Tech & AI

Obligatorisk entreprenörskapundervisning integrerad i tekniska projektöverskeningar ökar sannolikheten att kompetensen förankras på programnivå — inte bara genomförås som tillfälliga kurser. En femtonårs fallstudie från Linköpings universitet visar att när företagarskap är inbäddad i CDIO-baserade projektkurser och genomförts av över 3 000 studenter, utvecklas hållbar institutionalisering utan att tränga ut teknisk undervisning. Framgångsfaktorerna var obligatorisk struktur, stabila lärandemål med flexibel genomförande, och justering mot projektarbetets rytm. Forskaren Magnus Klofsten och sitt lag använde kvalitativ casestudier kombinerat med autoetnografi och analyserade studentomdömen från 2009 till 2025. För ledare inom teknikutbildning och deeptech-investerare är slutsatsen klar: entreprenörskapsutveckling kräver återkommande exponering genom ordinarie kurssekvenser, inte punktinsatser. Detta skapar både kompetensutveckling och programkoherens när nya talanger fostrars.

Engineers have cracked a stubborn problem in heavy-duty diesel engines: higher compression ratios boost fuel efficiency but increase heat loss to cylinder walls, negating gains. A simple offset orifice nozzle redirects flame distribution to cut cooling losses while maintaining efficiency improvements—a finding that could reduce fuel consumption and emissions across trucking and industrial equipment fleets.EN

Researchers have developed a novel mathematical framework that simplifies how physicists predict outcomes of particle collisions—a capability central to validating physics theories and designing particle accelerators. The breakthrough applies advanced algebraic structures to complex quantum calculations, potentially accelerating computational speed for experiments costing billions of dollars.EN

Researchers have created a computational model that simulates fermentation conditions for multiple food and beverage products simultaneously, predicting optimal outcomes and failure modes. The tool could accelerate product development and quality control for breweries, bakeries, and fermentation-based biotech firms while reducing costly trial-and-error experimentation.EN

Researchers have developed a faster method for solving complex planning problems in artificial intelligence by distributing computation across multiple processor cores. The breakthrough could accelerate everything from logistics optimization to robotics scheduling, with performance gains that multiply as more computing power is added.EN

Researchers have developed a technique called Gradient-Momentum Coupling that helps machine learning systems distinguish useful learning signals from random noise—a persistent problem that undermines AI training. The approach could improve how autonomous systems learn from data, making AI training faster and more reliable in real-world environments where noisy observations are common.EN

A new study reveals how fact-checkers in Cameroon, Kenya, Nigeria, and South Africa perceive their roles and obstacles. The findings matter to media companies, advertisers, and policymakers seeking to combat disinformation in high-growth markets where misinformation spreads faster than corrections.EN

Researchers have decoded how nature's slimy protective layers—found on fish and eels—absorb energy and reduce wear on moving surfaces. The discovery could enable tougher, self-healing coatings for machinery, robotics, and industrial equipment that perform better under stress without frequent replacement.EN

A new paper reveals that software engineering studies routinely reach wrong conclusions because they ignore important factors that shape outcomes. Researchers now have practical methods to detect and correct for these blind spots before publishing findings that could mislead engineering teams and technology investments.EN

A new study finds that immersive mixed reality experiences—where people inhabit virtual bodies—can reduce social barriers and foster genuine connection across different groups. The finding has implications for remote work, education, and customer engagement, suggesting companies may tap VR and AR to build more cohesive teams and communities.EN

Researchers have developed a more accurate way to measure silicon carbide chip layers—a critical material for electric vehicles and power electronics. The advance cuts measurement errors in half, potentially reducing costly defects in high-stakes manufacturing environments where precision directly affects device performance and profitability.EN

Researchers have mapped a practical path forward for continuous health monitoring devices that stick to skin or implant in the body, combining flexible materials with machine learning to catch health problems early. The breakthrough addresses long-standing barriers—battery life, data privacy, and clinical accuracy—that have kept these sensors out of hospitals and consumer markets.EN

Researchers have developed a new mathematical framework that transforms irregular, fractal-like spaces into simpler shapes without losing their essential energy characteristics. The breakthrough could accelerate computational modeling in machine learning, materials science, and data analysis by making complex geometries easier to process while maintaining their mathematical integrity.EN

Researchers have developed a method that can explain why tree-based machine learning models make specific predictions — and do so dramatically faster than existing approaches. For companies deploying AI in high-stakes decisions like credit approval or medical diagnosis, this breakthrough could reduce the computational burden of proving their systems' logic to regulators and customers.EN

Researchers have developed a machine-learning system that not only predicts when corroded pipelines will fail—but explains why in plain mathematical terms. The breakthrough matters to gas utilities and regulators because it combines accuracy with transparency, making it viable for real safety decisions where operators need to trust the technology.EN

Researchers have developed a mathematical framework that predicts how quantum states become more complex when conserved charges—like energy or particle number—are present. The work could improve designs for quantum computers and simulations by revealing fundamental speed limits on how fast quantum information can spread.EN

Medicinska databolag och forskningsmiljöer kan nu bryta dataksilos genom syntetisk datasyntes. Ramverket CLEO kombinerar Gaussiska blandningsmodeller med Q-learning för att generera patientdata som håller sekretess medan det bevarar analyskvalitet. Metoden testades på multincentrisk data från intrakraniella aneurysm-patienter. CLEO nådde ett kombinerat poäng på 0,9232 och överträffade konkurrenter som TVAE och CTGAN. Modeller tränade på den syntetiska datan uppnådde genomsnittlig AUC på 0,7376 i klinisk prognos, vilket bekräftar att beslutsdata bevarades. Närliggande-granne-granskning visade låg återidentifieringsrisk (0,4940). Initiativ från Taiyuan, Karolinska Institutet och Swansea Universitetet adresserar EU och svenska GDPR-krav. För produktchefer inom healthtech är detta relevant för att bygga kollaborativa ML-modeller utan att lagra patientdata centralt.

**Chipletarkitektur kräver exakt fysisk modellering för tillförlitlig simulering** Uppsala University presenterar DICE, ett verktyg för detaljerad end-to-end PHY-simulering av chipletbaserade system. Chiplets — modulära processorkärnor sammankopplade på samma substrate — växer som designparadigm för att hantera tillverkningskomplexitet och kostnad, men simuleringsmodeller har släpat efter. Befintliga verktyg förenklar gränssnittskommunikationen mellan chiplets, vilket leder till fel vid layout och timing. DICE modellerar signal- och strömöverföring mellan chiplets i detalj, från logik till fysisk implementering. Denna tillnärmning avslöjar problem som tvetydig latens och elektriska störningar tidigt — före tillverkning. För chipletleverantörer och SoC-designare är detta kritiskt: felaktig simulering kostar månader i redesign. Uppsala Universities arbete adresserar en faktisk flaskhals i chipletekosystemet när Intel, AMD och asiatiska foundries skalrar produktionen.

Regenererad cellulosa kan förbättra energiomvandlingen i tribo­elektriska nanogeneratorer (TENG) genom tunna metalloxidskikt applicerade via atomär lagerpåväxling (ALD). Forskare vid Mid Sweden University och RISE modifierade cellulosaöverflöet med aluminiumoxid för att stabilisera den elektriska prestanda. Metoden minskar ytans känslighet för variationer i gränsytegenskaper som påverkar utgångsspänningen. Resultaten visar att ALD-behandlingen ökar pålitligheten och minskar driften över tid — kritiska faktorer för kommersialisering av fibbaserade energiskörd. Arbetet växer fram ur samarbete mellan Sverige och Finland, finansierat av Vinnova och svenska forskningsråd. För tillverkare av mjuka elektronik och bärbar energiteknik är detta relevant: stabilare materialegenskaper förkortar utvecklingstiden från laboratorie till produktion och minskar variationen mellan batcher, vilket direkt påverkar skalningsbar tillverkning av hållbara kraftkällor.

A new comparative study of five Euro 6 vehicle types reveals that even advanced low-emission cars release toxic particles that damage lung cells. The findings suggest regulators and manufacturers may need to tighten standards further—particularly for diesel vehicles during filter cleaning cycles—to protect public health and avoid future costly recalls.EN

Researchers found widespread skepticism among ship crews about meeting international evacuation standards, citing inadequate training and psychological unpreparedness. The findings expose gaps between regulatory requirements and operational reality—a critical issue for cruise operators facing liability risks and maritime authorities responsible for passenger safety.EN

Researchers have built the first practical imaging system that works inside a dilution refrigerator, enabling scientists to visualize and test quantum devices at near-absolute-zero temperatures. The breakthrough removes a major engineering bottleneck in scaling up quantum networks—a technology race that could reshape computing and communications infrastructure over the next decade.EN

Researchers have developed improved methods for the generalized double pouring problem, a theoretical challenge with practical applications in resource allocation and logistics optimization. The work provides tighter computational bounds and new algorithms that could accelerate solutions for real-world distribution and scheduling problems across industries.EN

A new study of Sweden's smart-grid rollout reveals that treating digital infrastructure as immaterial masks a hard reality: equipment gets replaced long before it needs to be, driving unnecessary mineral demand and excluding poorer communities. Researchers say governance frameworks must now synchronize lifetimes—regulatory, financial, and technical—to prevent premature obsolescence.EN

Researchers have developed equations that predict how polyatomic gases behave when molecules move and vibrate at different temperatures simultaneously. The work could improve design of high-speed aircraft engines, industrial plasma reactors, and other systems operating under extreme conditions where standard gas models fail.EN