Fabrizio Palumbo

• Palumbo, Fabrizio & Krøvel, Roy (2026). Data Transparency as a Tool for Accountability: Norwegian Municipalities’ Approaches to Elderly Care Information, and Broader Implications. Scandinavian Journal of Public Administration (SPJA). ISSN 2001-7405. doi: 10.58235/sjpa.33214. Full text in Research Archive Show summary

  This paper presents an investigative analysis of Freedom of Information (FOI) compliance among Norwegian municipalities, focusing specifically on requests for elderly care data. Norway presents a particularly intriguing case, ranking first globally on the World Press Freedom Index whilst achieving only moderate performance on citizen access to public data. Through systematic FOI requests to all Norwegian municipalities, seeking data on reported accidents in elderly care, we assess transparency levels and institutional willingness to share information under the FOI Act. Our findings reveal compliance gaps, with significant variations in data quality and format. While the results underscore the need to improve transparency and accountability in order to foster high-quality elderly care, the research additionally provides valuable insights for understanding the practical limitations of the FOI legislation in democratic governance.

• Biermann, Daniel; Palumbo, Fabrizio; Goodwin, Morten & Granmo, Ole-Christoffer (2024). Streamlining Attention for Text Classification: Sequence Length Reduction with Pooling Attention. In Bramer, Max & Stahl, Frederic (Ed.), Artificial Intelligence XLI: 44th SGAI International Conference on Artificial Intelligence, AI 2024. Springer. ISSN 9783031779145. p. 61–73. doi: 10.1007/978-3-031-77915-2_5. Full text in Research Archive
• Aaby, Pernille; Biermann, Daniel; Yazidi, Anis; Mello, Gustavo Borges Moreno e & Palumbo, Fabrizio (2023). Exploring Multilingual Word Embedding Alignments in BERT Models: A Case Study of English and Norwegian. In Bramer, Max & Stahl, Frederic (Ed.), Artificial Intelligence XL: 43rd SGAI International Conference on Artificial Intelligence, AI 2023. Springer. ISSN 9783031479939. p. 47–58. doi: 10.1007/978-3-031-47994-6_4. Full text in Research Archive Show summary

  Contextual language models, such as transformers, can solve a wide range of language tasks ranging from text classification to question answering and machine translation. Like many deep learning models, the performance heavily depends on the quality and amount of data available for training. This poses a problem for low-resource languages, such as Norwegian, that can not provide the necessary amount of training data. In this article, we investigate the use of multilingual models as a step toward overcoming the data sparsity problem for minority languages. In detail, we study how words are represented by multilingual BERT models across two languages of our interest: English and Norwegian. Our analysis shows that multilingual models similarly encode English-Norwegian word pairs. The multilingual model automatically aligns semantics across languages without supervision. Additionally, our analysis also shows that embedding a word encodes information about the language to which it belongs. We, therefore, believe that in pre-trained multilingual models' knowledge from one language can be transferred to another without direct supervision and help solve the data sparsity problem for minor languages.

• Fridman, Marina; Krøvel, Roy & Palumbo, Fabrizio (2023). How (not to) Run an AI Project in Investigative Journalism. Journalism Practice. ISSN 1751-2786. doi: 10.1080/17512786.2023.2253797. Full text in Research Archive Show summary

  Data journalists are increasingly reliant on automation and artificial intelligence (AI) to process and analyse massive datasets. AI can contribute to journalism by creating visualizations, verifying accuracy of information, analysing historical data, monitoring social media, finding patterns and outliers, generating text and much more. However, the integration of AI into the newsroom comes with its own challenges. In this article, we take a practice- based approach to develop a deeper understanding of how to overcome such challenges. Our teams of data scientists, AI experts and journalists took on four projects incorporating data science and machine learning into investigative journalism. From those experiences, we found that access to data at scale, data quality and reworking the concept of “newsworthy” as a machine learning question were the most significant obstacles to deploying AI in the newsroom. We recommend closer collaborations between team members of different disciplines to create a truly trans-disciplinary approach, as well as some practical considerations for choosing projects to facilitate successful AI-assisted investigations.

• Krøvel, Roy; Palumbo, Fabrizio & Orgeret, Kristin (2023). Introduction: Understanding Roots and Betweenness Defining Safety of Journalists as a Sub-Field of Research. Reading between the Lines. Journalism Studies. ISSN 1461-670X. 24(7), p. 825–837. doi: 10.1080/1461670X.2023.2206494. Full text in Research Archive
• Opalic, Sven Myrdahl; Palumbo, Fabrizio; Goodwin, Morten; Lei, Jiao; Nielsen, Henrik Kofoed & Kolhe, Mohan Lal (2023). COST-WINNERS: COST reduction WIth Neural NEtworks-based augmented Random Search for simultaneous thermal and electrical energy storage control. Journal of Energy Storage. ISSN 2352-152X. 72(B). doi: 10.1016/j.est.2023.108202. Full text in Research Archive Show summary

  The combination of local renewable energy production, dynamic loads, and multiple energy storage systems with different dynamics requires sophisticated control systems to maximize the energy cost efficiency of the combined energy system. Battery and thermal energy storage systems can be combined to increase the local use of on-site renewable energy, reduce peak power demand, and exploit time-of-use energy pricing. In this paper, we focus on how the augmented random search algorithm and artificial neural networks can be used together to solve an energy cost optimization problem involving the control of a battery energy storage system and a thermal energy storage system at the same time in a smart warehouse. As part of this work, a simulated training environment made using the data from the smart warehouse’s operations. In addition to the energy storage systems, the warehouse energy system has integrated a large roof mounted photovoltaic power plant and an industrial-scale cooling system. The developed solution is able to minimize the energy costs by modulating both energy systems, depending on the situation. Additionally, when it is tested against the state-of-the-art solutions, our developed solution at worst matches performance when the alternative algorithm is allowed to increase training time by a factor of nearly three. On average, our presented solution doubles the performance of the benchmark algorithm with much less computational resource expenditure.

• Lien, Henrik; Biermann, Daniel; Palumbo, Fabrizio & Goodwin, Morten (2022). An Exploration of Semi-supervised Text Classification. In Iliadis, Lazaros; Jayne, Chrisina; Tefas, Anastasios & Pimenidis, Elias (Ed.), Engineering Applications of Neural Networks - 23rd International Conference, EAAAI/EANN 2022. Springer Nature. ISSN 9783031082221. p. 477–488. doi: 10.1007/978-3-031-08223-8_39. Full text in Research Archive Show summary

  Good performance in supervised text classification is usually obtained with the use of large amounts of labeled training data. However, obtaining labeled data is often expensive and time-consuming. To overcome these limitations, researchers have developed Semi-Supervised learning (SSL) algorithms exploiting the use of unlabeled data, which are generally easy and free to access. With SSL, unlabeled and labeled data are combined to outperform Supervised-Learning algorithms. However, setting up SSL neural networks for text classification is cumbersome and frequently based on a trial and error process. We show that the hyperparameter configuration significantly impacts SSL performance, and the learning rate is the most influential parameter. Additionally, increasing model size also improves SSL performance, particularly when less pre-processing data are available. Interestingly, as opposed to feed-forward models, recurrent models generally reach a performance threshold as pre-processing data size increases. This article expands the knowledge on hyperparameters and model size in relation to SSL application in text classification. This work supports the use of SSL work in future NLP projects by optimizing model design and potentially lowering training time, particularly if time-restricted.

• Palumbo, Fabrizio; Buji, Adam; Yazidi, Anis & Haugerud, Hårek (2022). Genetic Algorithms For Tightening Security. In Ukjent, Ukjent (Eds.), Proceedings of the 14th IFIP Wireless and Mobile Networking Conference. IEEE (Institute of Electrical and Electronics Engineers). ISSN 9783903176522. doi: 10.23919/WMNC56391.2022.9954297. Full text in Research Archive Show summary

  Proper configuration of operating systems and program parameters is known to be a key security factor in order to remove vulnerabilities. It is known that vulnerabilities can be caused by a human misconfiguration or by an improper chain of parameter settings. It is impossible to find an optimal combination manually due to the enormous number of possible configurations. In this article, we resort to a Genetic Algorithm equipped with a user-defined fitness function in order to compute a configuration of high fitness. Our work presents a two-fold contribution. First, we successfully use a GA to implement a moving target defense by alerting the configuration regularly in order to spoil an attacker's reconnaissance efforts. The GA tightens the security solution by evolving the fitness of the configuration over generations while maintaining diversity within generations across a pool of servers. This resulted in high-quality configurations crucial for a successful moving target defense strategy. Second, we try to find a compromise between tightening the security of the configuration and maintaining the Quality of Service (QoS) on a web server. In practice, usually tightening security on a web server comes at the cost of a decrease in QoS.

• Palumbo, Fabrizio; Serneels, Bram & Yaksi, Emre (2021). Optimized protocol for conditioned place avoidance learning in juvenile zebrafish. STAR Protocols. 2(2). doi: 10.1016/j.xpro.2021.100465. Full text in Research Archive Show summary

  Conditioned place avoidance assays are broadly used in mammals to study different cognitive aspects of operant learning. Here, we introduce a series of experimental designs for training juvenile zebrafish in short-term and long-term conditioned place avoidance assays. Our goal is to promote standardization of animal handling procedures and setup conditions to improve animal welfare and reproducibility while studying operant learning behaviors in juvenile zebrafish. For complete details on the use and execution of this protocol, please refer to Palumbo et al. (2020).

• Fore, Stephanie Liesbeth; Hinrichsen, Francisca Daniela Acuna; Mutlu, Aytac Kadir; Bartoszek-Kandler, Ewelina Magdalena; Serneels, Bram & Faturos, Nicholas Guy [Show all 14 contributors for this article] (2020). Functional properties of habenular neurons are determined by developmental stage and sequential neurogenesis. Science Advances. doi: 10.1126/sciadv.aaz3173. Full text in Research Archive Show summary

  The developing brain undergoes drastic alterations. Here, we investigated developmental changes in the habenula, a brain region that mediates behavioral flexibility during learning, social interactions, and aversive experiences. We showed that developing habenular circuits exhibit multiple alterations that lead to an increase in the structural and functional diversity of cell types, inputs, and functional modules. As the habenula develops, it sequentially transforms into a multisensory brain region that can process visual, olfactory, mechanosensory, and aversive stimuli. Moreover, we observed that the habenular neurons display spatiotemporally structured spontaneous activity that shows prominent alterations and refinement with age. These alterations in habenular activity are accompanied by sequential neurogenesis and the integration of distinct neural clusters across development. Last, we revealed that habenular neurons with distinct functional properties are born sequentially at distinct developmental time windows. Our results highlight a strong link between the functional properties of habenular neurons and their precise birthdate.

• Kermen, Florence; Darnet, Lea; Wiest, Christoph; Palumbo, Fabrizio; Bechert, Jack & Uslu, Ozge [Show all 7 contributors for this article] (2020). Stimulus-specific behavioral responses of zebrafish to a large range of odors exhibit individual variability. BMC Biology. 18(1). doi: 10.1186/s12915-020-00801-8. Full text in Research Archive Show summary

  Background Odor-driven behaviors such as feeding, mating, and predator avoidance are crucial for animal survival. The neural pathways processing these behaviors have been well characterized in a number of species, and involve the activity of diverse brain regions following stimulation of the olfactory bulb by specific odors. However, while the zebrafish olfactory circuitry is well understood, a comprehensive characterization linking odor-driven behaviors to specific odors is needed to better relate olfactory computations to animal responses. Results Here, we used a medium-throughput setup to measure the swimming trajectories of 10 zebrafish in response to 17 ecologically relevant odors. By selecting appropriate locomotor metrics, we constructed ethograms systematically describing odor-induced changes in the swimming trajectory. We found that adult zebrafish reacted to most odorants using different behavioral programs and that a combination of a few relevant behavioral metrics enabled us to capture most of the variance in these innate odor responses. We observed that individual components of natural food and alarm odors do not elicit the full behavioral response. Finally, we show that zebrafish blood elicits prominent defensive behaviors similar to those evoked by skin extract and activates spatially overlapping olfactory bulb domains. Conclusion Altogether, our results highlight a prominent intra- and inter-individual variability in zebrafish odor-driven behaviors and identify a small set of waterborne odors that elicit robust responses. Our behavioral setup and our results will be useful resources for future studies interested in characterizing innate olfactory behaviors in aquatic animals.

• Palumbo, Fabrizio; Serneels, Bram; Pelgrims, Robbrecht & Yaksi, Emre (2020). The Zebrafish dorsal habenula is required for updating learned behaviors. Cell reports. 32(8). doi: 10.1016/j.celrep.2020.108054. Full text in Research Archive Show summary

  Operant learning requires multiple cognitive processes, such as learning, prediction of potential outcomes, and decision-making. It is less clear how interactions of these processes lead to the behavioral adaptations that allow animals to cope with a changing environment. We show that juvenile zebrafish can perform conditioned place avoidance learning, with improving performance across development. Ablation of the dorsolateral habenula (dlHb), a brain region involved in associative learning and prediction of outcomes, leads to an unexpected improvement in performance and delayed memory extinction. Interestingly, the control animals exhibit rapid adaptation to a changing learning rule, whereas dlHb-ablated animals fail to adapt. Altogether, our results show that the dlHb plays a central role in switching animals’ strategies while integrating new evidence with prior experience.

• Foré, Stephanie; Palumbo, Fabrizio; Pelgrims, Robbrecht & Yaksi, Emre (2018). Information processing in the vertebrate habenula. Seminars in Cell and Developmental Biology. ISSN 1084-9521. 78, p. 130–139. doi: 10.1016/j.semcdb.2017.08.019. Full text in Research Archive

View all works in NVA

• Palumbo, Fabrizio; Kalu, Nelly; Wei, Sisi & Tshabalala, Tshepo (2025). Uncovering Wrongdoing Using AI: Methods, Techniques and Challenges. doi: https:/gijc2025.org/program/schedule/sessions/444405cbc31cd26f36182c0a59ead817/. Full text in Research Archive Show summary

  While AI can pose challenges, it has helped journalists to work with large data troves and find hidden patterns - and its tools are available to both large and small newsrooms. In this practical panel that draws on powerful investigative projects, data experts show how AI tools can be leveraged to comb through everything from court records to images and political speeches to help uncover stories of public interest.

• Palumbo, Fabrizio (2023). Datadriven Stories 2.0 from Excel to AI. Full text in Research Archive Show summary

  The lecture addressed the shifting landscape of data journalism by focusing on practical methodologies for moving beyond traditional spreadsheets and into advanced machine learning applications. Palumbo discussed bridging the tech gap by showing how investigative journalists can evolve their workflows from basic software like Excel into AI-driven data pipelines. Drawing from his academic research and firsthand newsroom collaborations, he shared insights on the realities of AI projects, emphasizing the framework outlined in his co-authored paper, How (not to) Run an AI Project in Investigative Journalism. Finally, he highlighted how to navigate project pivots through real-world case studies—such as partnering with Norwegian newsrooms to audit complex fisheries and vessel ownership data—to demonstrate how investigative teams must systematically build reliable code infrastructure rather than treating AI as a "magic bullet" fix.

• Kvalnes, Pål Arne; Palumbo, Fabrizio & Krøvel, Roy (2023). Kva skal journalistar med kunstig intelligens? [Internet]. oslomet.no. Full text in Research Archive
• Lal, Pradeep; Fiskum, Vegard; Kermen, Florence; Ostenrath, Anna Maria; Palumbo, Fabrizio & Jurisch-Yaksi, Nathalie [Show all 9 contributors for this article] (2022). The role of GnRH3 neurons in modulating olfactory computations and odor induced feeding behaviors. Full text in Research Archive
• Palumbo, Fabrizio (2022). Artificial Investigative Journalism, How much of it can actually be artificial? Full text in Research Archive
• Lal, Pradeep; Weth-Malsch, Daniela; Fiskum, Vegard; Kermen, Florence; Palumbo, Fabrizio & Pelgrims, Robbrecht [Show all 10 contributors for this article] (2018). Gonadotropin-releasing hormone (GnRH) neuromodulation of olfactory computations in zebrafish brain. Full text in Research Archive
• Lal, Pradeep; Weth-Malsch, Daniela; Fiskum, Vegard; Palumbo, Fabrizio; Kermen, Florence & Pelgrims, Robbrecht [Show all 10 contributors for this article] (2018). The function of gonadotropin-releasing hormone (GnRH) neurons in olfactory computations. Full text in Research Archive
• Palumbo, Fabrizio; Pelgrims, Robbrecht & Yaksi, Emre (2018). Studying conditioned place avoidance in juvenile zebrafish. Full text in Research Archive
• Lal, Pradeep; Kermen, Florence; Palumbo, Fabrizio; Pelgrims, Robbrecht & Yaksi, Emre (2017). Gondotropin-releasing hormone (GnRH) neuromodulation of olfactory computaions in zebrafish brain. Full text in Research Archive

View all works in NVA