Andrea Merlina

• Nawrath, Maximilian; Merlina, Andrea; Knight, Jemmima; Welch, Sam A.; Rashidian, Mahla & Seifert-Dähnn, Isabel (2026). Validating Large Language Models for Title-Abstract Screening in Low-Prevalence Systematic Reviews: An Environmental Science Case Study. Information. 17(5). doi: 10.3390/info17050501. Fulltekst i vitenarkiv Vis sammendrag

  Literature screening is a major bottleneck in systematic reviews, yet Large Language Models (LLMs) can substantially reduce workloads. However, performance varies across models and is sensitive to evaluation metrics, particularly in low-prevalence screening contexts. We validated five LLMs (GPT-4.1, Claude 3.5 Sonnet, Gemini 2.0 Flash, DeepSeek V3, and Mistral Large) against a 500-record gold-standard dataset (8 inclusions; 1.6% prevalence) using a conservative zero-shot prompt aligned with standard systematic review workflows. Performance was assessed through classification metrics (sensitivity, specificity, precision), logistic regression (GLM; Firth-penalised where separation occurred), and agreement indices (Cohen’s κ, MCC, PABAK, Gwet’s AC1). Gemini 2.0 Flash and Mistral Large showed no false negatives (1.00) but differed in specificity (0.858 vs. 0.697) and accuracy (0.860 vs. 0.702). GPT-4.1 and Claude 3.5 Sonnet performed identically (sensitivity 0.875; specificity 0.876; accuracy 0.876). In contrast, DeepSeek V3 maximised specificity (0.980) and accuracy (0.970) but demonstrated lower sensitivity (0.375). Regression analyses confirmed strong positive associations with human decisions (OR 28.9–49.5). Agreement indices revealed the expected low-prevalence artefact, with Cohen’s κ low despite high concordance while MCC, PABAK, and AC1 indicated substantially stronger agreement. Our results highlight a fundamental sensitivity-specificity trade-off, with conclusions dependent on the evaluation framework chosen. LLMs may meaningfully support title-abstract screening as decision-support tools, provided that human oversight is maintained and validation is transparent and reproducible.

• Song, Weijia; Yang, Yuting; Liu, Thompson; Merlina, Andrea; Garrett, Thiago & Vitenberg, Roman [Vis alle 10 forfattere av denne artikkelen] (2022). Cascade: An Edge Computing Platform for Real-time Machine Intelligence, ApPLIED '22: Proceedings of the 2022 Workshop on Advanced tools, programming languages, and PLatforms for Implementing and Evaluating algorithms for Distributed systems. Association for Computing Machinery (ACM). ISSN 9781450392808. s. 2–6. doi: 10.1145/3524053.3542741. Fulltekst i vitenarkiv Vis sammendrag

  Intelligent IoT is a prerequisite for societal priorities such as a smart power grid, smart urban infrastructures and smart highways. These applications bring requirements such as real-time guarantees, data and action consistency, fault-tolerance, high availability, temporal data indexing, scalability, and even self-organization and self-stabilization. Existing platforms are oriented towards asynchronous, out of band upload of data to the cloud: Important functionality, but not enough to address the need. Cornell's Cascade project seeks to close the gap by creating a new platform for hosting ML and AI, optimized to achieve sharply lower delay and substantially higher bandwidth than in any existing platform. At the same time, Cascade introduces much stronger guarantees - a mix that we believe will be particularly appealing in applications where events should trigger a quick and trustworthy response. This short paper is intended as a brief overview of the effort, with details to be published elsewhere.

• Merlina, Andrea; Vitenberg, Roman & Setty, Vinay (2022). A General and Configurable Framework for Blockchain-based Marketplaces. I Hong, Jiman; Bures, Miroslav; Park, Juw Won & Cerny, Tomas (Red.), SAC '22: Proceedings of the 37th ACM/SIGAPP Symposium on Applied Computing. Association for Computing Machinery (ACM). ISSN 9781450387132. s. 216–225. doi: 10.1145/3477314.3507039. Fulltekst i vitenarkiv
• Merlina, Andrea (2019). BlockML: a useful proof of work system based on machine learning tasks. I Nawab, Faisal & Riviere, Etienne (Red.), Middleware '19: Proceedings of the 20th International Middleware Conference Doctoral Symposium. Association for Computing Machinery (ACM). ISSN 9781450370394. s. 6–9. doi: 10.1145/3366624.3368156. Fulltekst i vitenarkiv

Se alle arbeider i NVA

• Norli, Marit; Jaccard, Pierre Franqois; Hansen, Peter Stig; Merlina, Andrea; Risbakk, Nathalie Marquesin & Vedal, Jens [Vis alle 8 forfattere av denne artikkelen] (2026). New IFCB pipeline. Fulltekst i vitenarkiv Vis sammendrag

  Presenting NIVAs new IFCB pipeline going from from images captured on the SOOP MS Color Fantasy using an IFCB, to data being stored as cell counts of phytoplankton taxa in the NIVA BioDB database. The pipeline includes use of an internally developed app to predict, annotate and retrain a CNN model.

• Merlina, Andrea; Vitenberg, Roman & Setty, Vinay Jayarama (2019). Towards Useful Proof of Work: combining Blockchain and Machine Learning. Fulltekst i vitenarkiv
• Sample, James Edward; Leirvik, Kim Nes; Bhakta, Debhasish & Merlina, Andrea (2025). Overview of the SeaBee data platform. Norsk institutt for vannforskning (NIVA). ISSN 9788257778538. Fulltekst i vitenarkiv Vis sammendrag

  SeaBee is the Norwegian national infrastructure for drone-based research, mapping and monitoring, funded by the Research Council of Norway. This report provides a high-level overview of the software, tools and data pipelines developed during Phase 1 of the SeaBee project (2020 to 2025).

Se alle arbeider i NVA