cx="751.82898"
- cy="682.39972"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="1_4"
- cx="751.82898"
- cy="890.71771"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="2_0"
- cx="1114.4091"
- cy="57.445866"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="2_1"
- cx="1114.4091"
- cy="265.76382"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="2_2"
- cx="1114.4091"
- cy="474.08179"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="2_3"
- cx="1114.4091"
- cy="682.39972"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="2_4"
- cx="1114.4091"
- cy="890.71771"
- style="fill:#f4e3d7;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="3_0"
- cx="1476.9894"
- cy="369.92279"
- style="fill:#ffffff;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="3_1"
- cx="1476.9894"
- cy="578.24078"
- style="fill:#ffffff;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="3_0-4"
- cx="388.60492"
- cy="369.82123"
- style="fill:#ffffff;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <ellipse
- class="node"
- id="3_0-8"
- cx="388.52823"
- cy="578.25275"
- style="fill:#ffffff;stroke:#333333;stroke-width:5.19381"
- rx="51.797153"
- ry="52.079494" />
- <g
- id="g105949"
- transform="translate(-34.658572,-1.5529019)">
- <text
- xml:space="preserve"
- style="font-size:42.6667px;font-family:sans-serif;-inkscape-font-specification:'sans-serif, Normal';letter-spacing:0px;fill:#000000;-webkit-tap-highlight-color:rgba(0, 0, 0, 0)"
- x="97.183601"
- y="384.22177"
- id="text30949"><tspan
- sodipodi:role="line"
- id="tspan30947"
- x="97.183601"
- y="384.22177"
- style="font-size:42.6667px">$x_1 \pm u(x_1) $</tspan></text>
- </g>
-</svg>
diff --git a/src/slides/main.tex b/src/slides/main.tex
deleted file mode 100755
index 78ec8203828e0b7e18e15dafde29f4ce739671cf..0000000000000000000000000000000000000000
--- a/src/slides/main.tex
+++ /dev/null
@@ -1,249 +0,0 @@
-%% LaTeX Beamer presentation template (requires beamer package)
-%% see http://tug.ctan.org/tex-archive/macros/latex/contrib/beamer/doc
-%% /beameruserguide.pdf
-
-% This either shows compressed or normal navigation bars
-%\documentclass[compress]{beamer}
-%\documentclass[draft]{beamer}
-%\documentclass[aspectratio=169,handout]{beamer}
-\documentclass[aspectratio=169]{beamer}
-% This speeds up building by only compiling the mentioned slides.
-%\includeonlyframes{current}
-
-% Check those default combinations here:
-% https://hartwork.org/beamer-theme-matrix/
-%\usetheme{Darmstadt}
-\usetheme{Frankfurt}
-\usecolortheme{beaver}
-
-
-\setbeamertemplate{navigation symbols}{}
-\addtobeamertemplate{footline}{
- \leavevmode%
- \hbox{%
- \begin{beamercolorbox}[wd=.333333\paperwidth,ht=2.25ex,dp=1ex,
- center]{author in head/foot}%
- \usebeamerfont{author in head/foot}\insertshortauthor
- \end{beamercolorbox}%
- \begin{beamercolorbox}[wd=.333333\paperwidth,ht=2.25ex,dp=1ex,
- center]{title in head/foot}%
- \usebeamerfont{title in head/foot}\insertframenumber /
- \inserttotalframenumber
- \end{beamercolorbox}%
- \begin{beamercolorbox}[wd=.333333\paperwidth,ht=2.75ex,dp=.5ex,right,
- rightskip=1em]{}%
- \usebeamercolor[fg]{navigation symbols}\insertslidenavigationsymbol%
- \insertframenavigationsymbol%
- \insertsubsectionnavigationsymbol%
- \insertsectionnavigationsymbol%
- \insertdocnavigationsymbol%
- \insertbackfindforwardnavigationsymbol%
- \end{beamercolorbox}%
- }%
- \vskip0.5pt%
-}{}
-
-\newenvironment{noheadline}{
- \setbeamertemplate{headline}{}
- \addtobeamertemplate{frametitle}{\vspace*{-0.9\baselineskip}}{}
- }{}
-
-% This displays an outline slide before each section.
-% \AtBeginSection[]
-% {
-% \begin{frame}<beamer>
-% \frametitle{Outline for section \thesection}
-% \tableofcontents[
-% currentsection,
-% currentsubsection,
-% subsectionstyle=show/show/hide,
-% subsubsectionstyle=show/show/show/hide
-% ]
-% \end{frame}
-% }
-
-% Shadyly show not yet visible lines of the current slide.
-\setbeamercovered{transparent}
-\usepackage[T1]{fontenc}
-\usepackage[english]{babel}
-%\usepackage{lmodern}
-\usepackage[scaled=.90]{helvet}
-\usepackage{xcolor}
-% Enalbe notes in pdfpc.
-\usepackage{pgfpages}
-% This actually is not needed, just PyCharm complains about it missing.
-\usepackage{amsfonts}
-\usepackage{amsmath}
-\usepackage{graphicx}
-\usepackage{csquotes}
-\usepackage{hyperref}
-\usepackage{import}
-\usepackage{ccicons}
-\usepackage[ruled,linesnumbered]{algorithm2e}
-
-\usepackage{caption}
-% Redefine the caption setup of the figures environment in the beamer class.
-% Minimize font size.
-\captionsetup{font=tiny, labelfont=tiny}
-% Remove label "Figure".
-\captionsetup[figure]{labelformat=empty}
-% This switches on figure numbering in case labels are shown.
-%\setbeamertemplate{caption}[numbered]
-
-\graphicspath{{images/}}
-%\setbeamertemplate{theorems}[numbered] %
-
-\theoremstyle{definition} % insert bellow all blocks you want in normal text
-
-% This shows the notes in pdfpg on the presenters screen if opened with
-% pdfpc introduction_to_neural_networks.pdf --notes=right
-%\setbeameroption{show notes on second screen}
-
-% Packages for the references.
-\usepackage[backend=biber,style=apa,natbib=true]{biblatex}
-\DeclareLanguageMapping{english}{english-apa}
-\addbibresource{GUM.bib}
-\addbibresource{references.bib}
-\usepackage{mathtools}
-
-\title{GUM-compliant neural network robustness verification}
-
-\subtitle{\ldots towards a thesis in the M.Sc. Mathematics programme}
-
-% - Use the \inst{?} command only if the authors have different
-% affiliation.
-%\author{F.~Author\inst{1} \and S.~Another\inst{2}}
-\author{Björn Ludwig}
-
-% - Use the \inst command only if there are several affiliations.
-% - Keep it simple, no one is interested in your street address.
-\institute[TU Berlin, ZIB and PTB]{Technical University of Berlin, Zuse Institute
-Berlin and Physikalisch-Technische Bundesanstalt}
-
-\date[July 2022]{12th July 2022}
-
-
-% This is only inserted into the PDF information catalog. Can be left
-% out.
-\subject{Talks}
-
-% This would be the logo to be used, but it doesn't suit all themes
-%\pgfdeclareimage[width=6 cm]{unilogo}{images/tu-berlin_logo}
-%\logo{\pgfuseimage{unilogo}}
-
-% If you wish to uncover everything in a step-wise fashion, uncomment
-% the following command:
-% \beamerdefaultoverlayspecification{<+->}
-
-\begin{document}
-
- \maketitle
-
-
- \section{Ideas}\label{sec:ideas}
-
- \begin{noheadline}
-
- \begin{frame}{Main idea in a nutshell}
- \begin{itemize}[<+->]
- \item Solve a classification task (e.g.,\ residual lifespan in predictive
- maintenance),
- \item employing a common multi-layer perceptron (MLP),
- \item with uncertain inputs,
- \item while propagating measurement uncertainty to the outputs,
- \item and verify robustness via MILP
- \end{itemize}
- \end{frame}
-
- \begin{frame}{1. Propagate inputs' uncertainties to outputs}
- \begin{figure}
- \centering
- \begin{figure}
- \centering
- \def\svgwidth{\columnwidth}
- \subimport{images/}{hidden_layers_with_inputs.pdf_tex}
- \caption{}
- \label{fig:propagate-inputs-uncertainties-to-outputs}
- \end{figure}
- \end{figure}
- \end{frame}
-
- \begin{frame}{1. Propagate inputs' uncertainties to outputs}
- \begin{itemize}[<+->]
- \item uncertainties \(u(\cdot)\) as referred to by the
- \textit{Guide to the expression of uncertainty in measurement (GUM)
- }~\citep{jcgm_evaluation_2008}
- \item uncertainty propagation via the \textit{law of propagation of uncertainty
- (LPU)} for uncorrelated inputs~\citep[chapter E.3]{jcgm_evaluation_2008}
- \[u^2(y) = \sum_{i=1}^N \left( \frac{\partial f}{\partial x_i} \right)^2 u^2
- (x_i)\]
- \item requires continuously differentiable activation function
- \end{itemize}
- \end{frame}
-
- \begin{frame}{2. Incorporate uncertainties into existing method}
- \begin{itemize}[<+->]
- \item take
- \begin{enumerate}
- \item the uncertain inputs and network's outputs with propagated
- uncertainties
- \item a state-of-the-art QP method which works on a neural network's input
- and output spaces
- \end{enumerate}
- \item extend the method to take the uncertainties into consideration
- \end{itemize}
- \end{frame}
-
- \begin{frame}{Robustness verification via MIP or QP}
- \begin{itemize}[<+->]
- \item Neural networks trained only to optimize training accuracy can often
- be fooled by adversarial examples
- \item Verification enables to gauge vulnerability to such attacks
- \item~\cite{tjeng_evaluating_2019} present a complete verifier
- via MILP and solve the verification task \enquote{two to three orders of
- magnitude quicker} than the state-of-the-art
- \item~\cite{kuvshinov_robustness_2022} compute the distance to the decision
- boundary for a given sample in a way, which they claim to provide
- \enquote{better or competitive results in comparison with a wide range of
- existing techniques}
- \end{itemize}
- \end{frame}
-
- \begin{frame}{Main challenges}
- \begin{itemize}[<+->]
- \item all methods so far exploit all kinds of linearities (ReLU, \(l_\infty\)
- or \(l_1\)-Norm)
- \item find a way to bound
- \end{itemize}
- \end{frame}
-
- \section{Outline}\label{sec:outline}
-
- \begin{frame}{Thesis outline (preliminary)}
- \begin{itemize}
- \item Abstract
- \item Introduction
- \item Methods
- \begin{itemize}
- \item GUM-compliant measurement uncertainty propagation in NN
- \item Activation functions
- \begin{itemize}
- \item Prerequisites for the activation functions
- \item Parametric Softplus
- \end{itemize}
- \item Robustness verification
- \end{itemize}
- \item Simulation/Numerical results
- \item Conclusions and outlook
- \item References
- \end{itemize}
- \end{frame}
-
-
- \end{noheadline}
-
- \begin{frame}{References}
- \printbibliography
- \end{frame}
-
-\end{document}
diff --git a/src/internship_results/Makefile b/src/thesis_presentation/Makefile
similarity index 55%
rename from src/internship_results/Makefile
rename to src/thesis_presentation/Makefile
index 82428a567d5a8697c90740907ce3e2d98e0addad..69d4b5848dd3ea30797873b69cae677cceada9af 100644
--- a/src/internship_results/Makefile
+++ b/src/thesis_presentation/Makefile
@@ -1,5 +1,5 @@
TARGET?=main
-SUBFOLDER?=src/internship_results
+SUBFOLDER?=src/thesis_presentation
default: build
@@ -9,14 +9,29 @@ clean:
$(TARGET).bbl \
$(TARGET).bcf \
$(TARGET).blg \
+ $(TARGET).lof \
$(TARGET).log \
+ $(TARGET).lot \
+ $(TARGET).nav \
$(TARGET).out \
$(TARGET).run.xml \
+ $(TARGET).snm \
$(TARGET).synctex.gz \
+ $(TARGET).synctex\(busy\) \
+ $(TARGET).thm \
$(TARGET).toc
+
build:
cd $(SUBFOLDER); \
xelatex -shell-escape $(TARGET).tex; \
biber $(TARGET); \
xelatex -shell-escape $(TARGET).tex; \
xelatex -shell-escape $(TARGET).tex
+
+single-build:
+ cd $(SUBFOLDER); \
+ xelatex -shell-escape $(TARGET).tex
+
+quick-build:
+ cd $(SUBFOLDER); \
+ xelatex -interaction=batchmode -shell-escape $(TARGET).tex
diff --git a/src/slides/references.bib b/src/thesis_presentation/references.bib
similarity index 100%
rename from src/slides/references.bib
rename to src/thesis_presentation/references.bib