2024

2023

2022

2021

2020

2019

2018

2017

Portefølje Naturvitenskap og teknologi ikon

The research system

Portefølje Humaniora og samfunnsvitenskap ikon

Ground-breaking research

Portefølje Samisk ikon

Sámi society and culture

Portefølje Demokrati, styring og fornyelse ikon

Democracy and global development

Portefølje Velferd, kultur og samfunn ikon

Welfare, culture and society

Portefølje Helse ikon

Health

Portefølje Klima- og polarforskning ikon

Climate and environment

Portefølje Landbasert mat, miljø og bioressurser ikon

Land-based food, the environment and bioresources

Portefølje Energi, transport og lavutslipp ikon

Energy and transport

Portefølje Muliggjørende teknologier ikon

Enabling technologies

Portefølje Industri og tjenestenæringer ikon

Innovation

LTP Climate, environment and environmentally friendly energy

Climate and loss of natural diversity

Internationalization

Societal security and civil preparedness

Renewal in the public sector

Velferdstjenester

Frontpage Image

Marine icon

Petroleum og mineraler

LTP Hav ikon

Oceans and coastal areas

LTP2 Muliggjørende og industrielle teknologier ikon

Enabling and industrial technologies

Trust and community

Professional environment

High quality and accessibility

LTP Climate

Strengthened competitiveness and innovation capacity

LPT Health

 An innovative business community

LTP3 Helse

Portefolje Petroleum Ikon

Portefølje Hav ikon

Portefølje Utdanning og kompetanse ikon

Portefølje Global utvikling og internasjonale relasjoner ikon

Portefølje Livsvitenskap ikon

LTP2 Styrket konkurransekraft og innovasjonsevne ikon

LTP2 Utvikle fagmiljøer av fremragende kvalitet

LTP2 Samfunnssikkerhet og samhørighet ikon

LTP2 Fornyelse i offentlig sektor ikon

LTP2 Klima, miljø og miljøvennlig energi ikon

Forsidebilde optimalisert

project-bank-technical

Et innovativt næringsliv ikon

Muliggjørende teknologier bakgrunn

Marin bakgrunnsbilde

Marin ikon

Forsidebilde

Statistics

Prosjektbanken presenterer statistikk og informasjon om prosjekter med finansiering fra Forskningsrådet fra 2004. Statistikken er fordelt på blant annet tema, år, geografi og organisasjon. Prosjektbanken gir også detaljert informasjon om de enkelte prosjektene.

...or see the summary and other information about the projects you have selected

Portfolios

En god beskrivelse av denne siden for søkemotoroptimalisering

Forside

How the grants are distributed between organisations, counties, subject areas etc....

Here are some examples of searches you can perform in the Project Databank

All long term plan objectives

About Text Queries

No thematic area or topic related to the project

has received funding in the programme period

No popular science description is available

have received funding in the programme period

The project with project number {{project number}} was not found

Technical description

Woops, something went wrong. Please refresh the page. If the error persists, please contact us.

We could not find what you were looking for

See statistics related to this LTP objective

See projects related to this LTP objective

related to this LTP objective have received grants

has been granted to projects related to this LTP objective

has been granted to projects related to this portfolio

related to this portfolio have received grants

About Prosjektbanken

Kunnskapsdepartement

Samferdselsdeparteme

Nærings- og fiskerid

Diverse

Justis- og beredskap

Energidepartementet

Klima- og miljødepar

Kommunal-og distrikt

Landbruks- og matdep

Forsvarsdepartemente

Digitaliserings- og 

Teknologi

Agder

LTP3 Styrket konkurransekraft og innovasjonsevne

LTP3 Et kunnskapsintensivt næringsliv i hele landet

LTP3 Høy kvalitet og tilgjengelighet

LTP3 Fagmiljøer og talenter

LTP3 Klima, miljø og energi

LTP3 Miljøvennlig energi og lavutslippsløsninger

LTP3 Muliggjørende og industrielle teknologier

LTP3 IKT og digital transformasjon

UoH-sektor

Næringsliv

Portefølje Forskningssystemet

Portefølje Muliggjørende teknologier

Portefølje Energi og transport

Portefølje Banebrytende forskning

Portefølje Innovasjon

Programmer

Frittstående prosjekter

Storbritannia

Forskerprosjekt

Innovasjonsprosjekt

Digitalisering og bruk av IKT

Politikk- og forvaltningsområder

Bransjer og næringer

Energi

Miljøteknologi

Naturmangfold og miljø

Anvendt forskning

IKT forskningsområde

Klimarelevant forskning

Utviklingsarbeid

Internasjonalisering

Globale utfordringer

Grunnforskning

Miljøvennlig energi

Algebraic Theory of Information Quantities

Information theory is a discipline at the intersection of statistics, engineering and computer science. As the study of information quantities, such as compression or communication capacities, information content or measures of statistical dependency, it is one of the theoretical underpinnings of data science.

Computational problems in information theory are highly non-linear. The goal of this project is to transfer state-of-the-art methods of computational non-linear algebra to information theory, to study the inherent algebraic complexity of information-theoretical problems and to provide tools for solving them in practice. The algebraic point of view has proven to be fruitful in seemingly unrelated areas, as witnessed by a surge of recent work in algebraic statistics, in particular on likelihood geometry. However, maximizing the likelihood function is the same as minimizing relative entropy — a specific information quantity. Hence, this project also aims at generalizing the techniques developed in likelihood geometry.

One focus is on the practical computation of information quantities using numerical and differential algebraic geometry. Such quantities are defined via non-linear optimization problems and we aim to pinpoint the algebraic complexity of these problems in instances of general interest, such as common information measures. The final objective is finding fundamental laws and limits of data science imposed by non-linear inequalities constraining the entropic region. These inequalities provide, by duality, universal bounds for many of the optimization problems studied in information theory.

WHeelchair Activity Monitoring project (WHAM)

Wheelchair users (WCU) are three to five times more inactive than the general population, leading to a high risk of developing preventable secondary conditions. Although physical activity (PA) guidelines have been specifically developed for WCU, we lack systematic approaches to implement them in practice, as well as valid algorithms to monitor PA change in WCU through sensor-based solutions. Accordingly, the overall scientific aim of the WHeelchair Activity Monitoring project (WHAM) is to advance PA research and practice in WCU by developing valid digital markers for monitoring of and providing feedback on PA, paving the way for implementation and intervention. I will accomplish this through combining the complementary research fields of 1) behavioural intervention studies focusing on increasing PA in WCU (core competence University of British Columbia, UBC; outgoing phase) and 2) sensor-based PA monitoring (core competence Norwegian University of Science and Technology, NTNU; returning phase – consolidated by non-academic placement at Maastricht Instruments, which develop research prototypes including PA monitors). WHAM’s scientific aim is achieved through three objectives: 1) Measure the effect of a pragmatic exercise intervention trial on self-reported physical activity levels and cardio-respiratory fitness, 2) Develop valid algorithms for tracking PA markers in WCU, and 3) Co-create a framework that outlines how PA can be conjointly monitored and provided feedback on in WCU. This fellowship will enable me to become established as an internationally recognized researcher in the field of “physical activity and disability”, while bringing together ongoing national disjointed research activities in this field. Altogether, my overall ambition is to contribute to reducing inequalities in daily-life participation among WCU compared to the rest of the population.

EOSC Beyond: advancing innovation and collaboration for research

EOSC Beyond overall objective is to advance Open Science and innovation in research in the context of the European Open Science Cloud (EOSC) by providing new EOSC Core capabilities allowing scientific applications to find, compose and access multiple Open Science resources and offer them as integrated capabilities to researchers. To do so, EOSC Beyond supports a new concept of EOSC: a federated and integrated network of Nodes operated at different levels, national, regional, international and thematic, to serve the specific scientific missions of their stakeholders. Further specific objectives of the project are to accelerate ‘time to product’ of new scientific applications with software adapters, enable Open Science with machine composability and dynamic deployment of shared resources, support innovation in EOSC with a testing and integration environment, and align the EOSC Core architecture and specifications to integrate with European dataspaces. The project extends the state of the art of the EOSC Core and adopts a co-design methodology, including requirements elicitation, software development and validation in collaboration with different use cases from EOSC national and regional initiatives (e-Infra CZ, Czechia, NFDI, Germany, and NI4OS, South East Europe region), thematic research infrastructures from Social Sciences and Humanities (CESSDA), Life Sciences (CNB-CSIC and Instruct-ERIC), Environmental Science (ENES and LifeWatch), and Health and Food (METROFood-RI). EOSC Beyond builds on the capacities of prospective EOSC Nodes and partners with multi-annual experience in developing solutions for large-scale federated digital infrastructures and aligns with the technical architecture and requirements of data spaces from different business sectors. Ultimately, EOSC Beyond supports Open Science in modern, data-intensive, and multidisciplinary research, facilitating resource discovery, access, and reuse across scientific communities, organisations, and countries.


Harmonization of Advanced Materials Ecosystems serving strategic Innovation Markets to pave the way to a Digital Materials & Product Passport

The overarching objective of DigiPass is to enhance the digital maturity of the European communities that develop materials and intermediate products. The project will develop recommendations and clear routes toward digitalized circular business models. The overarching key result of DigiPass is to create a sustainable platform which includes support for Digital Materials & Product Passport and for collaborative innovation-by-design processes in a circular economy served by advanced materials. A business model for operating such a platform completes the overarching objective.
DigiPass is harmonizing and synergizing collected materials data sources and digital infrastructures. This will enable interoperability of data exchange, and standardization of advanced materials knowledge representation at all maturity levels. Accelerating the design, development, and production of advanced, safe, and sustainable chemicals and materials, as they are necessary for innovative products, calls for a collaborative approach involving different stakeholders to support durability, repair and overhaul, reuse, and recyclability of products. DigiPass project impacts all materials development communities over the whole circular value chain. 


Enhanced X(cross)-disciplinary Community-driven Imaging Technologies for Earth and Environmental material research

The EXCITE² Network is revolutionizing Earth and environmental material science research by consolidating a critical mass of 18 research facilities in 12 European and associated partner countries into a unified infrastructure and providing transnational access to advanced imaging technologies. This enables scientists to decipher complex processes within Earth materials at scales from nanometres to hundreds of centimetres, gaining unparalleled insight into the chemical and physical processes that govern, among others, environmental toxicity and its impact on human health, critical metal extraction for renewable energies, and the utilisation of Earth materials as long-term storage for climate-altering gases. The consortium fosters knowledge-sharing and talent attraction, bolstering the problem-solving capacity of the European Research Area. Moreover, EXCITE² is driving interdisciplinary collaboration and cross-fertilisation across academic institutions, scientific disciplines, and industry, propelling Europe towards a sustainable future. The initiative offers dedicated training programmes to a new generation of researchers within the European open science landscape, paired with access to world-class imaging facilities and expertise for problem-solving research and innovation. EXCITE² is introducing innovative service developments, such as artificial intelligence and leading-edge imaging experiments, to increase the user's problem-solving capacity. In addition, the consortium is implementing customised project valorisation strategies to maximize the impact of research outcomes in academia, society, and the economy. Citizen science projects allow the public to participate directly within the EXCITE² Network, contributing to solving EU challenges. As such, EXCITE² is paving the way towards a sustainable future, driving scientific excellence, and creating positive societal impact.

Engineering carbon quantum dots for zinc ion hybrid capacitors

Zinc ion hybrid capacitors (ZIHCs) integrate two energy storage mechanisms of battery-type electrode and capacitive electrode, bearing the advantages in energy density, power density, and safety. Until now, the cycle retention rate of ZIHC's capacity reported is still not more than 80%, restricting its development and commercial application severely. Focusing on the issue, the project aims at overcoming the limitation of traditional electrode materials and re-defining the designing concept to construct Carbon Quantum Dots (C-QDs) for ZIHCs. Advanced methods will be used to harvest carbon quantum dots from coal tar pitch (a sort of industrial solid castoff). The bulk sp2-carbon network, innumerable edge sites, profuse surface functional groups, and regular size ensures that as-synthesis C-QDs have high electrical conductivity, sufficient ion transport channels and rich adsorption sites. Moreover, the new concept of dynamic evolution is introduced for insight into the structure and interface failure of ZIHC's electrode. The interface-reconstruction mechanism and the attenuation model of capacity will be established for the first time. Further, the methods to strengthen structures of C-QD electrode will be developed based on a trade-off strategy. The ultimate goal is to improve the energy storage efficiency of ZIHCs by 20% and to increase their cycle performance more than 95% unprecedentedly. This research will develop long-running high performance ZIHCs, for contributing European energy storage industry. The ER will achieve abundant research experience and scientific skills from the project and the capability to launch her own research group in future.

IMProving User experience, Long-term sustainability and Services of EU-OPENSCREEN

EU-OPENSCREEN (EU-OS), the distributed European Research Infrastructure for Chemical Biology and early drug discovery, provides world-class services in high-throughput compound screening and medicinal chemistry to users from academia and industry. As one of the largest open-access initiative globally, EU-OPENSCREEN plays an increasingly important role in facilitating early stage drug discovery in Europe. It recently integrated fragment-based drug discovery and chemoproteomics as two new services, but the increasing complexity of modern drug discovery projects requires an even more comprehensive portfolio and integration of emerging technologies. To ensure that EU-OS continues to offer a comprehensive suite of relevant services and support a growing user community to conduct forefront research, IMPULSE will develop, validate and integrate new services in AI/ML, new chemical modalities, and innovative disease-relevant cell models combined with CRISPR-Cas9/RNAi-based genetic screening. IMPULSE will work towards the implementation of common operational and data standards to increase data reproducibility and FAIRness. Bespoke outreach and training activities will raise awareness among key user groups of its new services and advance operational excellence and reproducibility in pre-clinical drug discovery. New models for pre-competitive drug discovery with industry or charities, and growing ERIC member community will reinforce the sustainability and accessibility of EU-OPENSCREEN.

Aqua Research Infrastructure Services for the health and protection of our unique, oceans, seas and freshwater ecosystems

AQUARIUS will provide a highly comprehensive suite of integrated research infrastructures appropriate to addressing significant challenges for the long-term sustainability of our unique oceans, seas and freshwater ecosystems. For the first time, diverse research infrastructures will be combined to facilitate the work of researchers and key stakeholders focused on challenges and opportunities for both marine and freshwater systems. An impressive range of 57 research infrastructure services will be made available to include research vessels, mobile marine observation platforms, aircraft,  drones, satellite, sensors, fixed freshwater and marine observatories and test sites, experimental facilities, and sophisticated data infrastructures. AQUARIUS will support the development phase of the EU Mission to Restore our Ocean and waters by 2030, the Sustainable Blue Economy Partnership, the European Green Deal, and international climate initiatives. It will also be an essential component in achieving the European Digital Twin of the Ocean and the UN Decade for Ocean Sciences. The needs of researchers will be met through a robust and transparent system of transnational access funding Calls, facilitated by centralised user-friendly access portal. The Call programme will be informed through stakeholder engagement and brokerage events. Projects to be selected for Access must convincingly integrate multiple infrastructures and contribute to the core policy objectives of Mission Ocean, that is, to protect and restore marine and freshwater ecosystems and biodiversity; to prevent and eliminate pollution of our oceans, seas and waters; and to ensure a sustainable, carbon-neutral and circular blue economy. A thematic and geographic focus will be the hallmark of the proposed transnational Calls, aligning with the Lighthouse Regions, that is, the Baltic and the North Sea Basins, Black Sea, Atlantic/Arctic, and Mediterranean Sea along with their associated rivers.

EOSC-ENTRUST: A European Network of TRUSTed research environments

The mission of EOSC-ENTRUST is to create a European network of trusted research environments for sensitive data and to drive European interoperability by joint development of a common blueprint for federated data access and analysis. 
Countries and institutions have made significant investments in secure and trusted data environments to meet the need for research and policy-driven analysis of sensitive datasets such as people's health and socio-economic status, the habitats of vulnerable species and geo-spatial locations of protected sites. There are now a large number of such secure environments in operation - linked to national centres, individual organisations or specific research communities. This fragmented landscape poses challenges for both users and providers: researchers are faced with a large number of different systems and access procedures; providers need to manage federated access across multiple, potentially incompatible, technology and governance frameworks.
EOSC-ENTRUST brings together providers of operational Trusted Research Environments from 15 European countries with a shared goal to implement, validate and promote their capabilities through a common European framework using shared standards and common legal, operational and technical language. This blueprint for interoperability is anchored in the EOSC Interoperability Framework spanning the four dimensions of Legal, Organisational, Technical and Semantic interoperability. EOSC-ENTRUST has identified four driver projects covering Genomics, Clinical trials, Social science, and public-private partnerships to benchmark capabilities, inform blueprint design and demonstrate secure  data analysis using federated workflows.
Targeted outreach activities will expand this open network with further providers and develop policy papers and guidelines for the full range of stakeholders to create a long-term operational TRE framework within the European Open Science Cloud.


Providing research infrastructure services to support Next Generation EU

This four-year project will harness outputs from key social science RIs in Europe and other reliable sources to support the aims of the ‘Next Generation EU’ programme and to inform EU youth policy. Virtual access will be provided to the outputs through special installations (portals) developed during the project. The core of the proposal is a set of existing research ‘services’, which will be re-purposed and customised to focus directly on the five themes: Make it Green; Make it Digital; Make it Healthy; Make it Strong; and Make it Equal. 

For each area, the consortium will work collaboratively to produce seven sets of resources:

i)	A searchable inventory of relevant variables drawn from multiple cross-national datasets; 

ii)	Analytical summaries, suitable for policy makers and analysts new to the field, as well as interactive resources available through the project portal;

iii)	Harmonised and merged extracts from these datasets that reduce the burden on analysts and increase sample sizes;

iv)	New data collected through the ESS’s established CRONOS web panel infrastructure. Eleven countries will field five waves of the panel, each one dedicated to a NextGenEU theme. Free and rapid access to the data will be available through the CRONOS portal alongside analytical summaries;

v)	In-person and virtual deliberative forums with young people (aged 18-34) in four countries covering the five Next Generation EU’ themes;

vi)	An educational tool (E-NextGen) channelling our data for use in classrooms and by the general public;

vii)	Comprehensive training materials related to all project outputs.


Midlene fra FORNY20-FORNY2020 fordelt på år

Midlene fra FORNY20-FORNY2020 fordelt på fylker

Midlene fra FORNY20-FORNY2020 fordelt på fagområde

Smart building management: enhancing energy efficiency and indoor comfort through Digital Twin and AFDD technologies.

SecureIoTM-prosjektet har som mål å omdefinere cybersikkerhetsstandardene for nettverkskoblede enheter. Når alt fra smarte kjøleskap til sporingssystemer inngår i IoT-økosystemet, blir deres sikkerhet av største betydning. Sentralt i denne innsatsen står Tsetlin Machine, en banebrytende metode innen maskinlæring som forener avansert beregningskraft med energieffektivitet.

I motsetning til tradisjonelle maskinlæringsrammeverk, setter Tsetlin Machine seg frem ved å kombinere høy ytelse med energieffektivitet. Med stadig flere enheter, fra essensielt medisinsk utstyr til kaffemaskiner, som integreres i IoT, setter SecureIoTM fokus på å styrke deres evne til å motstå cyberangrep og dermed beskytte både dataintegritet og privatliv.

Dette innovative prosjektet er et samarbeid mellom Center of Artificial Intelligence Research ved Universitetet i Agder, Sørlandet Hospital Trust, og Newcastle University. Det bygger på fire nøkkelpilarer for innovasjon.

Først og fremst vektlegger Kontinuerlig Læring enheters evne til stadig å lære og tilpasse seg, samtidig som tidligere kunnskap bevares. Adaptiv Kompleksitet refererer til Tsetlin-maskinens evne til dynamisk å endre sin modellkompleksitet etter oppgavens krav, balanserende presisjon og energibruk. Cybermotstandsdyktighet, som tredje pilare, går videre enn bare å oppdage trusler; det gir systemet kapasiteten til å aktivt bekjempe cybertrusler i sanntid. Til slutt sikrer Sømløs Integrasjon at SecureIoTMs løsninger kan integreres smidig inn i eksisterende IoT-nettverk, forsterkende deres innebygde intelligens og sikkerhet.

Skulle SecureIoTM realisere sine ambisjoner, kan det omdefinere dagens standarder for trusseldeteksjon innen IoT, med metoder som er mer adaptive, robuste og energieffektive.

The Internet of Things (IoT) has major security concerns because of its limited ability to dedicate system resources to security mechanisms. Combined with billions of resource-constrained devices in healthcare, agriculture, and industry, this offers cybercrime an economic incentive. IoT is heterogeneous systems, making traditional security considerably more difficult to implement. Intrusion Detection Systems (IDS) are particularly difficult since conventional stand-alone anomaly detection cannot run at low energy, required for power-hungry Wi-Fi.  We need IDS for IoT, as attackers can gain device access when passive security fails. 

Tsetlin Machines (TMs), a new machine learning (ML) technique based on the Tsetlin Automaton, can meet the energy requirements of IoT. TM hardware (HW) has proven over ten orders of magnitude less energy and faster learning compared to NNs, outperforming vanilla and binary forms of CNNs and Feed-Forward on well-established benchmarks. TMs extract common patterns in data via frequent pattern mining and resource allocation instead of output error minimization, which is prone to overfitting. With its ultra-low-energy architecture, TM is key to creating state-of-the-art IDS, but it requires study in three crucial areas. 

1) Continual Learning for TM is crucial to online training as new sensor data or WIFI signals enter without catastrophic forgetting.  Machine learning, particularly TM, needs better techniques to prevent catastrophic forgetting. 

2) Automatic tuning of hyperparameters for multimodal and temporal data is extremely difficult since IoT and its heterogeneity make individual devices modality-specific. This means that each device requires specific hyperparameter tuning, which is impossible for millions of IoT devices.

3) Cyber-Resilient Training is crucial to training robust models where attackers can't modify data undetected. An attack can readily deceive training without robust models, making anomaly detection ineffective.

SecureIoTM: Ultra-low-energy IoT Intrusion Detection Systems using Logic-based Tsetlin Machines

Forskere samler stadig flere bevis på at maskinlæringsmodeller basert på korrelasjon er skjøre. Disse modellene skiller ikke mellom korrelasjon og årsakssammenheng. Som sådan gir de begrenset innsikt og kan ikke resonnere om virkningene av handlinger. De tradisjonelle feltene logikk, kausal resonnering og dyp læring har slitt med å lære storskala årsaksmodeller i flere tiår. Hver tilnærming står overfor betydelige hindringer. CaReLearner-prosjektet skal utvikle en enhetlig tilnærming som skal løse kjerneutfordringene til hvert paradigme. Prosjektet bygger på Tsetlinmaskinen og tar for seg den uløste læringsutfordringen innen logikk, skaleringsutfordringen til kausal resonnering og korrelasjonsavhengigheten til dyp læring. Vi vil dermed muliggjøre kausale og kontrafaktiske resonnementer i stor skala. Dette innebærer sammenkobling av bilder, tidsserier, tabelldata og tekst i tid og rom, for å håndtere utfordrende multimodale datasett fra kunnskapsintensive anvendelsesområder. En medisinsk datasett vil validere i hvilken grad de lærte modellene fanger opp årsaksmekanismer gjennom multimodale observasjoner vurdert av domeneeksperter.

Researchers are accumulating increasing evidence that machine learning models based on correlation are brittle. These models do not distinguish between correlation and causation, and, thus, they provide limited insight and cannot reason about the effects of actions for decision-making. The traditional fields of logical engineering, causal inference, and deep learning have struggled with learning causal models at scale for several decades, each approach facing substantial obstacles. The CaReLearner project seeks to adopt a unified approach to address the core challenges from each paradigm, dealing with the unresolved learning challenge in logical engineering, the scaling challenge of probabilistic causal models, and the correlation-reliance of deep learning. The unified approach merges: (1) recursive logical Horne clauses for modeling all possible functions that can be calculated; (2) causal Tsetlin machine learning for distilling causal mechanisms from data into causal Horn clauses; (3) large-scale probabilistic causal reasoning over sparse truth tables in Horn clause form. We will evaluate the ability to do causal and counterfactual reasoning at scale, which entails interconnecting images, time series, tabular data, and natural language text across time and space, on challenging multi-modal datasets from knowledge-intensive domains involving safety-critical decision-making. Compared to deep learning architectures, the performance target measures are: (1) surpassing or maintaining accuracy, (2) reducing energy by 100x, (3) reducing memory by 10x, and (4) reducing the overall computation cost by 10x. A medical case will validate the degree to which the learned models capture causal mechanisms through multi-modal observations, assessed by human domain experts. Overall, the most far-reaching impact of CaReLearner will be to synergize the above three fields, unifying the artificial intelligence community under a new research paradigm.

CaReLearner: Causal Reasoning with Logical Interpretable Learning

Tsetlinmaskiner er en ny form for maskinlæring som er basert på setningslogikk og Tsetlin-automater (endelige tilstandsautomater for maskinlæring). De lærer ved å trekke ut typiske mønstre fra data og beskriver disse med logiske uttrykk. Enkelheten i de logiske uttrykkene muliggjør svært lavt energiforbruk og rask parallell maskinvare. Til tross for denne enkelheten har Tsetlin-maskiner overgått nøyaktigheten til enkle dype nevrale nettverk på flere datasett. I dette prosjektet tar vi sikte på å løse tre sentrale utfordringer i Tsetlin-maskinforskningen: 1) Nåværende maskinvareprototyper er begrenset til småskala maskinlæringsproblemer. 2) Det er uklart hvordan man kan utføre forsterkningslæring (læring fra straff og belønning), som er et grunnleggende maskinlæringsparadigme. 3) Vi vet ikke hvordan man kan trene TMer fra umerkede data, noe som er viktig for å håndtere flaskehalsen manuell merking av data skaper. Ved å overvinne disse tre ufordringene tar vi sikte på å lage et nytt programvare- og maskinvareøkosystem som overgår dagens maskinlæring. Dette vil muliggjøre kraftige logikkbaserte maskinlæringsapplikasjoner for alt fra tingenes internett til skyen. 

For den første utfordringen har vi vist hvordan vi kan redusere energiforbruket ved å innføre deterministiske læringstrinn. Vi har videreutviklet en skalerbar og asynkron arkitektur som kan utnytte parallell maskinvare. Våre nyeste maskinvaredesign skaleres nå for å støtte Tsetlin-maskinkonvolusjon for bildeanalyse, og vi introduserer en fullt asynkron Tsetlin-auttomatarkitektur med integrert ikke-flyktig minne. Når vi skalerer opp kompleksiteten til datasettene for maskinlæring, har vi lagt merke til betydelig minnetrafikk i maskinvaredesignene, noe som påvirker effektiviteten og gjennomstrømningen vår. For å redusere minneflaskehalsene har vi designet flere maskinvarearkitekturer i minnet som er i stand til å behandle lagringsdata i modulære minneenheter. For implementeringer i minnet har vi brukt to typer minneteknologier: tradisjonelle statiske RAM og nye ikke-flyktige resistive minner. Vi har også fokusert på databooleanisering for å bygge konsise og nøyaktige mønstre. Vi har utviklet visualiseringsmetoder for å forstå og forklare virkningen av booleanisering og hyperparametervalg på effektiviteten av maskinlæring. Ved siden av disse har vi foreslått maskinvaredesign ved bruk av såkalte Petri-nett, inkludert latensanalyse.

For den andre utfordringen har vi utviklet det første Tsetlin-maskinrammeverket for off- og on-policy forsterkningslæring evaluert på grid-verdener. Videre, for brettspillet Hex har vi designet representasjoner og tolkningsmekanismer som tillater logikkbasert vinnerprediksjon. Vi har også løst problemet med online forsterkningslæring i en kontekst (det såkalte kontekstuelle bandittproblemet) med en Tsetlin-maskin, og utkonkurrerte andre populære sammenlignbare metoder. For komplekse oppgaver har vi observert at de logiske uttrykkene kan bli ganske lange. For å løse dette problemet har vi foreslått en ny variant av Tsetlinmaskinlæring der du kan begrense størrelsen på de logiske uttrykkene. Dette har tillat oss å halvere størrelsen på uttrykkene som ble brukt for vinnerprediksjon i Hex, uten tap i nøyaktighet.

For den tredje utfordringen har vi utviklet en Tsetlin-maskinarkitektur som støtter kartlegging mellom flerdimensjonale inn- og utdataverdier. Denne tilnærmingen har muliggjort den første Tsetlinmaskin-baserte autoencoderen,  samt selvlæring av språkmodeller. Vi har også laget den første ikke-ledede Tsetlin-maskinarkitekturen for å finne tolkbare klynger i data. For å adressere dynamiske miljøer har vi undersøkt en logisk tilnærming til modellering og deteksjon av datadrift. Til slutt har vi utviklet en metode for å avdekke relasjoner og struktur i data ved hjelp av et Tsetlinmaskin-produsert Bayesiansk nettverk, samt en Tsetlinmaskin-arkitektur for behandling av tidsserier. For alle disse tilnærmingene har vi utforsket ulike applikasjoner, inkludert sansing, planlegging, naturlig språkbehandling, bildeanalyse og beslutningsstøtte.

Gjennom dette prosjektet har vi satt i gang flere forskningssamarbeid knyttet til individuelle utfordringer. Disse inkluderer effektiv bruk av visualisering for å utforske ulike algoritmer og deres atferdsdynamikk, sammen med utvikling av ensemblemetoder for å løse nøyaktighetsproblemer for maskinlæringsproblemer i stor skala. Funnene og resultatene av prosjektet har vært grunnleggende for å utforske nye veier for Tsetlinmaskinen.

Tsetlin Machines (TMs) are a new machine learning (ML) approach founded on the Tsetlin Automaton. TMs use frequent pattern mining and resource allocation to extract common patterns in the data, rather than relying on minimizing output error, which is prone to overfitting. Unlike the intertwined nature of pattern representation in neural networks (NNs), TMs decompose problems into self-contained patterns, each represented as a conjunctive clause. The clause outputs, in turn, are combined into a classification decision through summation and thresholding, akin to a logistic regression function, however, with binary weights and a unit step output function. TM hardware (HW) has demonstrated up to three orders of magnitude reduced energy and faster learning, compared to NNs alike. Logic-driven fundamental blocks, organized in lean parallel processing units, are major contributors to this comparative advantage over NNs that are heavily arithmetic-based. The TM further outperforms vanilla CNNs and LSTMs accuracy-wise on well-established benchmarks. While the reported results on TMs are promising in terms of scalability, training time, accuracy, and energy, we here address three major obstacles. 1) Current FPGA and ASIC prototypes lack scalable memory elements, constraining them to small-scale ML problems. 2) Reinforcement learning (RL) is key to many ML problems, such as playing board games, however, it is unclear how to model RL in the TM framework. 3) State-of-the-art deep learning models support pre-training on unlabelled data, which significantly improves the accuracy of following supervised learning, dealing with shortage of labelled data. It is unclear how to pre-train TMs from unlabelled data. By overcoming these three obstacles we aim to architect a new TM HW/SW ecosystem that outperforms state-of-the-art ML in terms of energy efficiency and scalability, parametrised by accuracy. This will enable powerful logic-based ML applications at the edge and in the cloud.

Logic-based Artificial Intelligence Everywhere: Tsetlin Machines in Hardware

En utfordring for vannkraftprodusenter er å finne ut hvordan man skal disponere tilgjengelige vannmengder i kraftmagasinene. Skal man løse dette problemet må mange faktorer vurderes. Eksempler på dette kan være tilsig til kraftmagasiner, snømengder i fjellet, forventet prisutvikling på kraft, med mere.

I dette prosjektet har vi undersøkt hvordan kunstig intelligens kan anvendes til å finne ut hvordan et gitt vannkraftsystem kan utnyttes mest mulig optimalt. Med dette menes at for et gitt vassdrag og kraftstasjoner, så ønsker man til enhver tid å kunne beregne hva som er optimal bruk av tilgjengelig vannmengde i magasinene. 

Prisen på elektrisk strøm har de seneste årene gått kraftig ned. En konsekvens av dette er at de norske vannkraftprodusentene opplever at inntektene er blitt redusert og at det er presserende behov for å minske kostnader, samt effektivisere driften.

Prosjektet har vist at det er mulig å benytte dyp læring til å forstå hvordan vannmagasinene skal disponeres optimalt. Hovedresultatene i prosjektet demonstrerte hvordan ulike søkealgoritmer kan kombineres med syntetiske data og brukes til å trene nevrale nett. De trente nevrale nettene kan brukes til å foreslå produksjonsplaner eller vannverdier basert på pris- og tilsigsinformasjon. Algoritmene har blitt testet ut i skarp drift for Kvinesdal kraftstasjon og Agder Energi benytter nå metodene i sitt daglige arbeide. Prosjektet viser et område hvor kunstig intelligens har et stort potensiale og i fremtiden vil prosjektresultatene kunne danne utgangspunkt for nye verktøy for resten av vannkraftbransjen.

The project is motivated by the fact that the Nordic electricity market for the last two years has experienced a huge fall in energy prices. The reasons for this fall are many, but the undisputable consequence for the Norwegian hydropower sector is that income is going down and there is an urgent need to lower costs and run the operations more efficiently. To this end, there is a need for more practical applicable simulation software. However, traditional solvers based on linear and dynamic programming techniques have significant shortcomings in operational use.

The underlying idea in this research project is to apply and adapt recent breakthroughs in so-called Deep Reinforcement Learning (DRL) to the hydropower scheduling problem. To our knowledge, this has never been done before. We will develop DRL based models, algorithms and an accompanying hydraulic-economic model that can utilize inexpensive and massive parallel computing platforms offered by Graphical Processing Units (GPUs). The new software will be tested in an operational setting at Agder Energi and compared to traditional optimization techniques based on linear- and dynamic programming methods (LP, DP).

Given the success in this project the value creation will be more efficient utilization of the water resources available at Agder Energi. Today this energy resource is about 4 percent of the Norwegian hydropower production and represents a huge part of Agder Energi's income. It is clear that even small incremental improvements in how water resources are used (1-2 %), will create huge increase in income for the Norwegian hydropower sector. Furthermore, societal costs related to flooding may be reduced with improved techniques for hydropower optimization.

FORNY20-FORNY2020

Smart building management: enhancing energy efficiency and indoor comfort through Digital Twin and AFDD technologies.

Awarded: NOK 0.50 mill.

Funding scheme:

FORNY20-FORNY2020

Funding Sources

Thematic Areas and Topics