{"id":25428,"date":"2020-02-27T00:00:00","date_gmt":"2024-09-13T08:48:43","guid":{"rendered":"https:\/\/wordpress-583806-4798031.cloudwaysapps.com\/3-x-ml-in-public-health-and-care\/"},"modified":"2024-11-15T16:21:51","modified_gmt":"2024-11-15T15:21:51","slug":"3-x-ml-in-public-health-and-care","status":"publish","type":"post","link":"https:\/\/informator.se\/en\/3-x-ml-in-public-health-and-care\/","title":{"rendered":"3 x ML in Public Health and Care"},"content":{"rendered":"\n

Health, pharma, and care expose Machine Learning to yet another stress test in practice, which adds another vital quality attribute to architects\u2019 QA-list (along with explainability, security, safety, accuracy, etc.)  – privacy-friendly ML.<\/em><\/p>\n\n\n\n

Three current examples from the health realm <\/strong><\/h3>\n\n\n\n

1.<\/strong> Most public authorities worldwide are using outdated analytics tools in forecasting and warning. Canadian AI-based warning system <\/a>BlueDo<\/a><\/strong>t<\/strong><\/a> predicted the corona<\/strong> epidemic<\/a> , and where it would travel to next. It was weeks ahead of the WHO (not to mention China: just scroll down on BBC<\/a> News, to \u201cAn epic political disaster\u201d). The persecuted and deceased whistleblower Dr Li Wenliang\u2019s<\/a> words \u201cA healthy society should not only have one voice\u201d happen to reflect a universal principle worth considering in architectures as well. To pick a handful: Triple Modular Redundancy (in mission-critical onboard systems etc.) , P2P (Peer to Peer architecture pattern), SOA, decision forests<\/a>, and not least, Federated Learning (FL, see below). The more \u201cbiodiversity\u201d, the more robust, fit, and sustainable the outcome; be ecosystems, social systems, public health systems, or ML-driven systems. <\/p>\n\n\n\n

2. <\/strong>AI-based selftests <\/strong>for elders will provide early warnings to persons at risk of cognitive <\/strong>disorders. An R&D<\/a> project at the Blekinge Institute of Technology will save time and resources by cutting lengthy diagnostics procedures from months\/years to hours. It\u2019s learning from a 20-year long series of relevant data about 10,000+ persons. The Nordic region is suited for medical ML, because data from long-term studies is available about several diseases. Often, the data spans decades, yet with a relatively small drop-off. <\/p>\n\n\n\n

3. Pattern recognition in image-based diagnostics.<\/strong> Another R&D<\/a> project at the Karolinska Institute in Stockholm is using mobile solutions and AI to make medical diagnostics accessible geographically & organizationally, safe, and accurate for several diseases: ML is learning to recognize cancer, malaria, schistosomiasis (bilharzia), soil-transmitted infections, pneumonia, or to classify burns. <\/p>\n\n\n\n

Privacy <\/strong>matters\nto both individuals and policymakers. Patient data takes privacy requirements\nan order of magnitude higher than online advertising or recommender engines\ndid. <\/p>\n\n\n\n

In addition to privacy, Federated ML (FL<\/strong>) addresses even security, access control, running or testing some nodes while training others, and learning from multiformat multitenancy data. A current paper by Ericsson<\/a> Research also points out that FL can reduce the size of training data needed, as well as the necessary transfers and thus network footprint in runtime (by up to 99%). This is sweet music for architects. So, how is FL orchestrated?<\/p>\n\n\n\n

First, let\u2019s step back to decision forests<\/strong> (see last diagram and paragraph of this<\/a> post), which use a simpler and more transparent distributed logic where independent decision trees are logical units (and in distributed environments, they might even correspond to HW\/MW units); however, the interactions (see Sequence Diagram) are one<\/strong>-way. Instead of sending (sensitive) training data, each node sends its decision (sometimes with a probability value) to the center. The center counts these as \u201cvotes\u201d, to select a \u201cwinner\u201d decision (sometimes using probabilities as weights of each tree\u2019s \u201cvote\u201d), which in turn becomes the output of the forest. <\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n
\"\"<\/figure>\n\n\n\n

FL <\/strong>can use a\nsimilar topology, where edge devices or servers become nodes, but interactions (see Sequence Diagram) go\nin both<\/strong> directions. The Ericsson team put it\na similar way a multimodel-database\nvendor in big data probably would: \u201cbring the\ncomputation to the data\u201d. Instead of bringing data (sensitive, especially in\nhealth) to a central node, FL trains an algorithm across several decentralized nodes\nthat own<\/strong> their training data locally<\/strong>, without sharing it. <\/p>\n\n\n\n

This results in a number of\nlocal models. These are sent (as vectors of weights) to the central node and\nreconciled there, into a common model (but extended algorithms can reconcile even\nwithin a Peer-to-Peer pattern). Then, this common weight vector is sent back<\/strong> to all nodes, each node upgrades\nits (neural) weights, and resumes the learning at this new level.<\/p>\n\n\n\n

This\nlearn-reconcile-update loop is repeated until the weights converge across\nnodes. Instead of the sensitive training data, only (intermediate) results of each\ntraining round are thus sent back and forth, as weights.<\/p>\n\n\n\n

We can expect some new \u201cmixed flavors\u201d in a near\nfuture to borrow features across ML techniques, such as NN-based leaf nodes (from\nsoft trees) or neural decision forests.<\/p>\n\n\n\n

FL<\/strong> combines the privacy <\/strong>of local <\/strong>ownership of data with the accuracy <\/strong>built by ML from big <\/strong>data. Along with explainability and accuracy, privacy<\/strong> is essential in health apps of AI. This is one of the main reasons why FL is more viable than big-data transfers. Even more so in the EU\/EES marketplace, where tech firms soon need to meet new<\/a> EU requirements<\/strong> on AI\/ML and big data. \u00a0\u00a0<\/p>\n<\/span><\/span>\r\n

Milan Kratochvil\u00a0<\/span><\/span><\/h4>\r\nTrainer at Informator, senior modeling and architecture consultant at Kiseldalen\u2019s, main author: UML Extra Light (Cambridge University Press) and Growing Modular (Springer), Advanced UML2 Professional (OCUP<\/a> cert level 3\/3).\n\nMilan and Informator collaborate since 1996 on architecture, modeling, UML, requirements, rules\/AI, and design. You can meet him in the coming months at public courses (in English or Swedish) on\n

AI, Architecture, and Machine Learning<\/a><\/p>\nModular Product Line Architecture<\/a>\n

Avancerad objektmodellering med UML<\/a><\/p>\nAgile Architechture Fundamentals<\/a>\n\nAgile Modeling with UML<\/a><\/span><\/span>","protected":false},"excerpt":{"rendered":"

Health, pharma, and care expose Machine Learning to yet another stress test in practice, which adds another vital quality attribute to architects\u2019 QA-list (along with explainability, security, safety, accuracy, etc.)  – privacy-friendly ML. Three current examples from the health realm 1. Most public authorities worldwide are using outdated analytics tools in forecasting and warning. Canadian […]<\/p>\n","protected":false},"author":6,"featured_media":26467,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_monsterinsights_skip_tracking":false,"footnotes":""},"categories":[570,922,572],"tags":[591,594,592],"class_list":["post-25428","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-blogg","category-blogi","category-it-arkitektur","tag-ai","tag-arkitektur","tag-machine-learning"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t