<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Elena Ranguelova on Medium]]></title>
        <description><![CDATA[Stories by Elena Ranguelova on Medium]]></description>
        <link>https://medium.com/@e.ranguelova?source=rss-1c30c46a7632------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ib-aSNE2_W_Vekyo4MIKWw.png</url>
            <title>Stories by Elena Ranguelova on Medium</title>
            <link>https://medium.com/@e.ranguelova?source=rss-1c30c46a7632------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 12:19:56 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@e.ranguelova/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[How to find your Artificial Intelligence explainer]]></title>
            <link>https://medium.com/escience-center/how-to-find-your-artificial-intelligence-explainer-dbb1ac608009?source=rss-1c30c46a7632------2</link>
            <guid isPermaLink="false">https://medium.com/p/dbb1ac608009</guid>
            <category><![CDATA[explainable-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Elena Ranguelova]]></dc:creator>
            <pubDate>Thu, 17 Mar 2022 08:44:52 GMT</pubDate>
            <atom:updated>2022-03-21T12:03:32.714Z</atom:updated>
            <cc:license>http://creativecommons.org/licenses/by/4.0/</cc:license>
            <content:encoded><![CDATA[<p>Our easy-to-use <a href="https://github.com/dianna-ai/dianna">DIANNA</a> library helps you figure out which method works for your research.</p><p><em>by </em><a href="https://orcid.org/0000-0002-9834-1756"><em>Elena Ranguelova</em></a><em> and </em><a href="https://orcid.org/0000-0002-1966-8460"><em>Yang Liu</em></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ocxIDV_ec--jDLfz_t8T5g.jpeg" /><figcaption>The logic of AI should be understandable by humans. [Image by <a href="https://www.shutterstock.com/g/PhonlamaiPhoto">Phonlamai Photo</a>/<a href="https://www.shutterstock.com/home">Shutterstock.com</a>.]</figcaption></figure><h4><strong>Why Explainable Artificial Intelligence?</strong></h4><p>Explainable Artificial Intelligence (or XAI) is AI that provides interpretable outputs, that can be easily understood by humans. Unfortunately, often there is a trade-off between interpretability and performance — for example, traditional methods like decision trees are very interpretable, but have lower performance compared to the “black-box” (Deep) Neural Networks (DNN). In addition, what is needed to make something “understandable” depends on several factors, like context and the method’s target audience.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*Ypd67-9B3N8HtZCaCKT_DA.png" /><figcaption>Trade-off between AI models accuracy and interpretability</figcaption></figure><p>Since we would like to have <em>both</em> high-performing <em>and</em> interpretable models, we need XAI. <strong>There are many XAI methods to choose from, but which is the “best”? </strong>One of the goals of the <a href="https://www.research-software.nl/projects/e2bf3bd0-c429-4caa-9c2c-1f41a188e8ca">Deep Insight and Neural Network Analysis (DIANNA) project</a> is to help answer this question for researchers in any scientific domain, especially the non-AI experts. DIANNA provides a standard API for several XAI methods to make them easier for researchers to use. This blog post describes our process of selecting which XAI methods to support.</p><h4><strong>An overview of explainable AI approaches</strong></h4><p>There are many ways to approach the explainability of complex opaque models. We have considered the large group of <em>post-hoc</em> (after training) XAI methods. They provide insight into an AI model’s reasoning on a given data item of interest by quantifying the <em>relevance</em> of individual data features (image pixels, words in a text, etc.) with respect to a selected prediction. The resulting <em>relevance scores</em>, overlaid on the inputs, indicate how the decision has been made. The input, in this case, is the (scientific) data which can be of different modalities (image, text, speech, etc.). The XAI visualizes the knowledge captured by the trained AI model, which therefore can become a source of scientific insights.</p><p>Many <em>post-hoc </em>XAI methods exist and some apply to many data modalities. Some examples are <a href="https://arxiv.org/abs/1602.04938">LIME</a>, <a href="https://arxiv.org/abs/1705.07874">SHAP</a>, <a href="https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0130140">Layer-wise Relevance Propagation (LRP)</a>, <a href="https://www.sciencedirect.com/science/article/pii/S0031320316303582">Deep Taylor Decomposition (DTD)</a>, <a href="https://arxiv.org/pdf/1704.02685.pdf">DeepLIFT</a>, <a href="https://arxiv.org/pdf/1704.02685.pdf">Pattern Attribution and PatternNet</a> and <a href="https://arxiv.org/abs/1806.07421">RISE</a>.</p><p>Here, we will use the LRP method to illustrate the principle of determining which features in an input contribute most strongly to the output of a trained DNN model. In this example, the input to the network is an <strong>image.</strong> We are interested in figuring out the <em>relevance</em> of each image pixel to a given image class. This could be the final DNN classification decision for that image (“cat”) or any other class known to the model (e.g. “dog”, “cow”, “bee”, etc. if the model is trained to classify animals). The output relevance is considered a weighted sum of the relevancies of all the <strong>pixels</strong> of the input image. The relevance at the output node in the network graph is decomposed by iterative propagation backward through the network layers until the input layer. The trained model weights are used in this decomposition. Using this <em>relevance propagation</em> process, the final <em>relevance scores map </em>(<em>or heatmap</em>) is obtained. The final relevance map gives us an indication for the DNN’s hidden logic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9jrocNtPKKoVN7xh7-CWeA.png" /><figcaption>Relevance scores of the image pixels in respect to the class “cat” obtained via relevance propagation.</figcaption></figure><p>Besides LRP, there are many other XAI methods, which use different ways of determining relevance. In order to determine which of the plethora of post-hoc XAI methods to include in DIANNA, we started by compiling an initial list of 18 XAI post-hoc feature-relevance methods following <a href="https://www.sciencedirect.com/science/article/abs/pii/S1566253519308103"><em>Arrieta et al. (2020)</em></a>:</p><ul><li><strong>Deep Taylor decomposition (DTD):</strong> <a href="https://www.sciencedirect.com/science/article/pii/S0031320316303582">Paper</a>| <a href="https://www.youtube.com/watch?v=gy_Cb4Do_YE">Presentation</a>| <a href="https://github.com/marcoancona/DeepExplain">Code</a></li><li><strong>Layer-wise Relevance Propagation (LRP): </strong><a href="https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0130140">Paper</a>|<a href="http://danshiebler.com/2017-04-16-deep-taylor-lrp/">Blog</a>| <a href="https://github.com/albermax/innvestigate">Code</a></li><li><strong>Integrated Gradients (IG): </strong><a href="https://arxiv.org/pdf/1703.01365.pdf">Paper</a>| <a href="https://captum.ai/docs/extension/integrated_gradients/">Tutorial</a>| <a href="https://github.com/pytorch/captum">Code</a></li><li><strong>Local Interpretable Model-agnostic Explanations (LIME):</strong> <a href="https://arxiv.org/abs/1602.04938">Paper</a>| <a href="https://towardsdatascience.com/lime-how-to-interpret-machine-learning-models-with-python-94b0e7e4432e">Blog</a>| <a href="https://github.com/marcotcr/lime">Code</a></li><li><strong>SHapley Additive exPlanations (SHAP): </strong><a href="https://arxiv.org/abs/1705.07874">Paper</a>| <a href="https://towardsdatascience.com/shap-how-to-interpret-machine-learning-models-with-python-2323f5af4be9">Blog</a>|<a href="https://github.com/slundberg/shap">Code</a></li><li><strong>PatternNet </strong>and<strong> PatternAttribution: </strong><a href="https://arxiv.org/pdf/1705.05598.pdf">Paper</a>| <a href="https://github.com/albermax/innvestigate">Code</a></li><li><strong>Deep Learning Important FeaTures (DeepLIFT):</strong> <a href="https://arxiv.org/pdf/1704.02685.pdf">Paper</a>| <a href="http://goo.gl/qKb7pL">Tutorial</a>| <a href="https://vimeo.com/238275076">Presentation</a>| <a href="https://github.com/marcoancona/DeepExplain">Code</a></li><li><strong>SmoothGrad: </strong><a href="https://arxiv.org/abs/1706.03825">Paper</a>| <a href="https://pair-code.github.io/saliency/">Demo</a>| <a href="https://github.com/pair-code/saliency">Code</a></li><li><strong>CLass-Enhanced Attentive Response(CLEAR): </strong><a href="https://openaccess.thecvf.com/content_cvpr_2017_workshops/w26/papers/Kumar_Explaining_the_Unexplained_CVPR_2017_paper.pdf">Paper</a></li><li><strong>(Adaptive) deconvolutional network (Deconvnet):</strong> <a href="https://link.springer.com/content/pdf/10.1007/978-3-319-10590-1_53.pdf%20https://www.matthewzeiler.com/mattzeiler/adaptivedeconvolutional.pdf">Paper</a></li><li><strong>Gradient-weighted Class Activation Mapping (Grad-CAM): </strong><a href="https://ieeexplore.ieee.org/document/8237336">Paper</a>| <a href="https://www.pyimagesearch.com/2020/03/09/grad-cam-visualize-class-activation-maps-with-keras-tensorflow-and-deep-learning/">Blog</a>|<a href="https://github.com/ramprs/grad-cam/">Code</a></li><li><strong>Randomized input sampling for explanation of black-box models (RISE): </strong><a href="https://arxiv.org/abs/1806.07421">Paper</a>| <a href="https://github.com/facebookresearch/TorchRay">Code</a></li><li>Extremal Perturbations (EP)| <a href="https://openaccess.thecvf.com/content_ICCV_2019/papers/Fong_Understanding_Deep_Networks_via_Extremal_Perturbations_and_Smooth_Masks_ICCV_2019_paper.pdf">Paper</a>| <a href="https://github.com/facebookresearch/TorchRay">Code</a></li><li>Saliency Analysis (SA): <a href="https://arxiv.org/pdf/1312.6034.pdf">Paper</a>| <a href="https://github.com/osu-xai/pytorch-saliency">Code</a></li><li>Anchors: <a href="https://homes.cs.washington.edu/~marcotcr/aaai18.pdf">Paper</a>| <a href="https://github.com/viadee/javaAnchorExplainer">Code</a></li><li>Deep visualization: <a href="https://arxiv.org/pdf/1506.06579.pdf">Paper</a>| <a href="https://yosinski.com/deepvis">Demo</a>| <a href="https://github.com/yosinski/deep-visualization-toolbox">Code</a></li><li>AutomaticSTRucture IDentification (ASTRID): <a href="https://arxiv.org/pdf/1707.07576.pdf">Paper</a>| <a href="https://github.com/bwrc/astrid-r">Code</a></li></ul><p>We excluded all XAI methods without a working and well-maintained open-source implementation or proven user community. That resulted in a subset of the <strong>13 methods</strong> highlighted above.</p><h4><strong><em>Evaluation criteria</em></strong></h4><p>The large variety of XAI approaches benefits the AI community by providing freedom in the search for explainability. However, it also increases the difficulty of choosing the most appropriate method for each use case. Every XAI approach has pros and cons. In order to select the best methods for our task, we needed a framework to evaluate these methods systematically. Fortunately, this topic has already been visited by the XAI researchers and there is an extensive list of objective criteria defined in the <a href="https://dl.acm.org/doi/abs/10.1145/3351095.3372870">Explainability fact sheet</a> paper [<em>Sokol, K., &amp; Flach, P. (2020)</em>]. The authors look at an abundant set of XAI aspects and cluster the requirements towards an explainer into <strong>5 groups </strong>(or <strong>dimensions</strong>):</p><ul><li>Functional (F1-F9)</li><li>Operational (O1-O10)</li><li>Usability (U1-U11)</li><li>Safety (S1-S4)</li><li>Validation (V1-V2)</li></ul><p>For each dimension, there are specific criteria, e.g. the “Functional” group has 9 and in total there are 36 criteria. We will describe our choices next.</p><h4><strong><em>The selection</em></strong></h4><p>In our <a href="https://github.com/dianna-ai/dianna">DIANNA project</a>, we built a Python package that brings XAI to non (X)AI experts and various scientific domain researchers. The XAI methods to be included in the package had to have the capacity to work with different types of data, various AI architectures and we chose to focus on the model explanation. Most importantly, the resulting explanations had to be easy to interpret. Based on our goals, we have compiled a shortlist of <strong>18 requirements</strong> from the 36 requirements in the <a href="https://dl.acm.org/doi/abs/10.1145/3351095.3372870">XAI fact sheet</a>. Using these relevant for our target users&#39; requirements, we evaluated the selected 13 methods listed in the XAI overview section.</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/89cc601bf64e0887b3a8ff452643a8e3/href">https://medium.com/media/89cc601bf64e0887b3a8ff452643a8e3/href</a></iframe><p>Apart from the requirements on the list, we have also considered the similarity and the “popularity” of these methods, the quality of the existing implementations, and the difficulty of integrating them in DIANNA. The evaluation procedure can be summarized with the table shown below: (<strong>Deconvnet</strong> and <strong>GradCAM</strong> are filtered out since they can only work with Convolutional NNs and are therefore not included in the table.)</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/89d11738dc574c6c3b9ad3a3d1224a83/href">https://medium.com/media/89d11738dc574c6c3b9ad3a3d1224a83/href</a></iframe><p>Finally, we narrowed down our list to <strong>5 methods:</strong></p><ul><li><strong>Layer-wise Relevance Propagation (LRP)</strong></li><li><strong>Local Interpretable Model-agnostic Explanations (LIME)</strong></li><li><strong>SHapley Additive exPlanations (SHAP)</strong></li><li><strong>PatternAttribution (PA)</strong></li><li><strong>Randomized input sampling for explanation of black-box models (RISE)</strong></li></ul><p>Within limited project time, we were able to include three of them in our <a href="https://github.com/dianna-ai/dianna">DIANNA library</a>: RISE, LIME &amp; SHAP. A preview of the explanations that we obtained using our DIANNA library is shown below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tsqE6UtGvVq2siVR.png" /><figcaption>Explanation for “0” with RISE, LIME, and SHAP using DIANNA library for a model trained on binary MNIST data. Red indicates high or positive relevance, blue- low or negative.</figcaption></figure><p>We could observe that the methods are complimentary to each other in respect to their output, which can be particularly useful for researchers seeking new insights to their data and problem. We also believe that simpler datasets, e.g. a binary (only 0s and 1s) MNIST should be used as “Hello world” example for the XAI (user) community!</p><p>Our library is open-source and it is ready to help you decrypt your neural network models. Wait, you want to ask which format do we support? TensorFlow? PyTorch? Don’t worry. We can work with all of them, as we support <a href="https://onnx.ai/">ONNX</a> — the de-facto DNN format standard! Any contributions to our repository are very welcome (talk is cheap, show me your pull request! Want to give it a try? Check out the link below our logo!</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*od59p5ZzK0j97kQC.png" /><figcaption><a href="https://github.com/dianna-ai/dianna">https://github.com/dianna-ai/dianna</a></figcaption></figure><h4>Acknowledgments</h4><p>The work described in this blog has been performed by the members of the DIANNA team: <a href="https://orcid.org/0000-0002-5529-5761">Christiaan Meijer</a>, <a href="https://orcid.org/0000-0001-8724-8372">Leon Oostrum</a>, <a href="https://orcid.org/0000-0002-1966-8460">Yang Liu </a>and <a href="https://orcid.org/0000-0002-9834-1756">Elena Ranguelova</a> with the big help of <a href="https://orcid.org/0000-0002-6033-960X">Patrick Bos</a> and <a href="https://orcid.org/0000-0002-0823-0121">Giulia Crocioni</a>. Special thanks to <a href="https://orcid.org/0000-0002-5529-5761">Christiaan Meijer</a>, <a href="https://orcid.org/0000-0002-0823-0121">Giulia Crocioni</a>, <a href="https://orcid.org/0000-0001-6022-0046">Matthieu Laneuville,</a> <a href="https://orcid.org/0000-0002-2855-940X">Pablo Rodríguez-Sánchez</a> and <a href="https://orcid.org/my-orcid?orcid=0000-0003-3381-2040">Lieke de Boer</a> for their valuable contributions. The DIANNA project is supported by <a href="https://www.esciencecenter.nl/">Netherlands eScience Center</a> and <a href="https://www.surf.nl/en">SURF</a>.</p><h4><em>References</em></h4><p><a href="https://www.sciencedirect.com/science/article/abs/pii/S1566253519308103"><em>[Arrieta et al. (2020)</em></a><em>] Arrieta, A. B., Díaz-Rodríguez, N., Del Ser, J., Bennetot, A., Tabik, S., Barbado, A., … &amp; Herrera, F. (2020). Explainable Artificial Intelligence (XAI): Concepts, taxonomies, opportunities and challenges toward responsible AI. Information fusion, 58, pp. 82–115.<br></em>[<a href="https://dl.acm.org/doi/abs/10.1145/3351095.3372870"><em>Sokol, K., &amp; Flach, P. (2020)</em></a>] <em>Sokol, K., &amp; Flach, P. (2020, January). Explainability fact sheets: a framework for systematic assessment of explainable approaches. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency, pp. 56–67.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dbb1ac608009" width="1" height="1" alt=""><hr><p><a href="https://medium.com/escience-center/how-to-find-your-artificial-intelligence-explainer-dbb1ac608009">How to find your Artificial Intelligence explainer</a> was originally published in <a href="https://medium.com/escience-center">Netherlands eScience Center</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I referred to the XAI paper in section 2 pointing to the reference as well to the link.]]></title>
            <link>https://medium.com/@e.ranguelova/i-referred-to-the-xai-paper-in-section-2-pointing-to-the-reference-as-well-to-the-link-56883a0bde68?source=rss-1c30c46a7632------2</link>
            <guid isPermaLink="false">https://medium.com/p/56883a0bde68</guid>
            <dc:creator><![CDATA[Elena Ranguelova]]></dc:creator>
            <pubDate>Fri, 25 Feb 2022 13:21:40 GMT</pubDate>
            <atom:updated>2022-02-25T13:21:40.097Z</atom:updated>
            <content:encoded><![CDATA[<p>I referred to the XAI paper in section 2 pointing to the reference as well to the link. Maybe do that for the XAI fact sheet as well? Now it&#39;s only the link in the text, then doesn&#39;t make sense to add it in the Reference section ;-). I have also struggled to add table to Medium, there are workarounds, but no one worked for one containing hyperlinks...</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=56883a0bde68" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Is an eScience Research Engineer simply a Computer Scientist in disguise?]]></title>
            <link>https://medium.com/escience-center/is-an-escience-research-engineer-simply-a-computer-scientist-in-disguise-f73f23741cb?source=rss-1c30c46a7632------2</link>
            <guid isPermaLink="false">https://medium.com/p/f73f23741cb</guid>
            <category><![CDATA[computer-science]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[research]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[science]]></category>
            <dc:creator><![CDATA[Elena Ranguelova]]></dc:creator>
            <pubDate>Thu, 06 Sep 2018 08:18:46 GMT</pubDate>
            <atom:updated>2018-09-06T11:23:47.812Z</atom:updated>
            <cc:license>http://creativecommons.org/licenses/by/4.0/</cc:license>
            <content:encoded><![CDATA[<h4>My personal view on the differences between Computer Scientists and eScience Research Engineers</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RPecqNsPMcKZ8I_tj16VCA.jpeg" /></figure><h3>Can a wolf change its mind by changing its coat?</h3><p>The origin of this proverbial question comes from the Latin motto:</p><blockquote>Lupus pilum mutat, non mentem.</blockquote><blockquote>The wolf can change its coat, but not its mind (character/disposition).</blockquote><p>Querying the ability of any person to change its innate being is a just question. Can a computer scientist change his mind by changing his title? Is an eScience Research Engineer simply a computer scientist in disguise?</p><p>I fell in love with computer science and engineering in high school and I have been in love since. The master degrees and a doctorate which followed were a mere formal manifestation and consequence of how I felt. In my career path, I had applied my knowledge in different ways reflected by different job titles: post-doctoral researcher, innovator, R&amp;D engineer. I consider myself a computer scientist and engineer and I currently work as an eScience Research Engineer (eSRE) and eScience coordinator. Many of my colleagues at the <a href="https://www.esciencecenter.nl/">Netherlands eScience Center</a> have similar careers.</p><h4><strong>Six aspects explaining my personal identity change</strong></h4><p>Working with many partners — other researchers, experts in their scientific domain — inevitably raises the question of who we are. Especially by partners who are computer scientists themselves! Being a “convert” myself, I see similarities but also differences between working as a computer scientist and as an eScience Research Engineer.</p><p>In this blog post I will explain my personal view on 6 aspects of this mysterious “identity change”:</p><ul><li><strong>Science:</strong> Is the focus on computer science, or on domain science?</li><li><strong>Software: </strong>Is the focus on the software itself, or on the impact of the software?</li><li><strong>Research:</strong> How much freedom do you have in developing your expertise area?</li><li><strong>Publication: </strong>Is publishing a main goal or not?</li><li><strong>Collaboration: </strong>Is collaboration an option or a requirement for doing research?</li><li><strong>Funding: </strong>Is applying for funding an option or a requirement for doing research?</li></ul><p>I will illustrate these six aspects in two stories: one about myself collaborating as a Computer Scientist, and one about myself collaborating as an eScience Research Engineer.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*iSf90cSRNy2SrF_I8hr7PA.jpeg" /><figcaption>Principal Investigators day in 2017. The eScience Research Engineers wear blue badges (contrasting nicely my red cardigan) and the various domain scientists wear white badges. Blue and white, the colors of the Netherlands eScience center, match nicely and the bearers of the different badges work nicely together!</figcaption></figure><h3>Collaborative science</h3><p>Modern science is a multidisciplinary effort. Computer science nowadays is often inspired by the needs and challenges of other sciences, which need computer science achievements in order to progress in the big data era. The keyword for a computer scientist and especially for an eSRE is <em>collaboration</em>.</p><h4>Collaborating as a computer scientist</h4><p>Not all computer scientists work with other domain scientists, but many do. While working as a computer scientist (computer vision researcher to be precise) I worked within <a href="https://cordis.europa.eu/project/rcn/58425_en.html">Europhlukes</a>, a joint project with marine biologists. They needed a tool for automated identification of individual humpback whales from images of their unique spots and patterns of the underside of their flukes. These spots and patterns are used for automated individual identification of these endangered mammals from large photo catalogs. The problem in biology gave me inspiration for computer vision research which would solve that problem. I have developed and published an <a href="http://muscle.ercim.eu/images/DocumentPDF/MP_260_Ranguelova_GVIP05_ext.pdf">algorithm for automatically identifying salient regions in an image</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/741/1*hzrWj-0ARNUFPPyQ4HmqAw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/751/1*8-asBtFELsEPiR0NjQmY4A.png" /><figcaption>Two pairs of images of two humpback whales. Each pair corresponds to the same individual. Every image is processed independently: after automatic segmentation of the flukes from the background, salient regions are detected automatically. The regions are then matched between a new image of a whale and all other images of known whales in a catalog. The matching regions in each pair of images are displayed using the same color.</figcaption></figure><p>Automating this step allowed the biologists to process faster their large image catalogs and spend their time and efforts on studying the population dynamics of the endangered species. I have met with the biologists a few times to discuss their problem, get feedback and train them on my software.</p><p>The same generic algorithm was later applied not only for identifying individual dolphins, newts, leather-back turtles and determining the wood type from microscopy images, but for identifying whether any digital photos subject to different distortions have captured the same object or scene.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/710/1*6cC32D_Xp6hKcWr9YYFKWg.png" /><figcaption>Regions detected independently on four images of the same scene. The first image is the original and the other three are being distorted by blur, decreased light and different camera angle respectively.</figcaption></figure><p>As a computer science researcher my aim was to publish my algorithm as a first author in a large computer vision conference or journal. Applying the algorithm to real images for real applications served as proof of the usability and increased the acceptance chances of my publications. I was the only developer of the software. I used a proprietary software (MATLAB) to create the tool and delivered an executable to my users.</p><h4>Collaborating as an eScience Research Engineer</h4><p>At the <a href="https://www.esciencecenter.nl/">Netherlands eScience Center</a>, I have worked on many projects requiring computer vision expertise. My first project in collaboration with neuroscientists from four medical centers, was <a href="https://www.esciencecenter.nl/project/biomarker-boosting">Biomarker Boosting</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/244/1*7ltJ6xYnoF6BQdJ9DlntAA.png" /></figure><p>A <em>biomarker</em> is a measurable indicator of some biological state or condition. For example, the volume of the <em>hippocampus</em> in the brain is a marker for Alzheimer’s disease. Smaller volume is strongly correlated with both age and cognitive impairment diseases. The larger the sample size, the better the predictive value of the biomarker, which can be achieved by combining data sets obtained by different medical centers. The aim of the Biomarker Boosting project was to build a reusable platform for sharing patient imaging data among medical centers to run a common analysis pipeline. I have worked on the core of the processing pipeline - automatic segmentation of the hippocampus.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/899/1*Fla0k1Hl4j7Szpl0nEEnbQ.png" /><figcaption>Visualizing the segmentation result of the hippocampus. The volume can be obtained by simple voxel count.</figcaption></figure><p>The developed algorithm was generic and could be used to segment any brain structure, but it was used with atlas data for the hippocampus. The pipeline was applied to four different cohorts collected to study dementia and mild cognitive impairment as that was the focus of the neurological researchers. I have worked together with the scientists in the same office on a regular basis (a day every week) during the project and we have kept in touch closely. We have presented and published the results of the project in neuroscience conferences and journals and I was often a co-author. I attended neurological conferences and made links with that community. We have worked together in the same software repository, the development was adhering to high code quality, testing and documentation standards. The goal was to deliver an open source software in order to increase the chances of community adoption and to serve as many neuroscientists as possible and even reach other domains.</p><h3>Similarities and Differences</h3><p>The above examples from my career illustrate the <em>similarities</em> and <em>differences </em>between working as a computer scientist and as an eScience Research Engineer as<em> </em>I experienced them.</p><p>On a first glance, the work of a computer scientist and an eSRE is the same, only the job title differs. But the main difference between the work of a computer scientist and an eSRE stem from their different <strong>focus.</strong></p><ul><li><strong>Science:</strong> Both a computer scientist and an eSRE work in scientific projects. Computer science is the science of interest of a computer scientist. Other scientific domains might present inspiration, challenge or serve as application domains, but are not the focus. eScience stands for enhanced science, an eSRE’s applies ICT expertise to facilitate breakthroughs in another science. The focus is on the domain research question.</li><li><strong>Software:</strong> Both aim to produce working software. Computer scientists write software implementing new research ideas and producing correct publishable results is sufficient. eRSE write software to solve a domain question. eRSEs take care of the usability and within (or even cross-) domain applicability of the software. While the computer scientists do not necessarily open their code (e.g. to have a competitive advantage), eSREs do this as main principle. The impact of the research software they produce should be as large as possible.</li><li><strong>Research:</strong> Computer scientists and eSRE are researchers. eSRE can do (applied) research in their expertise area if it fits the project, but always engineer research software. The main research in the project is the one aimed at answering the domain research question.</li><li><strong>Publication:</strong> Both publish their work. Publishing as a first author (at the beginning of the career) is the main goal of a computer scientist. eSRE also publish either as a first author about the eScience or often as a co-author in a domain publication, but that is not their major goal.</li><li><strong>Collaboration:</strong> Some computer scientists collaborate with other scientific domains researchers, but not all. For an eSRE this is one of the core job competencies — collaboration is much more intense and crucial for the success of the collaborative projects.</li><li><strong>Funding:</strong> Computer scientists (especially to progress in their career) apply as PI for research projects funding as their core task. Sometimes also eSREs apply for funding, although usually together with a domain scientist as a PI. The Netherlands eScience center is a funding organization, hence many of the (junior) eSREs do not need to worry about acquiring funds. Although by doing their job excellently, the happy PIs are investing money for further future work together.</li></ul><h3>What about the wolf?</h3><p>It is not so important what the job title is, what is important is the <strong>mission</strong> behind it. I have enjoyed working as a computer scientist, but looking back — Europhlukes, my favorite project, could be considered an eScience project. Working closely with domain scientists can be very challenging, but also very rewarding.</p><blockquote>Using ICT expertise to enable breakthroughs in science for the good of society is very rewarding.</blockquote><p>It seems that I haven’t changed my disposition, but I have changed into much more of a team player. If you are inspired by the possibilities to work <a href="https://www.esciencecenter.nl/funding">with</a> the eScience Research Engineers or want to be <a href="https://www.esciencecenter.nl/about/careers">part of the team</a>, come to work with us!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f73f23741cb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/escience-center/is-an-escience-research-engineer-simply-a-computer-scientist-in-disguise-f73f23741cb">Is an eScience Research Engineer simply a Computer Scientist in disguise?</a> was originally published in <a href="https://medium.com/escience-center">Netherlands eScience Center</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>