Research news 2020 2021 - Department of Computer Science
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Contents Research News 16 Video games evoke emotions that 4 even music and cinema fail to reach What prevents us from making the most of the IoT? Aalto 17 researchers developed ways Artificial intelligence for to make IoT truly ubiquitous dementia prevention 5 18 Finns’ use of data is growing How did Runeberg’s career progress? rapidly and increasing electricity AcademySampo helps find the consumption in the entire ICT sector answer from a huge amount of data 6 19 Artificial intelligence produces Alumna Juulia Suvilehto is studying data synthetically to help treat how coronavirus is affecting diseases like COVID-19 social behaviour and touching 7 20 Sleep apnea is a threat for Artificial intelligence model developed public health - students by Finnish researchers predicts developed an application that which key of the immune system recognizes its signs at home opens the locks of coronavirus 8 21 How vaping companies Will AI make us better humans? exploit Instagram for youth- Why we must be cautious oriented marketing? 22 9 What archaeological objects have Practices identified at Terveyskylä been found in different parts of for facilitating the adoption of Finland? Researchers have developed online health care services FindSampo to bring together the discoveries made by citizens 10 University collaboration brings 23 a scope of new possibilities AI predicts if storms will cause for Planmeca’s business blackouts many days in advance 11 23 Scientists develop a sound device Following e-cigarette conversations to allow monkeys living in a Finnish on Twitter using artificial intelligence zoo to play sounds and music Research news of 2020-2021 from the Department of Computer Science 24 at Aalto University 12 How is Instagram used by the leading New epidemic modelling facilitates health agencies to engage with the assessment of corona strategies public during the COVID-19 crisis? Copyright © 2021 Aalto University 13 Towards trustworthy AI and an autonomous Europe Design and cover illustration: Humans of Computer Science Matti Ahlgren 14 27 Published by Department of Computer Pekka Marttinen: It is very Science, Aalto University From her own little world to important to take good care of the other side of the globe cs.aalto.fi health and social services data 15 29 Anna Cichonska uses data science AI predicts which drug to develop better healthcare combinations kill cancer cells 1
31 Computer Science News 59 By studying software and ‘I’m happy I didn’t go anywhere service engineering, you will 50 else’ – HAIC scholars very satisfied learn truly useful skills Ten new Academy projects awarded with their studies at Aalto to the School of Science 33 61 Maarit Käpylä turned her 51 Forbes AI Award to research on childhood hobbies into a living Aalto students help develop digital image-creating GAN models services for women’s charities 35 and family support groups 62 Besides coders, computer School of Science announced science needs professionals 52 2021 equality implementation interested in humans Flexibility and enthusiastic plan, and funds new diversity teaching make the SECCLO and inclusivity projects 37 programme a great option Everyday choices: Nitin Sawhney, how 62 do we cooperate in times of crisis? 53 From Mars to machine learning The coronavirus pandemic encouraged 39 hundreds of experienced professionals 63 Meet the data agent: Richard Darst to study change leadership ‘For many people, it’s always been a crisis’ – a new Aalto 40 54 collective engages intersecting Meet the data agent: Enrico Glerean Nitin Sawhney receives an Academy dimensions of crises of Finland grant for analyzing and 41 reconstructing crisis narratives 64 Tuomas Sandholm applies game Alumnus Kyunghyun Cho wants theory in his companies and 54 to support women who study data encourages everyone to study AI Digital humanities information science and artificial intelligence infrastructures receive the Academy 43 of Finland’s roadmap status 65 Former Nokia executive Tero Ojanperä Career course gathered a diverse believes intelligent platforms could 55 range of views, alumni and students be next success story for Finland ELLIS inaugurates 30 research units around user experience at leading institutions across Europe 44 67 ‘The atmosphere at Aalto is 55 CS S4 Student Symposium served comfortable and you can do European Commission grants remote pizza, socializing and things at your own pace’ €4M funding for SECCLO – interesting discussions on AI Master’s Programme in Security 45 and Cloud Computing 67 If you study computer science, Samuel Kaski receives major you may end up developing better 56 international funding for the cancer treatments by applying AI CodeRefinery went online and development of AI methods attracted over a hundred participants 47 from all over the Nordics 68 Teaching creativity to computers Assistant Professor in Machine - Christian Guckelsberger wants 57 Learning Arno Solin nominated to equip AI with the motivation Mario Di Francesco elected Computer for Young Academy Finland to discover the unexpected Science teacher of the year 2020 69 48 57 Follow Computer Science Alexandru Paler: Creating software SCI Awards 2020 celebrate our research in Aalto University for quantum computers successes in a challenging year 70 58 Kahvit näppikselle Doctoral thesis: Technology that’s better at detecting hate speech may also increase security risks 2
Research News What prevents us from making the most of the IoT? Aalto researchers developed ways to make IoT truly ubiquitous For his doctoral dissertation, Óscar Novo Díaz researched the technical challenges that prevent us from using the full potential of the Internet of Things Research & Studies, published: 27.5.2020 In recent years, the Internet of ty, management, and data processing. implementing a system designed to Things (IoT) has become a reality in Many of those challenges are scattered enable interoperability across hetero- our society. It has helped automatize across the various layers of the Inter- geneous semantic IoT domains. our homes, healthcare, and industrial net protocol stack, making it difficult Due to the COVID-19 pandemic, Ós- manufacturing, to mention just a few to identify the crucial mechanisms car Novo’s public defence took place application areas. In the home envi- contributing to the cause. online. He was physically together in ronment, this can mean that your en- As an important part of his doctor- the same lecture hall with the kustos, ergy supplier reads your energy meter al research work, Óscar Novo and his Professor Antti Ylä-Jääski, and the automatically, or that you have a se- colleagues developed several solutions opponent, Professor Sasu Tarkoma, curity system in your home, which al- that address these problems. Particu- but everyone else followed the con- lows you to monitor and manage your larly, the researchers implemented an ference over Zoom. Novo thinks that property, simply by using a mobile app. IP-agnostic mechanism that enables everything went well and having the Óscar Novo Díaz, Doctor of Science conference online had its benefits. (in Technology) who recently graduat- ‘Normally, public defences are quite ed from Aalto University, says that IoT heavy and some people may be more can be seen as an extension of the In- interested in the presentation, while ternet, and one of its most essential as- others want to hear the discussion pects resides in its ubiquitous nature. with the opponent. Now, people could IoT devices are broadly available and “Right now, we have pro- just join and leave the conference fully connected to each other and the tocols for the Internet of whenever they wanted.’ Internet. Things, but we have to en- Professor Ylä-Jääski kept track of However, one of IoT’s main challeng- sure that all those proto- the number of participants during the es, too, lies in its ubiquitous nature; cols – in terms of security, event and he told Novo that, at some currently some technical challenges accessibility, and so on point, around 20–30 people were fol- prevent us from making the most of – are totally invisible and lowing it online. ‘That felt really nice.’ IoT’s potential. Óscar Novo recently work very smoothly“ Óscar Novo has been working on the defended his doctoral dissertation on IoT field for years, for which reason it this topic, and he focused on four spe- felt natural to select it as his disserta- cific categories; ubiquitous commu- tion topic, too. Currently, he is working nication, access control, concurrency as a researcher at a private company. ‘I control, and ubiquitous semantics. would like to move towards another ‘Right now, we have protocols for research topic but still stay connect- the Internet of Things, but we have ed to IoT and use my IoT knowledge. to ensure that all those protocols – in the connectivity and communication For example, artificial intelligence is a terms of security, accessibility, and so of IoT devices behind Network Ad- very promising and interesting area of on – are totally invisible and work very dress Translations. Moreover, they research.’ smoothly, without spending too many designed and implemented a resource Óscar Novo Díaz defended his disser- resources of the constrained IoT de- concurrency framework to manage the tation on 24 April 2020. The title of his vices,’ he explains. By the term con- concurrent access of shared resources dissertation is ‘Improving the Ubiq- strained, he means that inexpensive in IoT environments. uitous Capabilities of the Internet of devices, such as small sensors, run To improve the current access con- Things’. on batteries and have a relatively low trol situation, they also designed a computation power. decentralized access control manage- The decentralization of IoT, its het- ment system that relies on blockchain erogeneity in terms of protocols, tech- technology to control the authoriza- nologies and standards, has created tion policy of the IoT devices. Finally, several interoperability questions they addressed the problem of seman- from the perspective of connectivi- tic interoperability, proposing and By Anu Haapala 4
Research News Finns’ use of data is growing rapidly and increasing electricity consumption in the entire ICT sector The use of mobile data is growing fast in Finland, compared to other countries, re- veals a recent report. Research, published: 8.6.2020 IP traffic and data use by the indus- try. Consumers’ use of data services, creasingly significant part of Finland’s try and consumers have increased sig- streaming services in particular, is electricity consumption. While elec- nificantly in the last fifteen years. In a factor that has increased IP traffic tricity in Finland is a relatively pure Finland, the use of fixed data grew sev- most significantly. form of energy, other energy sources en-fold and the use of mobile data 25- As states, companies and different in the ICT industry should also be set fold during 2011–2017. In the last few types of organisations seek a shift to- on a sustainable basis. During a power years, the growth has slowed down, wards a low-carbon economy, the role cut, for example, reserve power comes but researchers believe that the use of the ICT sector is divided. On the one from heat and fuels based on fossil fu- of fixed and mobile data will keep on hand, it has been proposed that the ICT els that strain the environment,’ notes growing in the coming years. sector would seek more efficient ener- Timo Seppälä, Chief Research Scien- The information is revealed by a re- gy consumption and form a basis for a tist at Etla and Professor of Practice at cent report by Aalto University and greener economy. On the other hand, Aalto University. ETLA Economic Research (Etla), the growth in the ICT sector’s energy ‘Research shows that despite the which was released on Monday, 8 use and massive use of rare metals in strong growth of IP traffic and the June. In their report, researchers at usable devices has been perceived as a amount of data the electricity con- Aalto and Etla address how the energy threat to sustainable development. sumption has not increased at an and electricity consumption in the ICT equivalent rate. Instead the relative sector increased during 2011–2017. share of the ICT sector’s electricity In the last decade, the information consumption of the total electricity sector and software-based applica- consumption in European countries tions acquired an increasing role in the “In Finland, electricity seems to be strongly linked to the num- everyday lives of humans and organi- consumption represents ber of data centers. In other words, in sations. Digital services, mobile appli- as much as 68.7 per cent the light of the information that was cations and software are increasingly of the energy consump- gathered, the growth in electricity con- utilised in everyday communication as tion in the entire ICT sumption in the ICT sector is strongly well as companies’ operations. sector” focused on data centers, not so much In Finland, the use of mobile data on telecommunications networks,’ alone increased on average 72 per cent says Kari Hiekkanen, Research Fel- each year in 2011-2017. The share of low at Aalto University. total data use in Finland is currently at The researchers were able to form a approximately 29 per cent. Its share in In Finland, electricity consumption good overview of the energy and elec- relation to fixed network data is grow- represents as much as 68.7 per cent of tricity consumption in the ICT sector ing more rapidly than in the countries the energy consumption in the entire using open data from Statistics Fin- under comparison. The phenomenon ICT sector. In 2011-2017, electrici- land and Eurostat. Since open statis- is explained by two factors: the avail- ty consumption of the ICT sector in- tical data was available in great extent ability of fixed-price subscriptions creased a total of 2.2 per cent annual- in Finland as well as elsewhere in Eu- with limitless data as well as Finland’s ly. Within the ICT sector, the growth rope, they were also able to compare strong role as a developer of mobile has been particularly strong in data European countries. technologies. processing where the electricity con- The report has been written as part The increasing use of mobile data sumption increased by 130.9 per cent. of the Finnish Innovation Fund Si- among consumers has not, however, Data processing services include for tra’s theme of carbon-neutral circular resulted in growing household elec- example data centres, the number of economy and the Digital Disruption tricity consumption but as increased which keeps on increasing in Finland. of Industry research project, which is energy consumption in the entire ICT In 2017, the share of the ICT sector in funded by the Academy of Finland. sector. Watching videos on the phone, Finland’s entire electricity consump- in other words, does not increase the tion totalled one per cent. electricity consumption of just a sin- ‘The energy and electricity consump- gle consumer but the entire ICT indus- tion in the ICT sector will form an in- By Anu Haapala 5
Research News Artificial intelligence produces data synthetically to help treat diseases like COVID-19 The ability to produce data synthetically makes studying of the COVID-19 disease significantly easier. Research, published: 25.6.2020 Data driven technologies and ‘big veal trade secrets and inventions to er functionality. ‘There are still many data’ are revolutionizing many indus- their competitors. This is especially things we don’t know about the new tries. However, in many areas of re- true in drug development, which re- coronavirus: for example, we do not search – including health and drug quires lots of financial risk. If phar- know well enough what the virus caus- development – there is too little data maceutical companies could share es in the body and what the actual risk available due to its sensitive nature their data with other companies and factors are. When researchers have and the strict protection of individ- researchers without disclosing their synthetic data, we start understanding uals. When data are scarce, the con- own inventions, everyone would ben- these things better,’ says Kaski. clusions and predictions made by re- efit. FCAI researchers are now working searchers remain uncertain, and the The ability to produce data synthet- on a project in which they use syn- coronavirus outbreak is one of these thetic data to construct a model that, situations. based on certain biomarkers, predicts ‘When a person gets sick, of course, whether a test subject’s coronavirus they want to get the best possible care. test is positive or negative. Biomark- Then it would be important to have the “I myself would not ers can be for example certain types of best possible methods of personalized like to give insurance molecules, cells, or hormones that in- healthcare available,’ says Samuel companies my own dicate a disease. Kaski, Academy Professor and the Di- genomic information” ‘The original data set with which we rector of the Finnish Center for Artifi- do this has been publicly available. cial Intelligence FCAI. However, developing such methods of personalized healthcare requires a lot of data, which is difficult to obtain because of ethical and privacy issues ically solves these problems. In their surrounding the large-scale gathering previous study, which is currently be- “To solve this issue, re- of personal data. ‘For example, I my- ing peer-reviewed, FCAI researchers searchers at FCAI have self would not like to give insurance found that synthetic data can be used developed a new machine companies my own genomic informa- to draw as reliable statistical conclu- learning-based method tion, unless I can decide very precisely sions as the original data. It allows that can produce research what the insurance company will do researchers to conduct an indefinite data synthetically” with the information,’ says Professor number of analyses while keeping the Kaski. privacy of the individuals involved in To solve this issue, researchers at the original experiment secure. FCAI have developed a new machine The application that was published learning-based method that can pro- at the end of June works like this: The duce research data synthetically. The researcher enters the original data set method can be useful in helping de- into the application, from which the Now we are trying to reproduce the re- velop better treatments and to under- application builds the synthetic data- sults of the original research with the stand the COVID-19 disease, as well as set. They can then share their data to help of synthetic data and build a pre- in other applications. The researchers other researchers and companies in a dictive model from the synthetic data recently released an application based secure way. that was achieved in the original re- on the method that allows academics The application was released on the search,’ explains Joonas Jälkö, doctor- and companies to share data with each fastest possible schedule so that re- al researcher at Aalto University. other without compromising the pri- searchers investigating the Coronavi- The research conducted at FCAI is vacy of the individuals involved in the rus pandemic would have access to it funded by the Academy of Finland. study. as early as possible. Researchers are Many industries want to protect further improving the application, their own data so that they do not re- to make it easier to use and add oth- By Anu Haapala 6
Research News Sleep apnea is a threat for public health - students developed an application that recognizes its signs at home The app is currently being tested, and its usability and reliability will be improved based on test users’ feedback. Research, published: 2.7.2020 Sleep apnea is a very common sleep er’s snoring through a microphone and forming a home recording, it is not disorder characterised by breathing the sleeping position using the sen- possible to control that the user per- pauses and periodic snoring. Accord- sors of a smartphone or smartwatch. forms the measurement under optimal ing to some studies, as much as ten The most common clinical form of conditions. The accuracy of the apnea percent of the population suffer from mild and moderate sleep apnea is po- alerts has not been yet studied. Fortu- it, but up to nine out of ten cases re- sition-dependent; the apneas occur nately, the application makes it easy to main undiagnosed because symptoms mainly when sleeping on the back. If make repeated measurements, which are sometimes mild and there are a compensates for the situation and in- lack of resources in diagnostics. As un- creases the reliability of the results,’ treated apnea can cause severe health says Joni Gardemeister, one of the de- issues, having simple and inexpensive “Nine out of ten cases velopers. ways to monitor and screen the dis- remain undiagnosed In the morning, the app forms an ease would be important. because symptoms are overall result of the night’s events, giv- To solve this issue, Finnish computer sometimes mild and there ing an idea of the amount of snoring science students designed an Android are a lack of resources in and the number of breathing pauses application, which helps to identify diagnostics” in different positions. The results are the signs of sleep apnea at home. Joni saved in the application, making it easy Gardemeister, Jonatan von Martens to track them over a longer period. and Sowrov Doha, who study artificial SnoreTracker is based entirely on intelligence, machine learning and the features of an Android smart de- data science at Aalto University, devel- vice; hence, the user does not need any oped the app in collaboration with Dr. breathing pauses occur while the user additional devices. The app is suitable Tapani Salmi, a medical doctor, PhD, is sleeping on their back, the applica- for anyone who suspects they may be and specialist in Clinical Neurophysi- tion detects the situation and alerts suffering from the symptoms of sleep ology and Sleep Medicine. the sleeper to change position. apnea, or who simply wants to follow The SnoreTracker application, de- ‘Sensors on smart devices are sensi- their sleep. ‘It allows you to examine veloped by the team, monitors the us- tive and reliable. However, when per- your own snoring, pauses in breathing 7
Research News and sleeping at home on a low thresh- old – even if you have no opportunity How vaping companies exploit for actual medical sleep study,’ Garde- meister adds. Instagram for youth-oriented Diagnosing sleep apnea requires a marketing? polysomnography, usually performed Researchers use artificial intelligence to analyse hun- only during one night, either at home dreds of thousands of Instagram posts about vaping or in a hospital sleep laboratory. Mea- suring is laborious and expensive, and Research, published: 13.7.2020 in some countries, there are hardly any resources for making such mea- surements. For these reasons, the app E-cigarettes are highly addictive nicotine products with unclear health impacts, can be of significant help, as it makes particularly on youth. Instagram is a visual social media platform which is wildly it possible to monitor the sleep easily popular among adolescents. Researchers interested in public health at Aalto Uni- at home. versity in Finland studied how vaping is represented on the platform. By using ar- If someone suspects they have sleep tificial intelligence, they were able to analyse hundreds of thousands of posts from apnea, they should however always a 6-month period last year, and found that a large portion of posts are promoting contact health care as well. Untreated controversial flavoured e-liquids to young audiences. sleep apnea can cause daytime somno- The research analyzed Instagram posts shared between June to November 2019 lence, make the person more suscepti- with a caption “#vaping”. ‘We knew this would be predominantly promotional ble to accidents but also cause serious material,’ said Dr. Aqdas Malik from the Department of Computer Science who health risks. It increases the risk of studies social media and user behaviors with a specific interest in online discourse health problems such as hypertension, about public health ‘but we were interested in what types of images these would cardiac and neurological diseases. be, and who was posting them.’ In the end, the research team generated a database Sleep apnea is most common in peo- of over half a million pictures, which were sorted using a neural network, and sub- ple middle aged and over, and obesity sequently grouped the images into categories representing similar features. is the major risk factor. As popula- What the neural network tions age and levels of obesity increase showed was that over 40% in countries with poor resources for of the images – the larg- “By using artificial intelligence, sleep medicine, there is a risk of a est proportion out of the 6 they were able to analyse hun- sleep apnea epidemic, says Dr. Tapani categories – were of e-liq- dreds of thousands of posts from Salmi. uids or e-juices. These were a 6-month period last year, and The application is based on years predominantly posted by found that a large portion of posts of research work aimed at identify- Instagram profiles listed are promoting controversial fla- ing breathing disorders and develop- as business accounts. The voured e-liquids to young audi- ing algorithms and led by Dr. Salmi. prominence of posts about ences.” The accuracy of the methods has been e-liquids is interesting from studied at Helsinki University Cen- a public health perspective tral Hospital and the results have been because, although many published in the United States in the e-cigarette brands market themselves as “smoking cessation” devices, it has been Journal of Clinical Sleep Medicine, shown by other studies that flavoured liquids are strongly linked to adolescent Salmi says. ‘I suppose SnoreTracker users taking up vaping in the first place. The USA banned the sale of flavoured is the first project that helps monitor liquids at the beginning of this year specifically to help tackle the huge growth of and treat a disease this common with teenage vaping, and other countries are looking to do the same. the use of a mobile device,’ he notes. ‘While print and broadcast media has clear rules and regulations about what can The developers emphasize that and cannot be advertised, and what constitutes advertising, we don’t see this on SnoreTracker is not a medical de- social media,’ said Dr. Malik. ‘I believe we need stricter laws to regulate the visi- vice but an application that promotes bility of various substances - including vaping products on these networks. Any health, sleep and well-being. Howev- 12-year-old with a phone can easily create a social media account and bypass the er, with the help of feedback collected age-rules for seeing what’s posted there, and the potential health implications are from users, its usability and reliability significant.’ 60% of all the posts using the hashtag #vaping were from business ac- can be improved and developed. Peo- counts. Over 70% of Instagram users are under 35, and over 35% of its users are ple interested in the app can present- under 24 years old. ‘It’s a huge grey area in terms of advertising regulations, espe- ly download the beta version for free cially regarding promotion towards younger audiences’ Dr. Malik said. from Google Play app store. The research has been published in the International Journal of Medical Infor- matics, and is available to read online. By Anu Haapala, Illustration Matti Ahlgren By Anu Haapala 8
Research News Practices identified at Terveyskylä for facilitating the adoption of online health care services It is important to involve professionals in the fields of information technology and health care as well as patients in the development of services Research, published: 19.8.2020 The adoption of online health care on usability in the design process. In- sponsibility of just one person within services has been found very difficult volving patients, which is often con- an organisation. globally: an estimated 45 per cent of sidered a difficult feat, is also import- ‘Professionals are motivated by ser- services fail on account of resistance ant. vices that make it easier to work – for from health care professionals. In a ‘If we can involve even the small- example, having to enter data only study by Aalto University and Ter- est group for testing, for example, the once – as well as help patients. Accord- veyskylä that has just been published, feedback will help us to improve the ing to the interviewees, interest has numerous factors for facilitating the usability of services. It would also been shown towards a digital care path adoption of the services have been dis- useful to include professionals of us- that helps to identify a serious skin covered. The information will help in er-centered design or service design. symptom from home with the help of launching new online health care ser- They can identify hidden needs that the patient sending a photograph,’ Ku- vices in the future. the users are unable to report as well jala says. The study focused on the Terveys- as combine the needs of the different Researchers have known for a long kylä.fi online service used within spe- time that motivating health care pro- cialised Finnish private health care to fessionals as well as patients to use offers health care related information new online health care services is dif- and support for citizens, along with “It is necessary to rec- ficult. ‘The introduction of new digital tools for health care professionals. In ognise that adopting services causes stress in a hectic care addition to virtual houses focusing on services takes time and work environment, and professionals different health issues, accessible to resources” have not always received enough sup- everyone, Terveyskylä includes digital port in the change.’ care paths for certain patient groups. Yet digital services do, when func- ‘Since there are new digital care paths tioning properly, help to offer better on the way, we wanted to gather les- services and support for profession- sons from the first care paths to bene- parties into solutions. Patients can be als in their work. Services also cost fit the later ones,’ says Research Fellow included as part of the treatment on Finland a great deal. Therefore, it is and the first writer of the article Sari a continuous basis on a small scale, important to get professionals as well Kujala. but also in collaboration with patient as patients to use them. Doctors and The researchers interviewed em- organisations, thesis workers and re- nurses do not always remember to ployees of the university hospitals of searchers,’ Kujala says. communicate about the services to Helsinki, Turku, Oulu, Tampere and The study also states that services their patients. According to Kujala, Kuopio, inquiring after their experi- and changes related to them must be this would be important: according to ences regarding the adoption of ser- communicated continuously. Ade- research, having a professional recom- vices. The interviewees were mainly quate training must be provided to mend services to their patients has a nurses responsible for the adoption of professionals, along with guidance on huge impact on the use of the services. the digital care paths in their place of how to use the services, and any con- The study is a part of the DigiIN proj- work. They were selected among dif- cerns they raise must be taken seri- ect, which seeks to get all people in- ferent care paths and different hospi- ously. User feedback should also be volved in a digital society by renewing tals for the purpose of gathering les- collected and the services tested at all the service culture. The study is fund- sons from a range of environments. stages in order to enable continuous ed by the Strategic Research Coun- Based on the results they gathered, improvement of the services and to en- cil (SRC). The research article was the researchers conclude that when sure that everyone can understand the recently published in the Journal of new digital services are introduced in services. It is necessary to recognise Medical Internet Research. health care, it is essential to include that adopting services takes time and professionals of health care and infor- resources. This is why the adoption mation technology as well as experts should be planned well and not the re- By Anu Haapala 9
Research News University collaboration brings a scope of new possibilities for Planmeca’s business Comparison of the model segmentation and the ground truth, from the secondary test data annotations, for a CBCT scan. For further explanation, see the research article in Nature Scientific Reports. Research, published: 2.11.2020 Collaboration between the Finn- and make their work faster and more meca. ish Center for Artificial Intelligence efficient, allowing them to focus on The new deep learning method is FCAI, dental equipment manufactur- more challenging cases,’ says Profes- based on the training of deep neural er Planmeca and Tampere University sor Kimmo Kaski from Aalto Univer- networks with a dataset consisting of Hospital (TAYS) will soon make the sity. 3D cone beam CT (CBCT) scans. The work of radiologists, mouth and jaw Planmeca is one of the world’s lead- research results were published in the surgeons and dentists easier, in addi- ing companies when it comes to man- prestigious Nature Scientific Reports tion to making the treatment of mouth ufacturing dental equipment. For publication in spring 2020. illnesses even safer. Planmeca, the collaboration means Based on the results, the model can Together with experts from Plan- new and significant business opportu- effectively identify designated struc- meca and TAYS, researchers at FCAI nities. The company plans to include a tures in images. The method located have developed a method to help lo- new feature in its existing software to mandibular canals on a detailed level cate the nerve canal of the lower jaw, ensure that the method will be used by and was able to beat statistical shape i.e. mandibular canal. This facilitates healthcare professionals. models, which have previously been the placement of teeth implants, for Vesa Mattila, Senior Advisor of determined as the best automated example. A dentist placing a tooth im- method of locating mandibular canals. plant must know the exact location of It was important for Planmeca that the mandibular canal in order to plan “The new method will the collaboration involved FCAI, Aal- the size and position of the implant be a tool for healthcare to University and Tampere Univer- and the entire procedure. professionals, not their sity Hospital. Scientific publications Specialised dentist Jorma Järnst- replacement” are almost indispensable to verify the edt from Tampere University Hospi- functionality of the method and gain tal specialises in radiology and den- research status. This and the collabo- tal, oral and maxillofacial diseases. He Medical Imaging at Planmeca Group, ration with Tampere University Hos- reveals that for a hospital radiologist, has also followed the project with in- pital helped to gain patient data for re- locating the nerve canal is a routine terest. He highlights that the entire search purposes, which is essential in procedure that must be done for each collaboration began with the needs of this type of study. patient. professionals of clinical work. ‘At the Vesa Mattila points out that because Normally the task is very time-con- time, the project had a great chance of the field of dental care has moved suming, and automating it reliably succeeding, not just technically and into 3D imaging, the amount of data with the help of artificial intelligence operationally, but also commercially.’ has jumped significantly. At the same makes the work of healthcare profes- time, there is a need for more radiol- sionals faster and easier, in addition “The industry needs tools ogists. ‘Thus the industry needs tools to making the treatment safer. ‘An al- such as this one” such as this one.’ gorithm that is medically proven to All parties are grateful for their col- be reliable makes the patient’s treat- Artificial intelligence methods have laboration partners. Although collab- ment safe while reducing unpleasant, taken significant leaps in the last oration with instances from different although very rare, surgical injuries,’ years. As for Planmeca, the company fields is always challenging at first, Järnstedt says. has for long been developing technol- initial problems were quickly over- The benefit of artificial intelligence ogy that is important for the company, come. ‘In the initial stage, it felt like we is that it does the work accurately ev- 3D X-ray imaging. ‘3D imaging is the weren’t speaking the same language, ery time. The new method will be a pillar of our business, and we want to but we are now. This is very important tool for healthcare professionals, not utilise it and commercialise the prod- for the project,’ Mattila says. their replacement. ‘The goal of the re- ucts that have been created using it,’ search work is not to replace radiolo- says Vesa Varjonen, Vice President gists but to create a tool to assist them of Research and Technology at Plan- By Anu Haapala 10
Research News Scientists develop a sound device to allow monkeys living in a Finnish zoo to play sounds and music The use of sounds is a promising way to improve the life and wellbeing of animals living in captivity Research, published: 7.10.2020 Sound is a promising way to stim- the sound system regularly throughout toral researcher at Aalto University, ulate zoo animals and increase their the study and, after the first few days, says that the result was both unex- welfare, as shown by a study from also began to sleep, groom, and so- pected and expected. ‘I’ve been work- Aalto University collaborating with cialize with other monkeys inside the ing with animals for a long time and Helsinki’s Korkeasaari Zoo. Research- sound device. learned to keep an open mind. Howev- ers built a device that plays different Kirsi Pynnönen-Oudman, re- er, given that the sound of traffic is so sounds for a group of white-faced saki search coordinator at Korkeasaari unfamiliar to the sakis and not related monkeys to listen to in their enclosure. Zoo, says that food is often used to to their normal daily life -- unlike, for They were interested in the saki mon- enrich the lives of animals in zoos. example, the sound of rain -- it came as key’s behaviour: did they want to use However, non-food stimuli must also a surprise that they were interested in the device, when would they use it, and be provided, especially for small an- those sounds.’ what would they choose to listen to? imals that are prone to gain weight. According to Dr. Hirskyj-Douglas, Little is known so far about the sound ‘Sounds are really important to many the result suggests that scientists need preferences of primates, so the re- to think carefully about what tech- searchers wanted to choose as var- nologies work for animals, keeping an ied a soundtrack as possible. The re- open mind and leaving the human per- searchers chose to play four different ception at the door. soundtracks, avoiding sounds of ani- In addition to influencing what mals that prey on sakis, or sudden and “Sounds are really im- sounds they want to listen to, the loud noises. portant to many animals white-faced sakis were able to influ- The Animal-Computer Interaction in their communication ence the final form of the device. The research team at Aalto University’s with each other. The researchers initially tested two proto- Department of Computer Science rainforest is full of dif- types and built the final sound device designed and built a tunnel-shaped ferent sounds and little from the option that the monkeys used sound device of wood and plastic for monkeys are supposed to significantly more. small monkeys, which was placed in be sensitive to different The tunnel-like box with a plywood their own residential area in the Am- sounds” floor and a transparent acrylic roof azon House of Korkeasaari Zoo. The was the sakis’ favourite. Both mate- monkeys were able to turn the device rials were already familiar to white- on themselves and therefore decide to faced sakis, and thanks to the shape listen to sounds or remain in silence. and materials of the box, the tunnel ‘We didn’t measure welfare fac- acoustics were good without having to tors, but from a welfare perspective, completely close the space. The shape it was essential to give animals de- animals in their communication with and material were suitable for their cision-making power and indepen- each other. The rainforest is full of dif- habitat and to support the species-spe- dence. These issues have been linked ferent sounds and little monkeys are cific behaviour of the animals. to well-being in other studies. Instead supposed to be sensitive to different Dr. Pynnönen-Oudman from Kork- of just playing music and other sounds sounds.’ easaari Zoo says that sound stimuli in their living environment and see- The fact that animals are so inter- have been used in zoos to some extent, ing how it works, we gave them a sys- ested in sounds is also reflected in the but so far quite little has been tailored tem that they could approach and use results of the study. The sakis were al- to particular species. ‘This could very themselves,’ says Roosa Piitulainen, lowed to choose how much either calm well be the way of the future to stimu- the first author of the research paper music, fast-paced electronic music, late the life of zoo animals.’ and doctoral candidate at Aalto Uni- sounds of rain, silence, or traffic nois- versity. es they prefer to listen. The sakis pre- The researchers followed how the ferred the sounds of traffic over all oth- monkeys used the tunnel for several er sounds. months. They found that the sakis used Ilyena Hirskyj-Douglas, postdoc- By Anu Haapala 11
Research News New epidemic modelling facilitates assessment of corona strategies Carried out as a joint project involving Finnish, Swedish and Norwegian researchers, the model also takes into account network structures and human mobility. Research, published: 13.11.2020 The NordicMathCovid project aims Traditional epidemic modelling does skelä explains. to model corona and future epidemics not take into account the network In addition to examining the spread more extensively than has been previ- structure, geographical location or hu- of communicable diseases, the data ously attempted. It also builds towards man mobility. Modern network theory and models may also be used to com- long-term cooperation in mathemati- provides computational methods for pare diverse vaccination strategies. cal modelling and extensive collection modelling population contact struc- ‘Mathematical modelling allows of health data. tures, which is needed in order to as- to explore and try to understand the ’One of the purposes of the project is sess, for example, the contribution of spread of the epidemic and also the school closures towards slowing down question of who should be vaccinat- the epidemic. ed first,’ says Professor Tapio Ala- ‘We are studying large populations. Nissilä. We do not assume that individuals The research can also be used to ex- are associated to each other on an en- amine the structure of the infectious tirely random basis; instead, we apply disease situation. knowledge about how social networks ‘Our research can tell us, for exam- are usually shaped: some people, such ple, how working remotely affects the as superspreaders, have more con- spread of the disease,’ Kivelä explains. tacts than others. In addition, social The project is led by Professor Tom networks are clustered, which means Britton from the University of Stock- that the connections are interlaced,’ holm and includes researchers from “Mathematical model- explains Professor Mikko Kivelä. Aalto University and the University of ling allows to explore The large variations in contacts, mo- Oslo as well as the countries’ national and try to understand bility and social activity in different health institutes: the Finnish Insti- the spread of the ep- population groups have a significant tute for Health and Welfare (THL), idemic and also the impact on the spread of the epidemic the Public Health Agency of Sweden question of who should and the formation of immunity. In or- and the Norwegian Institute of Public be vaccinated first” der to understand these phenomena, Health (NIPH). In addition to the sto- the project will develop new stochas- chastics research group led by Lasse tic models. Leskelä, the project also includes Mik- ko Kivelä’s network science research Data from different sources group and Tapio Ala-Nissilä’s compu- tational physics research group. The researchers are utilising an as The budget for the two-year project diverse as possible range of realistic comes in at just under EUR 1 million. and real-time medical, physical and The project is funded by NordForsk. social data. At the general level, Statis- The name of the project is ‘Data tics Finland is providing data related streams and mathematical modelling to people’s mobility, with telecommu- pipelines to support preparedness and nications operators providing more decision making for COVID-19 and fu- detailed data. Vehicle data can be ob- ture pandemics’. An objective for the to compare different corona models tained from road traffic. Local author- future is to extend the cooperation in and scenarios in different countries. ities can provide structural data on mathematical modelling to also in- For example, we can apply Swedish schools, which provides information clude Denmark, Iceland and the Baltic figures to conditions in Finland and on which areas the pupils of different countries. Norway or see what would have hap- schools are drawn from. pened if Sweden had acted differently,’ ‘The data is always stored on the se- says Professor Lasse Leskelä from Aal- cure servers that belong to its owner or to University. to the CSC IT Center for Science’, Le- By Tiina Aulanko-Jokirinne 12
Research News Towards trustworthy AI and an autonomous Europe Professor Michela Milano is the Deputy President of EurAI and one the keynote speakers at AI Day 2020 Research, published: 15.10.2020 Europe needs to be autonomous in start-ups in Europe after they become straightforward, showing causal re- terms of artificial intelligence (AI) and successful, I believe Europe will have lations – that outcome Y happens be- create its own AI technology that re- a very important and competitive ad- cause of the factor X and these events spects citizens’ privacy, says Michela vantage with respect to other big play- do not simply co-occur – is much more Milano, Professor at the University of ers.’ demanding. Bologna and one of the leading AI re- Taking privacy issues seriously is ‘There are many, many aspects that searchers in Europe. crucial. “For example in China, priva- are considered important from the The European Commission wants cy is not considered important at all European point of view. Europe wants European experts to build AI that peo- and personal data are shared and used to shape a strategy that differentiates ple can trust, and according to EU’s without any consideration. Europe from other economies, like the US and ethical guidelines, trustful AI is law- should keep on taking these values and China that have made huge invest- ful, ethical, and robust. As the Depu- aspects very carefully into account to ments on AI. Europe is lagging behind ty President of European Association make something really different.” a bit, but I think that the strategy that for Artificial Intelligence (EurAI), Building trustworthy AI is more com- is going to build trustworthy AI in re- Professor Milano has a central role in sponse to ethical principles is a good this work and she supports the com- path.’ mission in shaping its AI strategies. Professor Milano has been an AI re- She leads also the interdepartmental searcher for much longer than AI has research institute for human-centred been the hot topic it is these days. She artificial intelligence, Alma AI, in Bo- “Europe wants to shape remembers the time when people com- logna, and in November, she will speak a strategy that differen- monly thought that AI methods were about the European AI strategies at AI tiates from other econ- useless and researchers even avoided Day 2020. omies, like the US and using the term to avoid scaring funders Europe needs to be able to compete China that have made away. ‘But now of course everyone un- against large economies, such as the huge investments on AI” derstands that it’s useful and can bring US and China, in terms of AI. Accord- really important results and impacts ing to Milano, its advantages include in all aspects of our life.’ the strong university system, the long Milano researches decision-support history in conducting research of top systems and especially systems that quality, and an industrial ecosystem support – rather than replace – hu- that includes start-ups, big corpora- man experts. What Milano thinks is tions, and everything in between. plex than what it may sound. Milano particularly fascinating in AI is that ‘We have all the components that are points out that many AI systems “sim- researchers can embed human knowl- needed to create a significant strate- ply work”, while we do not complete edge in AI models. ‘You can really use gy and to really create an autonomous understand why they work. To create these systems to put together knowl- Europe. But there are problems, too: trustworthy AI systems, researchers edge from different experts, different investments and the fact that start- need to develop AI that is understand- domains, and glue it all together with ups can start here but as soon as they able and explainable to humans. data-driven models.’ become successful, they are bought ‘We also need to have more collabora- One aspect that fascinates her is the by Google, Facebook, or another big tion between AI systems and experts, interdisciplinary nature of AI and, for American player,’ Milano notes. which is very difficult. That requires example, ethical aspects that call for Another significant problem for Eu- encapsulating the collaboration and multidisciplinary approach. ‘When rope is brain drain. ‘Talents are leaving interaction with an expert. The system you are in a driverless car, you need to Europe because they can find better needs to learn from humans how they consider aspects that are not just al- salaries and better conditions outside solve problems, and that knowledge gorithmic or technological. You really of Europe. We really need to retain tal- needs to be injected in the AI systems.’ have to understand that there are diffi- ent. If we have a strategy with an eco- When we are developing such sys- cult decisions to make.’ system that is favourable for them, and tems, we also face the issue of causal- When e.g., lives are in question, mod- for creating start-ups and keeping the ity; while calculating correlations is elling a system is particularly difficult 13
Research News Pekka Marttinen: It is very important to take good care of health and social services data The DataLit project develops understandable and reliable practices for using health and social services data and ethical and moral principles should come into place – and at the same time, Research, published: 16.11.2020 there are many situations in which peo- ple from different parts of the world do Assistant Professor Pekka Marttinen is part of the new DataLit research con- not follow any global moral guidelines. sortium headed by Professor Petri Ylikoski of the University of Helsinki. The While East Asians may think that it is project aims to develop comprehensible and reliable practices for using Finnish most important to protect the elderly, data on health and social services. Europeans put children’s safety first. ‘My task is to develop machine learning methods and models for health and so- How to code these types of decisions cial services data. With the help of these methods and models it is possible to will be one of the key challenges in AI examine societal questions related to health care’, Pekka Marttinen says. research in the near future. On a posi- The DataLit project brings together skilled people from many different fields. tive note, a machine can be more ratio- ‘In addition to the methods, it is important to examine what the models are used nal than human decision makers can, for and who use them, talk about them, and make decisions on how they are used. as it can process information so quickly As it is often said, the decisions must not be left to the engineers alone’, Martti- that it can genuinely base its decisions nen says. on reasoning, while humans use their The project analyses key ethical concepts – such as privacy – in a range of tech- instinct. nical, societal, and legal contexts, while examining how officials can use the data In the last few years, artificial intelli- in a socially sustainable manner. DataLit also produces tools for the responsible gence systems have developed at an ex- and reliable use of register data, for example forecast models to support deci- tremely fast pace. Professor Milano is sion-making, and methods for anonymising data. eager to see what types of results will be ‘It is especially important to take good care of data. This means data security, achieved in the next ten years; she hopes consideration of ethical questions, the implementation of data analysis in a pro- to witness a European AI field that is tected environment – in general, the secure use of data. The use must be trans- truly strong and competitive. parent and acceptable’, Marttinen says. Another thing she wishes to see is The first phase of the project will take three years and its total budget is € 3.9 that researchers start using AI systems million. In addition to the University of Helsinki and Aalto University, the Uni- more to solve problems that help make versity of Eastern Finland and several other cooperative partners are involved in the Earth a better place for everyone; the project: The Finnish Institute for Health and Welfare (THL), The Social In- technology can help us reduce pollution surance Institution (Kela), the Joint municipal authority for North Karelia so- and emissions and hit the zero emission cial and health services (Siun sote), the Cancer Society of Finland, the Ministry and carbon neutrality targets. of Finance, Statistics Finland, the Digital and Population Data services Agen- ‘It is time for us to use technology for cy, and the Helsinki Institute for Social Sciences and Humanities (HSSH). The the good and help the environment and project is implemented in close cooperation with the Finnish Center for Artifi- improve our citizens’ quality of life. cial Intelligence (FCAI). Technological solutions really help us ect they have tested the new solution with several collaborators. One of those to go towards a better planet for us, our collaborators is Elisa, a Finnish company offering telecommunications services, children, and the next generations to which announced a demo application developed during the collaboration. come.’ One aim of the Aalto computer scientists is to try to find the customers who AI Day 2020 took place online on 26 will gain most benefit from the solution. ‘Our goal is to start a company that November 2020. brings the solution to the market.’ says Olavi Mertanen. The CloudXR project in question is a TUTLI (From Research to Business) project funded by Business Finland and Aalto University that ended at the end of March 2020. The team has focused on exploring potential commercialization paths for XR applications’ remote rendering and developing the technology fur- ther. By Anu Haapala By Anu Haapala 14
Research News AI predicts which drug combinations kill cancer cells A machine learning model can help us treat cancer more effectively Research, published: 1.12.2020 When healthcare professionals treat curately predicts how combinations drug combination selectively inhibits patients suffering from advanced can- of different cancer drugs kill various particular cancer cells when the effect cers, they usually need to use a combi- types of cancer cells. The new AI mod- of the drug combination on that type of nation of different therapies. In addi- el was trained with a large set of data cancer has not been previously tested. tion to cancer surgery, the patients are obtained from previous studies, which ‘This will help cancer researchers to often treated with radiation therapy, had investigated the association be- prioritize which drug combinations medication, or both. tween drugs and cancer cells. ‘The to choose from thousands of options Medication can be combined, with model learned by the machine is ac- for further research,’ says research- different drugs acting on different tually a polynomial function familiar er Tero Aittokallio from the Insti- cancer cells. Combinatorial drug ther- from school mathematics, but a very tute for Molecular Medicine Finland apies often improve the effective- complex one,’ says Professor Juho (FIMM) at the University of Helsinki. ness of the treatment and can reduce Rousu from Aalto University. The same machine learning ap- the harmful side-effects if the dosage The research results were published proach could be used for non-cancer- of individual drugs can be reduced. in the prestigious journal Nature ous diseases. In this case, the model However, experimental screening of Communications, demonstrating that would have to be re-taught with data drug combinations is very slow and the model found associations between related to that disease. For example, expensive, and therefore, often fails drugs and cancer cells that were not the model could be used to study how to discover the full benefits of combi- observed previously. ‘The model gives different combinations of antibiot- nation therapy. With the help of a new very accurate results. For example, ics affect bacterial infections or how machine learning method, one could the values of the so-called correlation effectively different combinations of identify best combinations to selec- coefficient were more than 0.9 in our drugs kill cells that have been infected tively kill cancer cells with specific ge- experiments, which points to excel- by the SARS-Cov-2 coronavirus. netic or functional makeup. lent reliability,’ says Professor Rousu. Researchers at Aalto University, In experimental measurements, a cor- University of Helsinki and the Uni- relation coefficient of 0.8-0.9 is con- versity of Turku in Finland developed sidered reliable. By Anu Haapala, a machine learning model that ac- The model accurately predicts how a Illustration Matti Ahlgren 15
You can also read