Institution profile

SINTEF

Academic institutioneurope · no
Official website
Research library66linked papers
Opportunities0open roles
Selected work

Representative Papers

An Agentic Operationalization of DISARM for FIMI Investigation on Social Media

Jan 21, 2026

This work proposes the first DISARM-oriented multi-agent AI system to address the challenges of automated detection and standardized categorization in large-scale application of the DISARM framework against foreign information manipulation and interference (FIMI) on social media. The system integrates natural language processing with knowledge mapping techniques, enabling collaborative agents to automatically identify manipulative content and transparently map it to the DISARM standard taxonomy in an interpretable manner. Experimental evaluation on two real-world annotated datasets demonstrates that the proposed approach significantly enhances FIMI analysis efficiency, strengthens situational awareness, and facilitates cross-organizational data interoperability.

1 citations1 influentialRead paper

Sashimi-Bot: Autonomous Tri-manual Advanced Manipulation and Cutting of Deformable Objects

Nov 14, 2025

This work addresses the challenge of autonomous manipulation and high-precision cutting of natural, deformable 3D objects—exemplified by salmon fillets. Key difficulties include substantial inter-object geometric and dimensional variability, unknown viscoelastic material properties, and slippery, compliant surfaces prone to slippage. To overcome these, we propose a coordinated three-arm robotic framework integrating vision–tactile perception with deep reinforcement learning for real-time, adaptive in-hand tool manipulation and dynamic in-hand cutting. To our knowledge, this is the first system achieving stable multi-point grasping, deformation compensation, pose adjustment, and thin-slice cutting of soft-bodied targets via tri-arm coordination. Experiments demonstrate robust handling of highly heterogeneous salmon fillets, achieving sub-millimeter slicing accuracy; pick-up success rate and slice quality approach human-level performance. The framework establishes a scalable, generalizable paradigm for automated processing of deformable food products.

1 citations1 influentialRead paper

Virtual Temperature Sensors in Power Transformers Using Neural Ordinary Differential Equations

Aug 13, 2026

Existing thermal modeling approaches for power transformers struggle to simultaneously achieve real-time performance, generalization capability, and physical consistency. This work proposes a virtual temperature sensor based on Neural Ordinary Differential Equations (Neural ODEs), which, for the first time, embeds a simplified heat conduction equation into the Neural ODE framework to enable physics-informed continuous-time dynamic modeling. The resulting method offers interpretability, a standardized architecture, and cross-device generalization, making it suitable for heterogeneous transformers. Validation on fifteen transformers in Norway—spanning diverse designs and cooling strategies—demonstrates that the model robustly, accurately, and physically consistently predicts thermal dynamics.

0 citationsRead paper
Recent publications

Latest Papers

Virtual Temperature Sensors in Power Transformers Using Neural Ordinary Differential Equations

Aug 13, 2026

Existing thermal modeling approaches for power transformers struggle to simultaneously achieve real-time performance, generalization capability, and physical consistency. This work proposes a virtual temperature sensor based on Neural Ordinary Differential Equations (Neural ODEs), which, for the first time, embeds a simplified heat conduction equation into the Neural ODE framework to enable physics-informed continuous-time dynamic modeling. The resulting method offers interpretability, a standardized architecture, and cross-device generalization, making it suitable for heterogeneous transformers. Validation on fifteen transformers in Norway—spanning diverse designs and cooling strategies—demonstrates that the model robustly, accurately, and physically consistently predicts thermal dynamics.

0 citationsRead paper

IO Factory: Simulating AI-Enabled Influence Campaigns at Scale

Aug 11, 2026

This study addresses the challenge of detecting AI-driven coordinated influence operations, which are difficult to identify through isolated data points. To this end, the authors propose a traceable and reproducible closed-loop simulation framework that models influence campaigns as an end-to-end process encompassing roles, actions, exposure, evaluation, and adaptation. Built upon a multi-agent system, the framework integrates action planning with feedback-driven adaptation mechanisms and incorporates belief variables alongside structured assessment methodologies. The system was successfully deployed at scale, simulating over 100,000 agents and generating auditable exposure pathways and belief evolution trajectories. This work represents the first large-scale simulation capable of capturing the full lifecycle of AI-coordinated influence operations, demonstrating both the scalability of the architecture and its analytical efficacy.

0 citationsRead paper

I wanted it to feel more personal: Customization of social AI as AI individualism in practice

Jul 20, 2026

This study addresses a critical gap in understanding how users actively customize social AI to reflect personal preferences. Introducing the novel concept of “AI individualism,” the research employs reflexive thematic analysis on open-ended survey responses from 169 participants, identifying seven key motivations for customization. Findings reveal that users perceive social AI as a personalized social resource, and their customization practices extend beyond functional adjustments to constitute a socio-technical practice through which they construct self-extension and a sense of pseudo-autonomy. This process enhances perceived support, autonomy, and intimacy, while reinforcing feelings of individualism, freedom, and control. However, it may also foster an illusion of system controllability, potentially obscuring the underlying algorithmic constraints and power dynamics inherent in AI systems.

0 citationsRead paper