Mulugeta Weldezgina Asres

• Balapuwaduge, Indika A.M. & Asres, Mulugeta Weldezgina (2026). Prompt Compliance and Detectability in Text-to-Video Abnormal Incident Simulation for Safety-Critical Computer Vision. I Misra, Sanjay; Kjølerbakken, Kai Morgan & Braarud, Per Øivind (Red.), International Conference on Human Centric Artificial Intelligence (ICHCAI): First International Conference, ICHCAI 2026. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9798331551186. doi: 10.1109/ICHCAI70183.2026.11607542. Fulltekst i vitenarkiv Vis sammendrag

  Open-source text-to-video (T2V) synthesis has been employed for simulating abnormal-incident scenarios for safety-critical applications, where real-world data are often difficult to collect and share because of privacy, operational, and ethical constraints. Before such synthetic videos can be considered for downstream model development, their semantic fidelity and incident recognizability should be assessed in a transparent and reproducible way. In this work, we evaluate open-source T2V models for abnormal incident video generation using two complementary upstream metrics. First, we propose a prompt compliance ratio that measures the proportion of frames that semantically align with the intended prompt over time. Second, we employ a zero-shot incident classification approach to estimate whether synthesized videos preserve recognizable incident categories without task-specific training. We also examine the impact of key video generation parameters, such as inference steps and subjective video quality using controlled prompts. Our method provides indicators of whether synthetic incident videos are sufficiently reliable for safety and security computer vision research workflows. The results indicate that the proposed metrics are effective for assessing both prompt compliance and zero-shot incident recognition in the generated videos. In addition, the study shows that parameter tuning plays an important role in improving separability and overall video quality.

• Asres, Mulugeta Weldezgina; Omlin, Christian Walter Peter & Collaboration, The CMS-HCAL (2026). Scalable temporal anomaly causality discovery in large systems: achieving computational efficiency with binary anomaly flag data. European Physical Journal C. ISSN 1434-6044. 86. doi: 10.1140/epjc/s10052-026-15611-5. Fulltekst i vitenarkiv Vis sammendrag

  Extracting anomaly causality facilitates diagnostics once monitoring systems detect system faults. Identifying anomaly causes in large systems involves investigating a broader set of monitoring variables across multiple subsystems. However, learning graphical causal models (GCMs) comes with a significant computational burden that restrains the applicability of most existing methods in real-time and large-scale deployments. In addition, modern monitoring applications for large systems often generate large amounts of binary alarm flags, and the distinct characteristics of binary anomaly data—the meaning of state transition and data sparsity—challenge existing causality learning mechanisms. This study proposes an anomaly causal discovery approach (AnomalyCD), addressing the accuracy and computational challenges of generating GCMs from temporal binary flag datasets. The AnomalyCD presents several strategies, such as anomaly data-aware causality testing, sparse data and prior link compression, and edge pruning adjustment approaches. We validate the performance of the approach on two datasets: monitoring sensor data from the readout-box system of the Compact Muon Solenoid experiment at CERN, and a public dataset from an information technology monitoring system. The results on temporal GCMs demonstrate a considerable reduction of computation overhead and a moderate enhancement of accuracy on the binary anomaly datasets. Source code: https://github.com/muleina/AnomalyCD

• Asres, Mulugeta Weldezgina; Lei, Jiao & Omlin, Christian Walter Peter (2025). Low-Latency Video Anonymization for Crowd Anomaly Detection: Privacy Versus Performance. IEEE Transactions on Information Forensics and Security. ISSN 1556-6013. 21, s. 1–16. doi: 10.1109/TIFS.2025.3630347. Fulltekst i vitenarkiv Vis sammendrag

  Recent advancements in artificial intelligence hold ample potential for monitoring applications using surveillance cameras. However, concerns about privacy and model bias have made it challenging to utilize them in public. Although de-identification approaches have been proposed in the literature, aiming to achieve a certain level of anonymization (AN), most of them employ deep learning models that are computationally demanding for real-time edge deployment. This study revisits conventional AN solutions for privacy protection and real-time video anomaly detection (VAD) applications. We propose a lightweight adaptive AN for VAD (LA3D) that employs dynamic adjustment to enhance full-body privacy protection. We have evaluated privacy protection and VAD utility retention efficacy using several publicly available datasets to examine the strengths and weaknesses of different AN methods and highlight the promising leverage of our approach. Our experiment demonstrates that the LA3D enables substantial improvement in privacy AN without severely degrading VAD efficacy, outperforming conventional and deep learning approaches.

• Asres, Mulugeta Weldezgina; Omlin, Christian Walter Peter; Wang, Long; Yu, David; Parygin, Pavel & Dittmann, Jay (2025). Data Quality Monitoring for the Hadron Calorimeters Using Transfer Learning for Anomaly Detection. Sensors. 25(11). doi: 10.3390/s25113475. Fulltekst i vitenarkiv Vis sammendrag

  The proliferation of sensors brings an immense volume of spatio-temporal (ST) data in many domains, including monitoring, diagnostics, and prognostics applications. Data curation is a time-consuming process for a large volume of data, making it challenging and expensive to deploy data analytics platforms in new environments. Transfer learning (TL) mechanisms promise to mitigate data sparsity and model complexity by utilizing pre-trained models for a new task. Despite the triumph of TL in fields like computer vision and natural language processing, efforts on complex ST models for anomaly detection (AD) applications are limited. In this study, we present the potential of TL within the context of high-dimensional ST AD with a hybrid autoencoder architecture, incorporating convolutional, graph, and recurrent neural networks. Motivated by the need for improved model accuracy and robustness, particularly in scenarios with limited training data on systems with thousands of sensors, this research investigates the transferability of models trained on different sections of the Hadron Calorimeter of the Compact Muon Solenoid experiment at CERN. The key contributions of the study include exploring TL’s potential and limitations within the context of encoder and decoder networks, revealing insights into model initialization and training configurations that enhance performance while substantially reducing trainable parameters and mitigating data contamination effects.

• Asres, Mulugeta Weldezgina; Omlin, Christian Walter Peter; Dittmann, Jay; Parygin, Pavel; Hiltbrand, Joshua & Cooper, Seth I. [Vis alle 8 forfattere av denne artikkelen] (2024). Lightweight Multi-System Multivariate Interconnection and Divergence Discovery. IEEE International Conference on System of Systems Engineering (SoSE). ISSN 2835-3307. 19(1), s. 36–43. doi: 10.1109/SOSE62659.2024.10620930. Fulltekst i vitenarkiv Vis sammendrag

  Identifying outlier behavior among sensors and subsystems is essential for discovering faults and facilitating diagnostics in large systems. At the same time, exploring large systems with numerous multivariate data sets is challenging. This study presents a lightweight interconnection and divergence discovery mechanism (LIDD) to identify abnormal behavior in multi-system environments. The approach employs a multivariate analysis technique that first estimates the similarity heatmaps among the sensors for each system and then applies information retrieval algorithms to provide relevant multi-level interconnection and discrepancy details. Our experiment on the readout systems of the Hadron Calorimeter of the Compact Muon Solenoid (CMS) experiment at CERN demonstrates the effectiveness of the proposed method. Our approach clusters readout systems and their sensors consistent with the expected calorimeter interconnection configurations, while capturing unusual behavior in divergent clusters and estimating their root causes.

• Asres, Mulugeta Weldezgina; Omlin, Christian Walter Peter; Wang, Long; Yu, David; Parygin, Pavel & Dittmann, Jay [Vis alle 14 forfattere av denne artikkelen] (2023). Spatio-Temporal Anomaly Detection with Graph Networks for Data Quality Monitoring of the Hadron Calorimeter. Sensors. 23(24). doi: 10.3390/s23249679. Fulltekst i vitenarkiv Vis sammendrag

  The Compact Muon Solenoid (CMS) experiment is a general-purpose detector for high-energy collision at the Large Hadron Collider (LHC) at CERN. It employs an online data quality monitoring (DQM) system to promptly spot and diagnose particle data acquisition problems to avoid data quality loss. In this study, we present a semi-supervised spatio-temporal anomaly detection (AD) monitoring system for the physics particle reading channels of the Hadron Calorimeter (HCAL) of the CMS using three-dimensional digi-occupancy map data of the DQM. We propose the GraphSTAD system, which employs convolutional and graph neural networks to learn local spatial characteristics induced by particles traversing the detector and the global behavior owing to shared backend circuit connections and housing boxes of the channels, respectively. Recurrent neural networks capture the temporal evolution of the extracted spatial features. We validate the accuracy of the proposed AD system in capturing diverse channel fault types using the LHC collision data sets. The GraphSTAD system achieves production-level accuracy and is being integrated into the CMS core production system for real-time monitoring of the HCAL. We provide a quantitative performance comparison with alternative benchmark models to demonstrate the promising leverage of the presented system.

• Asres, Mulugeta Weldezgina; Cummings, Grace; Khukhunaishvili, Aleko; Parygin, Pavel; Cooper, Seth I. & Yu, David [Vis alle 8 forfattere av denne artikkelen] (2022). Long Horizon Anomaly Prediction in Multivariate Time Series with Causal Autoencoders. Proceedings of the European Conference of the Prognostics and Health Management Society (PHME). 7(1), s. 21–31. doi: 10.36001/phme.2022.v7i1.3367. Fulltekst i vitenarkiv Vis sammendrag

  Predictive maintenance is essential for complex industrial systems to foresee anomalies before major system faults or ultimate breakdown. However, the existing efforts on Industry 4.0 predictive monitoring are directed at semi-supervised anomaly detection with limited robustness for large systems, which are often accompanied by uncleaned and unlabeled data. We address the challenge of predicting anomalies through data-driven end-to-end deep learning models using early warning symptoms on multivariate time series sensor data. We introduce AnoP, a long multi-timestep anomaly prediction system based on unsupervised attention-based causal residual networks, to raise alerts for anomaly prevention. The experimental evaluation on large data sets from detector health monitoring of the Hadron Calorimeter of the CMS Experiment at LHC CERN demonstrates the promising efficacy of the proposed approach. AnoP predicted around 60% of the anomalies up to seven days ahead, and the majority of the missed anomalies are abnormalities with unpredictable noisy-like behavior. Moreover, it has discovered previously unknown anomalies in the calorimeter's sensors.

• Asres, Mulugeta Weldezgina; Cummings, Grace; Parygin, Pavel; Khukhunaishvili, Aleko; Toms, Maria & Campbell, Alan [Vis alle 10 forfattere av denne artikkelen] (2021). Unsupervised Deep Variational Model for Multivariate Sensor Anomaly Detection, 2021 IEEE International Conference on Progress in Informatics and Computing (PIC). IEEE (Institute of Electrical and Electronics Engineers). ISSN 9781665426558. s. 364–371. doi: 10.1109/PIC53636.2021.9687034. Fulltekst i vitenarkiv Vis sammendrag

  The ever-increasing detector complexity at CERN triggers a call for an increasing level of automation. Since the quality of collected physics data hinges on the quality of the detector components at the time of data-taking, the rapid identification and resolution of detector system anomalies will result in a better amount of high-quality particle data. Therefore, this study proposes CGVAE, a data-driven unsupervised anomaly detection using a deep learning model, for detector system monitoring from multivariate time series sensor data. The CGVAE model is composed of a variational autoencoder with convolutional and gated recurrent unit networks for fast localized feature extraction, long temporal characteristics capturing, and descriptive representation learning. Furthermore, to mitigate signal reconstruction overfitting on anomalous patterns, the CGVAE employs encoded latent feature- and reconstruction-based metrics for anomaly detection. Moreover, the model integrates feature attribution algorithms to explain the contribution of the input sensors to the detected anomalies. The experimental evaluation on large sensor data sets of the Hadron Calorimeter of the CMS experiment demonstrates the efficacy of the proposed model in capturing temporal anomalies.

Se alle arbeider i NVA

• Dorotic, Matilda; Stagno, Emanuela & Asres, Mulugeta Weldezgina (2026). Roundtable Discussion with Stakeholders on Public AI Safety. Fulltekst i vitenarkiv
• Asres, Mulugeta Weldezgina (2026). Privacy-Aware CCTV Safety Monitoring Solutions: For Policy Decision Stakeholders. Fulltekst i vitenarkiv Vis sammendrag

  Privacy and safety security are the cornerstones of a thriving society and are often intricately linked to one another. Robust security measures can safeguard privacy, and respect for privacy can foster responsible security practices and accountability. Striking the right balance between privacy and security is essential, but an ongoing challenge. While automation provides faster response times, better accuracy, and a scalable solution across thousands of cameras, a key challenge remains: concerns about privacy, surveillance, power and control, respect for regulations, and potential model bias. - How to mitigate limitations and enhance protection while maintaining policy utility? - What are the challenges and solutions of visual anonymization methods, and - What are the performance trade-offs between privacy protection, safety monitoring, and explainability accuracy?

• Asres, Mulugeta Weldezgina (2026). Anonymized Crowd Monitoring: Solutions and Challenges within Privacy-Aware CCTV Applications. Fulltekst i vitenarkiv Vis sammendrag

  How can AI enhance public safety without compromising privacy and public trust? Privacy, trust, and security are the foundations of a resilient society—and they are deeply interconnected. Strong security can protect privacy and trust, while privacy-by-design fosters accountability and responsible use of technology. Yet, striking the right balance remains an ongoing challenge. We explore how AI-driven, real-time #CCTV monitoring can detect public safety risks—medium, high, and critical—while preserving privacy, ensuring transparency, and earning public trust. In this talk, I will address critical questions: 💡 What ethical risks arise from AI-based surveillance automation? 💡 Can personal visual data lead to misdetections in today’s anomaly detection models? 💡 Why common anonymization approaches like blurring, pixelization, or even Gen-AI can fall short considerably? 💡 How can real-time XAI improve alert explanations and trust? 💡 What trade-offs exist between privacy protection and alert detection performance? Finally, we will introduce Privacy-Aware CCTV Applications (PACA), a UiA-developed system that pushes the frontier of CCTV video anomaly detection by combining privacy protection, accurate alert detection, explainable alerts, and real-time processing.

• Stagno, Emanuela & Asres, Mulugeta Weldezgina (2024). Responsible AI for Citizen Safety in the Future. Fulltekst i vitenarkiv
• Asres, Mulugeta Weldezgina & Lei, Jiao (2023). AI for Policing-Aware Utility-Preservation on CCTV Anonymization. Fulltekst i vitenarkiv Vis sammendrag

  CCTV surveillance cameras have become increasingly widespread in safety and security, generating vast volumes of data. However, privacy concerns have made it challenging to utilize these technologies. Several de-identification approaches have been proposed for CCTV footage to address this issue, aiming to achieve different levels of anonymization. Nevertheless, most existing solutions focus on generic anonymization, which may need to be more effective in meeting policing requirements, such as anonymous attributes, preserved attributes, and algorithm complexity. In this presentation, we discuss CCTV anonymization challenges and solutions that balance the protection of privacy attributes with maintaining the necessary footage utilities for crowd policing applications. We present approaches that utilize deep learning models to implement end-to-end security and anonymization, which can effectively de-identify CCTV images and video data. We emphasize a versatile design that caters explicitly to the varied requirements of policing applications. It provides the necessary footage while ensuring flexibility in attribute manipulation and preservation control mechanisms. The results we are presenting are based on publicly available surveillance datasets that have been approved by Norwegian authorities or presented in published articles.

• Asres, Mulugeta Weldezgina; Wang, Long & Yu, David (2022). Spatio-Temporal Anomaly Detection for the DQM of the CMS Experiment via Graph Networks. doi: https:/indico.cern.ch/event/1078970/contributions/4833337. Fulltekst i vitenarkiv Vis sammendrag

  The Data Quality Monitoring (DQM) is in place to spot and diagnose particle physics data problems as promptly as possible to avoid data loss in the CMS experiment of CERN. Several studies have proposed leveraging the DQM automation using machine learning algorithms. However, only a few efforts have explored temporal characteristics to underpin system monitoring automation of the CMS detectors via anomaly detection models thus far. Moreover, the DQM for the HCAL detector of the CMS experiment poses multidimensional challenges, yet it is relatively unexplored with machine learning models. In this study, we propose a time-aware deep learning model for anomaly detection on the multidimensional spatial quantity of the DQM for the HCAL detector. The model employs convolutional, recurrent and graph networks to learn three-dimensional spatial characteristics, exploit physical connections of the QIE channels into RBX of the detector systems, and temporal evolution, respectively. Performance evaluation on artificially generated anomalies such as dead, hot and degrading channels has demonstrated the efficacy of the proposed model in detecting and robustly localizing anomalies in temporal and spatial contexts on digioccupacy histograms. Finally, we have carried-out comparison among several models that shows the substantial gain of the proposed model architecture and temporal modeling.

• Asres, Mulugeta Weldezgina & Lei, Jiao (2022). Anonymization for CCTV footage: Challenges and Possible Solutions. Fulltekst i vitenarkiv Vis sammendrag

  An enormous amount of CCTV surveillance cameras produces huge volumes of data When gathered and analyzed, it can be a great contributing factor to a smart city. While we are eager to embrace new technology, we are not willing to intrude on our personal privacy. Privacy and safety security are the cornerstones of a thriving society and are often intricately linked to one another. Safety and security help reduce crime and build trust and cooperation, leading to economic prosperity. Privacy enables freedom of expression, individual autonomy, and protection from abuse. Several de-identification approaches have been proposed for CCTV footage to address this issue, aiming to achieve different levels of anonymization. In this presentation, we discuss the state of the art of CCTV AI-based anonymization approaches, outlining their potential leverages and challenges. We also present how different de-identification methods meet policing requirements for real-time safety monitoring.

• Asres, Mulugeta Weldezgina (2024). Anomaly Detection, Prognostics, and Diagnostics : Machine Learning for the Hadron Calorimeter at the CMS Experiment. Universitetet i Agder. ISSN 9788284271941. Fulltekst i vitenarkiv Vis sammendrag

  Machine Learning (ML) tools have gained immense popularity due to the proliferation of sensor data for monitoring, prognostic, and diagnostic applications in various industrial domains. The growing system complexity and monitoring data volumes of the Large Hadron Collider (LHC) at CERN accentuates the need for automation through advanced ML tools. Detection, identification, and resolution of anomalies are essential to generate more physics collision data of the highest quality. Developing ML tools for complex systems often involves expensive data curation and modeling efforts; it requires adequate, cleaned, and annotated data sets, and addresses the challenges of heterogeneity and curse-of-dimensionality of large data sets. The Compact Muon Solenoid (CMS) experiment—one of the large general-purpose colliders at the LHC—has dedicated substantial monitoring efforts for detector systems and particle data quality; the control and safety systems (DCS/DSS) actively monitor safety-critical problems, and the data quality monitoring (DQM) system mitigates data loss by identifying and diagnosing physics data problems. The existing monitoring systems need to incorporate a wide range of monitoring variables and adapt to the evolving conditions of the detectors. This dissertation focuses on the development of unsupervised anomaly detection (AD), anomaly prediction (AP), and root-cause analysis (RCA) on multivariate time series data sets. We have developed deep learning models for frontend electronics of the Hadron Calorimeter (HCAL) of the CMS detector using diagnostic sensors and high-dimensional particle acquisition channel-monitoring data sets. We have employed subsystem-granularity modeling using a divide-and-conquer approach to monitor the complex HCAL systems with thousands of sensors. Our monitoring tools have detected and identified previously unknown and hard-to-monitor anomalies, and extended the monitoring, diagnostics, and prognostics automation of the HCAL. The developed tools are deployed at CERN and are currently providing essential real-time and offline anomaly monitoring and diagnostics on the frontend electronics of the HCAL and the online DQM system. Our scientific contribution in tackling the challenges for complex system monitoring includes: 1) enhancing multivariate sensor AD, 2) a promising AP approach, 3) context-aware high-dimensional spatio-temporal AD, 4) transfer learning on multi-network deep learning models, 5) lightweight interconnection and divergence discovery for multi-systems with multivariate sensors, and 6) enhancing computational efficiency of anomalies causality discovery on binary anomaly data.

Se alle arbeider i NVA