Frank Eliassen

简介

We present PM2Lat, a fast and generalized framework for accurately predicting the latency of deep neural network models on GPUs, with special focus on NVIDIA. Unlike prior methods that rely on deep learning models or handcrafted heuristics, PM2Lat leverages the Single-Instruction-Multiple-Thread architecture of GPUs to model execution time of DNN models. First, we conduct fine-grained GPU operation modeling by analyzing computational behavior and memory access patterns. Through this analysis, we observe that different GPU kernels can exhibit significant performance disparities, even when serving the same purpose. Hence, the core idea of PM2Lat is to differentiate kernels based on their configurations and analyze them accordingly. This kernel-aware modeling enables PM2Lat to achieve consist

代表成果

• Scientific articles and book chapters
• Ahmed, Awadelrahman Mohamedelsadig Ali & Eliassen, Frank (2026). Causal inference framework for smart grids with application to demand response. Applied Energy. ISSN 0306-2619. 424. doi: 10.1016/j.apenergy.2026.128495. Full text in Research Archive
• Le, Thanh; La, Hoang Loc; Taherkordi, Amirhosein; Eliassen, Frank; Ha, Hoai Phuong & Guan, Peiyuan (2026). PM2Lat: Highly Accurate and Generalized Prediction of DNN Execution Latency on GPUs. IEEE/ACM International Symposium on Cluster, Cloud, and Grid Computing. ISSN 2376-4414. p. 604–614. doi: 10.1109/ccgrid68966.2026.00070. Full text in Research Archive Show summary We present PM2Lat, a fast and generalized framework for accurately predicting the latency of deep neural network models on GPUs, with special focus on NVIDIA. Unlike prior methods that rely on deep learning models or handcrafted heuristics, PM2Lat leverages the Single-Instruction-Multiple-Thread architecture of GPUs to model execution time of DNN models. First, we conduct fine-grained GPU operation modeling by analyzing computational behavior and memory access patterns. Through this analysis, we observe that different GPU kernels can exhibit significant performance disparities, even when serving the same purpose. Hence, the core idea of PM2Lat is to differentiate kernels based on their configurations and analyze them accordingly. This kernel-aware modeling enables PM2Lat to achieve consistently low prediction error across diverse data types and hardware platforms. In addition, PM2Lat generalizes beyond standard matrix multiplication to support complex GPU kernels such as Triton, Flash Attention, and Cutlass Attention. Experimental results show that PM2Lat consistently achieves error rates below 10% across different data types and hardware platforms on Transformer models, outperforming the state-ofthe-art NeuSight by 10-20% for FP32 and by at least 50% for BF16. When applying to diverse kernels, the error rate is maintained at 3-8%.
• Foroughi, Mehdi; Bagherpour, Matin; Eliassen, Frank & Owe, Olaf (2026). PSDO: Privacy-Preserving Distributed Optimization for Autonomous Prosumer Energy Management. IEEE Transactions on Dependable and Secure Computing. ISSN 1545-5971. 23(3), p. 6874–6889. doi: 10.1109/tdsc.2026.3669022. Full text in Research Archive Show summary In the evolving energy landscape where prosumers play an increasingly important role, establishing a secure data exchange architecture is essential for building a resilient and efficient energy infrastructure. Current privacy-preserving systems suffer from inadequate adversarial models, dependence on centralized components, and inability to adapt to evolving threats. This paper introduces the Privacy-Sensitive Distributed Optimization (PSDO) framework, a decentralized management scheme designed to prioritize privacy safeguards in prosumer-driven systems. To achieve this goal, the PSDO framework combines decentralized optimization techniques with differential privacy. This integration serves a dual purpose: preserving prosumers’ control over their energy management and ensuring privacy of their sensitive information. By leveraging decentralized optimization, the PSDO framework enables prosumers to maximize the benefits derived from decentralized systems, promoting improved autonomy in energy management. Simultaneously, prosumers’ sensitive data is protected through the implementation of differential privacy measures. Through implementation on the IEEE 33-bus radial distribution system, the PSDO algorithm demonstrated its capability to converge to the optimal solution while rigorously upholding differential privacy. PSDO advances beyond the chosen baseline (DP-ADMM) by delivering superior privacy protection while incurring minimal utility loss. Moreover, the framework successfully accommodates diverse privacy preferences while maintaining system-wide efficiency, establishing its effectiveness for heterogeneous prosumers.
• Sharma, Jivitesh; Vallejo, Islen; Ødegård, Rune Åvar; Le, Truong Thanh; Taherkordi, Amirhosein & Eliassen, Frank (2025). Physics-Informed Deep Learning for Wind Downscaling over Oslo. Proceedings - International Conference on Tools with Artificial Intelligence (ICTAI). ISSN 1082-3409. 37, p. 270–276. doi: 10.1109/ICTAI66417.2025.00042. Full text in Research Archive Show summary Running a numerical weather model such as WRF at kilometre or sub-kilometre grid spacing over a regional domain is computationally expensive. We present physics-informed deeplearning models that ingest a single 9km WRF wind field and simultaneously predict two finer-scale wind fields at 3 km and 1 km resolution via dual decoder heads. Four representative architectures are benchmarked-Deep Residual U-Net (DeepRU), DEVINE, a bespoke 3-D Transformer, and a Fourier Neural Operator (FNO)-each trained with divergence-free, vorticity, and Navier-Stokes residual constraints plus Charbonnier and gradient perceptual losses. We train and validate our models on the city of Oslo for the year 2018. DeepRU achieves R2=0.94 (RMSE =0.050) at 3km and R2=0.89(RMSE=0.065) at 1 km. DEVINE, Transformer 3-D, and FNO yield 3 km scores of 0.91−0.93, with 1km scores lower by 0.02−0.08, illustrating the increased difficulty of finer-scale reconstruction. Physicsinformed losses improve all models compared to MSE-only baselines, and the residual architecture (DeepRU) remains most effective for this dual-scale task.
• Zhang, Min; Eliassen, Frank; Taherkordi, Amirhosein; Jacobsen, Hans-Arno; Li, Yushuai & Zhang, Yan (2025). Self-Determination Theory and Deep Reinforcement Learning for Personalized Energy Trading in Smart Grid. IEEE Transactions on Systems, Man & Cybernetics. Systems. ISSN 2168-2216. 55(6), p. 4216–4229. doi: 10.1109/TSMC.2025.3551667. Full text in Research Archive
• Foroughi, Mehdi; Bagherpour, Matin; Eliassen, Frank & Poudineh, Rahmatallah (2025). Autonomy as empowerment: A taxonomic framework for analyzing energy autonomy in local flexibility markets. Applied Energy. ISSN 0306-2619. 389. doi: 10.1016/j.apenergy.2025.125777. Full text in Research Archive
• Le, Truong Thanh; Taherkordi, Amirhosein; Eliassen, Frank & Guan, Peiyuan (2024). Optimal Distribution of ML Models Over Edge for Applications with High Input Frequency. IEEE International Conference on Cloud Computing Technology and Science (CloudCom). ISSN 2330-2194. p. 143–150. doi: 10.1109/cloudcom62794.2024.00033. Full text in Research Archive Show summary The rise of complex and sizeable Machine Learning (ML) models challenges traditional cloud computing models with respect to the high volume of incoming data which results in increased bandwidth usage and network congestion, as well as delays in inference. Such ML models are being rapidly developed thanks to advances in computing platforms and the real-time computing demands of ML-driven applications such as autonomous vehicles and video processing. To mitigate these challenges, ML model distribution and inference offloading to computing devices close to data sources have been explored, especially through partitioning the models across the IoT-Edge-Cloud continuum. Existing efforts in this area have not successfully mastered the fully automatic and efficient determination of optimal partition points. Additionally, they have not effectively integrated Early-Exit layers that allow for early termination of model inference at earlier stages when feasible. In this paper, we introduce a novel partitioning algorithm designed to distribute ML models across edge devices with the goal of reducing response time when facing high-rate input data streams. Our proposed approach leverages the principles of Dynamic Programming to determine optimal partition points and establish appropriate exit thresholds for Early-Exit layers. Our evaluation results reveal that, in the context of continuous, high-rate input data, our method consistently lowers the maximum round-trip time for processing inference requests compared to state-of-the-art methods such as NeuroSurgeon and Genetics.