2026
Other
Open Access
Retrieval-augmented generation for predicting cellular responses to gene perturbation
Di Francesco Andrea Giuseppe, Rubbi Andrea, Lio Pietro
Predicting how cells respond to genetic perturbations is fundamental to understanding gene function, disease mechanisms, and therapeutic development. While recent deep learning approaches have shown promise in modeling single-cell perturbation responses, they struggle to generalize across cell types and perturbation contexts due to limited contextual information during generation. We introduce PT-RAG (Perturbation-aware Two-stage Retrieval-Augmented Generation), a novel framework that extends Retrieval-Augmented Generation beyond traditional language-model applications to cellular biology. Unlike standard RAG systems designed for text retrieval with pre-trained LLMs, perturbation retrieval lacks established similarity metrics and requires learning what constitutes relevant context, making differentiable retrieval essential. PT-RAG addresses this through a two-stage pipeline: first, retrieving candidate perturbations K using GenePT embeddings, then adaptively refining the selection through Gumbel-Softmax discrDOI: 10.48550/arxiv.2603.07233
Metrics:
Di Francesco Andrea Giuseppe, Rubbi Andrea, Lio Pietro
Predicting how cells respond to genetic perturbations is fundamental to understanding gene function, disease mechanisms, and therapeutic development. While recent deep learning approaches have shown promise in modeling single-cell perturbation responses, they struggle to generalize across cell types and perturbation contexts due to limited contextual information during generation. We introduce PT-RAG (Perturbation-aware Two-stage Retrieval-Augmented Generation), a novel framework that extends Retrieval-Augmented Generation beyond traditional language-model applications to cellular biology. Unlike standard RAG systems designed for text retrieval with pre-trained LLMs, perturbation retrieval lacks established similarity metrics and requires learning what constitutes relevant context, making differentiable retrieval essential. PT-RAG addresses this through a two-stage pipeline: first, retrieving candidate perturbations K using GenePT embeddings, then adaptively refining the selection through Gumbel-Softmax discrDOI: 10.48550/arxiv.2603.07233
Metrics:
2026
Conference article
Open Access
When reducing representations improves performance
Pasin Andrea, Faggioli Guglielmo, Ferro Nicola, Perego Raffaele, Tonellotto Nicola
Neural models have transformed Information Retrieval (IR) by enabling semantic search, representing queries and documents as dense embeddings in latent spaces. However, recent works indicate the contribution of single dimensions in these representations to ranking quality is uneven: some dimensions are essential, while others may even degrade performance. Dimension IMportance Estimators (DIMEs) are heuristics to guide the search for the subsets of dimensions that induce an optimal subspace where retrieval is more effective. To explore these subspaces, DIMEs rely on two simplifying assumptions: the linearity of subspaces and the independence of dimensions. In this paper, we move a step forward by relaxing the independence assumption and employing genetic algorithms to select the optimal set of dimensions. We show that selecting optimal dimensions for individual queries can achieve up to 0.981 nDCG@10 and 0.831 AP using state-of-the-art dense retrieval models on the considered datasets. Additionally, we identify subsets of dimensions that improve ranking quality across multiple queries simultaneously. Finally, we show that a dataset-specific subset of dimensions enables dense retrieval models to generalize across other datasets without loss of performance.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16483, pp. 466-480. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21289-4_30
Metrics:
Pasin Andrea, Faggioli Guglielmo, Ferro Nicola, Perego Raffaele, Tonellotto Nicola
Neural models have transformed Information Retrieval (IR) by enabling semantic search, representing queries and documents as dense embeddings in latent spaces. However, recent works indicate the contribution of single dimensions in these representations to ranking quality is uneven: some dimensions are essential, while others may even degrade performance. Dimension IMportance Estimators (DIMEs) are heuristics to guide the search for the subsets of dimensions that induce an optimal subspace where retrieval is more effective. To explore these subspaces, DIMEs rely on two simplifying assumptions: the linearity of subspaces and the independence of dimensions. In this paper, we move a step forward by relaxing the independence assumption and employing genetic algorithms to select the optimal set of dimensions. We show that selecting optimal dimensions for individual queries can achieve up to 0.981 nDCG@10 and 0.831 AP using state-of-the-art dense retrieval models on the considered datasets. Additionally, we identify subsets of dimensions that improve ranking quality across multiple queries simultaneously. Finally, we show that a dataset-specific subset of dimensions enables dense retrieval models to generalize across other datasets without loss of performance.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16483, pp. 466-480. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21289-4_30
Metrics:
See at:
CNR IRIS 
| link.springer.com | doi.org 
| CNR IRIS | CNR IRIS
2026
Conference article
Open Access
Multivector Reranking in the era of strong first-stage retrievers
Martinico Silvio, Nardini Franco Maria, Rulli Cosimo, Venturini Rossano
Learned multivector representations power modern search systems with strong retrieval effectiveness, but their real-world use is limited by the high cost of exhaustive token-level retrieval. Therefore, most systems adopt a gather-and-refine strategy, where a lightweight gather phase selects candidates for full scoring. However, this approach requires expensive searches over large token-level indexes and often misses the documents that would rank highest under full similarity. In this paper, we reproduce several state-of-the-art multivector retrieval methods on two publicly available datasets, providing a clear picture of the current multivector retrieval field and observing the inefficiency of token-level gathering. Building on top of that, we show that replacing the token-level gather phase with a single-vector document retriever—specifically, a learned sparse retriever (LSR)—produces a smaller and more semantically coherent candidate set. This recasts the gather-and-refine pipeline into the well-established two-stage retrieval architecture. As retrieval latency decreases, query encoding with two neural encoders becomes the dominant computational bottleneck. To mitigate this, we integrate recent inference-free LSR methods, demonstrating that they preserve the retrieval effectiveness of the dual-encoder pipeline while substantially reducing query encoding time. Finally, we investigate multiple reranking configurations that balance efficiency, memory, and effectiveness, and we introduce two optimization techniques that prune low-quality candidates early. Empirical results show that these techniques improve retrieval efficiency by up to 1.8× with no loss in quality. Overall, our two-stage approach achieves over 24× speedup over the state-of-the-art multivector retrieval systems, while maintaining comparable or superior retrieval quality.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16485, pp. 49-65. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21324-2_4
Metrics:
Martinico Silvio, Nardini Franco Maria, Rulli Cosimo, Venturini Rossano
Learned multivector representations power modern search systems with strong retrieval effectiveness, but their real-world use is limited by the high cost of exhaustive token-level retrieval. Therefore, most systems adopt a gather-and-refine strategy, where a lightweight gather phase selects candidates for full scoring. However, this approach requires expensive searches over large token-level indexes and often misses the documents that would rank highest under full similarity. In this paper, we reproduce several state-of-the-art multivector retrieval methods on two publicly available datasets, providing a clear picture of the current multivector retrieval field and observing the inefficiency of token-level gathering. Building on top of that, we show that replacing the token-level gather phase with a single-vector document retriever—specifically, a learned sparse retriever (LSR)—produces a smaller and more semantically coherent candidate set. This recasts the gather-and-refine pipeline into the well-established two-stage retrieval architecture. As retrieval latency decreases, query encoding with two neural encoders becomes the dominant computational bottleneck. To mitigate this, we integrate recent inference-free LSR methods, demonstrating that they preserve the retrieval effectiveness of the dual-encoder pipeline while substantially reducing query encoding time. Finally, we investigate multiple reranking configurations that balance efficiency, memory, and effectiveness, and we introduce two optimization techniques that prune low-quality candidates early. Empirical results show that these techniques improve retrieval efficiency by up to 1.8× with no loss in quality. Overall, our two-stage approach achieves over 24× speedup over the state-of-the-art multivector retrieval systems, while maintaining comparable or superior retrieval quality.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16485, pp. 49-65. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21324-2_4
Metrics:
See at:
CNR IRIS | link.springer.com | doi.org | CNR IRIS | CNR IRIS
2026
Book
Restricted
Large-scale HPC approaches and applications on highly distributed platforms
Antelmi Alessia, Carlini Emanuele
The ever-increasing complexity of scientific and industrial challenges due to the enormous amount of data available nowadays requires advanced high-performance computing (HPC) solutions capable of processing and analyzing data efficiently on highly distributed platforms. Traditional centralized HPC systems frequently fall short of the demands of contemporary large-scale applications (e.g., large language models), prompting a move towards more flexible and scalable distributed computing environments. Furthermore, the growing emphasis on the environmental impact of large-scale computing has highlighted the need for sustainable computing practices that minimize energy consumption and carbon footprint. This special issue targets contributions that investigate both the challenges and the opportunities arising from this evolution. The accepted articles highlight enhancements in five key areas: (i) HPC in the cloud continuum, (ii) heterogeneous HPC architectures, performance tools, and programming models, (iii) parallel and distributed algorithms and applications, (iv) data management and storage systems, and (v) sustainable and energy-efficient HPC systems. In total, 29 submissions were received, and 20 papers were selected after a rigorous peer-review process. Collectively, these contributions provide a representative snapshot of current research efforts towards resilient, efficient, and sustainable HPC approaches and applications on highly distributed platforms.Source: FUTURE GENERATION COMPUTER SYSTEMS, vol. 179 (issue 108365)
DOI: 10.1016/j.future.2025.108365
Metrics:
Antelmi Alessia, Carlini Emanuele
The ever-increasing complexity of scientific and industrial challenges due to the enormous amount of data available nowadays requires advanced high-performance computing (HPC) solutions capable of processing and analyzing data efficiently on highly distributed platforms. Traditional centralized HPC systems frequently fall short of the demands of contemporary large-scale applications (e.g., large language models), prompting a move towards more flexible and scalable distributed computing environments. Furthermore, the growing emphasis on the environmental impact of large-scale computing has highlighted the need for sustainable computing practices that minimize energy consumption and carbon footprint. This special issue targets contributions that investigate both the challenges and the opportunities arising from this evolution. The accepted articles highlight enhancements in five key areas: (i) HPC in the cloud continuum, (ii) heterogeneous HPC architectures, performance tools, and programming models, (iii) parallel and distributed algorithms and applications, (iv) data management and storage systems, and (v) sustainable and energy-efficient HPC systems. In total, 29 submissions were received, and 20 papers were selected after a rigorous peer-review process. Collectively, these contributions provide a representative snapshot of current research efforts towards resilient, efficient, and sustainable HPC approaches and applications on highly distributed platforms.Source: FUTURE GENERATION COMPUTER SYSTEMS, vol. 179 (issue 108365)
DOI: 10.1016/j.future.2025.108365
Metrics:
See at:
Future Generation Computer Systems | CNR IRIS | CNR IRIS | www.sciencedirect.com
2026
Journal article
Open Access
Getting off the DIME: dimension pruning via dimension importance estimation for dense information retrieval
Faggioli Guglielmo, Ferro Nicola, Perego Raffaele, Tonellotto Nicola
Dense Information Retrieval (IR) systems rely on neural networks to embed documents and queries within a latent low-dimensional space. Among the Dense IR approaches, bi-encoders are particularly popular, as they achieve state-of-the-art performance and allow for efficient encoding of documents and queries. Nevertheless, using this class of systems, by construction, all the documents and queries are represented using the same set of dimensions. In this article, we introduce the Manifold Clustering (MC) hypothesis which states that, for each query, there exists a query-dependent manifold of the original embedding space where the query and documents relevant to it cluster more effectively. We empirically validate the MC hypothesis showing that it is possible to find a query-dependent linear subspace of the original embedding space where high retrieval effectiveness is achieved.Source: ACM TRANSACTIONS ON INFORMATION SYSTEMS, vol. 44 (issue 1), pp. 1-34
DOI: 10.1145/3765619
Metrics:
Faggioli Guglielmo, Ferro Nicola, Perego Raffaele, Tonellotto Nicola
Dense Information Retrieval (IR) systems rely on neural networks to embed documents and queries within a latent low-dimensional space. Among the Dense IR approaches, bi-encoders are particularly popular, as they achieve state-of-the-art performance and allow for efficient encoding of documents and queries. Nevertheless, using this class of systems, by construction, all the documents and queries are represented using the same set of dimensions. In this article, we introduce the Manifold Clustering (MC) hypothesis which states that, for each query, there exists a query-dependent manifold of the original embedding space where the query and documents relevant to it cluster more effectively. We empirically validate the MC hypothesis showing that it is possible to find a query-dependent linear subspace of the original embedding space where high retrieval effectiveness is achieved.Source: ACM TRANSACTIONS ON INFORMATION SYSTEMS, vol. 44 (issue 1), pp. 1-34
DOI: 10.1145/3765619
Metrics:
See at:
dl.acm.org | CNR IRIS | ACM Transactions on Information Systems | CNR IRIS
2026
Journal article
Open Access
Projection-displacement-based query performance prediction for embedded space of dense retrievers
Datta Suchana, Faggioli Guglielmo, Ferro Nicola, Ganguly Debasis, Muntean Cristina Ioana, Perego Raffaele, Tonellotto Nicola
Recent advances in representation learning have enabled neural Information Retrieval (IR) systems to use learned dense representations for queries and documents to effectively handle semantics, language nuances, and vocabulary mismatch problems. In contrast to traditional IR systems that rely on word matching, dense IR models exploit query/document similarity in dense latent spaces to account for semantics. This requires substantial training data and comes with increased computational demands. Thus, it would be beneficial to predict how a system will perform for a given query to decide whether a dense IR model is the best option or alternatives should be used. Traditional Query Performance Prediction (QPP) models are designed for lexical IR approaches and perform sub-optimally when applied to dense neural IR systems. Therefore, there has been a renewed interest in QPP methods to improve their effectiveness for dense neural IR models. While the results of the new QPP methods are generally encouraging, there is ample room for improvement in absolute performance and stability. We argue that by using features more aligned with the underlying rationale of dense IR models, we can enhance the performance of QPP. In this respect, we propose the Projection-Displacement-Based QPP (PDQPP), which exploits the geometric properties of dense IR models, projects queries and retrieved documents onto subspaces defined by pseudo-relevant documents, and considers changes in retrieval scores within them as a proxy for retrieval coherence. Minor score changes suggest robust and coherent retrieval, while significant alterations indicate semantic divergence and potentially poor performance. Results over a wide range of experimental settings on both traditional (TREC Robust) and neural-oriented (TREC Deep Learning) test collections show that PDQPP mostly outperforms the state-of-the-art QPP baselines.Source: ACM TRANSACTIONS ON INFORMATION SYSTEMS, vol. 44 (issue 1), pp. 1-30
DOI: 10.1145/3765617
Metrics:
Datta Suchana, Faggioli Guglielmo, Ferro Nicola, Ganguly Debasis, Muntean Cristina Ioana, Perego Raffaele, Tonellotto Nicola
Recent advances in representation learning have enabled neural Information Retrieval (IR) systems to use learned dense representations for queries and documents to effectively handle semantics, language nuances, and vocabulary mismatch problems. In contrast to traditional IR systems that rely on word matching, dense IR models exploit query/document similarity in dense latent spaces to account for semantics. This requires substantial training data and comes with increased computational demands. Thus, it would be beneficial to predict how a system will perform for a given query to decide whether a dense IR model is the best option or alternatives should be used. Traditional Query Performance Prediction (QPP) models are designed for lexical IR approaches and perform sub-optimally when applied to dense neural IR systems. Therefore, there has been a renewed interest in QPP methods to improve their effectiveness for dense neural IR models. While the results of the new QPP methods are generally encouraging, there is ample room for improvement in absolute performance and stability. We argue that by using features more aligned with the underlying rationale of dense IR models, we can enhance the performance of QPP. In this respect, we propose the Projection-Displacement-Based QPP (PDQPP), which exploits the geometric properties of dense IR models, projects queries and retrieved documents onto subspaces defined by pseudo-relevant documents, and considers changes in retrieval scores within them as a proxy for retrieval coherence. Minor score changes suggest robust and coherent retrieval, while significant alterations indicate semantic divergence and potentially poor performance. Results over a wide range of experimental settings on both traditional (TREC Robust) and neural-oriented (TREC Deep Learning) test collections show that PDQPP mostly outperforms the state-of-the-art QPP baselines.Source: ACM TRANSACTIONS ON INFORMATION SYSTEMS, vol. 44 (issue 1), pp. 1-30
DOI: 10.1145/3765617
Metrics:
See at:
dl.acm.org | CNR IRIS | ACM Transactions on Information Systems | CNR IRIS
2026
Journal article
Restricted
Dynamic workload balancing in decentralized edge systems: a marginal cost approach
Carlini Emanuele, Dazzi Patrizio, Ferrucci Luca, Massa Jacopo, Mordacchini Matteo
The rise of edge computing poses resource management issues, especially in decentralized systems where scalability and responsiveness are crucial. This paper introduces a cost-driven framework for collaborative resource management using the marginal computing cost per user. It applies the economic principle of marginal cost to assess edge data centers' (EDCs) ability to support more users, enabling efficient resource allocation. Simulations with PureEdgeSim and real-world data such as Alibaba Trace demonstrate substantial enhancements in resource use, latency, and active instance reduction, maintaining scalability and adaptability with high user demands.Source: FUTURE GENERATION COMPUTER SYSTEMS, vol. 176
DOI: 10.1016/j.future.2025.108167
Project(s): NOUS
, FutureHPC & BigData, OSMWARE
Metrics:
Carlini Emanuele, Dazzi Patrizio, Ferrucci Luca, Massa Jacopo, Mordacchini Matteo
The rise of edge computing poses resource management issues, especially in decentralized systems where scalability and responsiveness are crucial. This paper introduces a cost-driven framework for collaborative resource management using the marginal computing cost per user. It applies the economic principle of marginal cost to assess edge data centers' (EDCs) ability to support more users, enabling efficient resource allocation. Simulations with PureEdgeSim and real-world data such as Alibaba Trace demonstrate substantial enhancements in resource use, latency, and active instance reduction, maintaining scalability and adaptability with high user demands.Source: FUTURE GENERATION COMPUTER SYSTEMS, vol. 176
DOI: 10.1016/j.future.2025.108167
Project(s): NOUS
, FutureHPC & BigData, OSMWAREMetrics:
See at:
Future Generation Computer Systems | CNR IRIS | CNR IRIS | CNR IRIS | www.sciencedirect.com
2026
Journal article
Open Access
A time penalty for the Global South? Inequalities in visa appointment wait times at german embassies and consulates worldwide
Deutschmann Emanuel, Gabrielli Lorenzo, Orlova Alexandra, Harder Niklas, Recchi Ettore
Visas are a key tool for states to regulate incoming mobility from abroad, which can have ramifications for the establishment and perpetuation of global inequalities. In this article, we systematically analyze visa appointment wait times in German embassies and consulates worldwide. Using computational methods, we collect—and publish—fine-grained longitudinal data on the closest available appointment dates for various visa types, covering a total of 16,182 visa appointment requests. Our analysis reveals strong and systematic variance: the poorer the country a diplomatic mission is based in, the longer the wait time and the lower the chances of finding an available appointment (which ranges from almost 0 to 100 percent). We also argue that Germany's system is quite opaque compared to other established immigration countries such as the U.S. These core findings raise important questions in light of current debates about global justice, legal pathways to migration, and efforts to attract foreign talent.Source: POLITICAL GEOGRAPHY
DOI: 10.1016/j.polgeo.2025.103440
Metrics:
Deutschmann Emanuel, Gabrielli Lorenzo, Orlova Alexandra, Harder Niklas, Recchi Ettore
Visas are a key tool for states to regulate incoming mobility from abroad, which can have ramifications for the establishment and perpetuation of global inequalities. In this article, we systematically analyze visa appointment wait times in German embassies and consulates worldwide. Using computational methods, we collect—and publish—fine-grained longitudinal data on the closest available appointment dates for various visa types, covering a total of 16,182 visa appointment requests. Our analysis reveals strong and systematic variance: the poorer the country a diplomatic mission is based in, the longer the wait time and the lower the chances of finding an available appointment (which ranges from almost 0 to 100 percent). We also argue that Germany's system is quite opaque compared to other established immigration countries such as the U.S. These core findings raise important questions in light of current debates about global justice, legal pathways to migration, and efforts to attract foreign talent.Source: POLITICAL GEOGRAPHY
DOI: 10.1016/j.polgeo.2025.103440
Metrics:
See at:
Political Geography | Cadmus, EUI Research Repository | CNR IRIS | www.sciencedirect.com | CNR IRIS
2026
Other
Open Access
Quando il mare si illumina
Rapisarda Beatrice
La bioluminescenza è presente in tutti gli oceani del Pianeta, dalle acque illuminate dalla luce solare fino alle profondità più oscure degli abissi. Già nel Seicento i naturalisti restavano affascinati dal misterioso chiarore notturno del mare, ma solo nel corso del Novecento si è scoperto che quella luce incantata è il risultato di una precisa reazione chimica
Rapisarda Beatrice
La bioluminescenza è presente in tutti gli oceani del Pianeta, dalle acque illuminate dalla luce solare fino alle profondità più oscure degli abissi. Già nel Seicento i naturalisti restavano affascinati dal misterioso chiarore notturno del mare, ma solo nel corso del Novecento si è scoperto che quella luce incantata è il risultato di una precisa reazione chimica
See at:
almanacco.cnr.it | CNR IRIS | CNR IRIS
2026
Other
Open Access
A standardized framework for evaluating gene expression generative models
Rubbi Andrea, Di Francesco Andrea Giuseppe, Lotfollhai Mohammad, Lio Pietro
The rapid development of generative models for single-cell gene expression data has created an urgent need for standardised evaluation frameworks. Current evaluation practices suffer from inconsistent metric implementations, incomparable hyperparameter choices, and a lack of biologically-grounded metrics. We present Generated Genetic Expression Evaluator (GGE), an open-source Python framework that addresses these challenges by providing a comprehensive suite of distributional metrics with explicit computation space options and biologically-motivated evaluation through differentially expressed gene (DEG)-focused analysis and perturbation-effect correlation, enabling standardized reporting and reproducible benchmarking. Through extensive analysis of the single-cell generative modeling literature, we identify that no standardized evaluation protocol exists. Methods report incomparable metrics computed in different spaces with different hyperparameters. We demonstrate that metric values vary substantially depending on implementation choices, highlighting the critical need for standardization. GGE enables fair comparison across generative approaches and accelerates progress in perturbation response prediction, cellular identity modeling, and counterfactual inference.DOI: 10.48550/arxiv.2603.11244
Metrics:
Rubbi Andrea, Di Francesco Andrea Giuseppe, Lotfollhai Mohammad, Lio Pietro
The rapid development of generative models for single-cell gene expression data has created an urgent need for standardised evaluation frameworks. Current evaluation practices suffer from inconsistent metric implementations, incomparable hyperparameter choices, and a lack of biologically-grounded metrics. We present Generated Genetic Expression Evaluator (GGE), an open-source Python framework that addresses these challenges by providing a comprehensive suite of distributional metrics with explicit computation space options and biologically-motivated evaluation through differentially expressed gene (DEG)-focused analysis and perturbation-effect correlation, enabling standardized reporting and reproducible benchmarking. Through extensive analysis of the single-cell generative modeling literature, we identify that no standardized evaluation protocol exists. Methods report incomparable metrics computed in different spaces with different hyperparameters. We demonstrate that metric values vary substantially depending on implementation choices, highlighting the critical need for standardization. GGE enables fair comparison across generative approaches and accelerates progress in perturbation response prediction, cellular identity modeling, and counterfactual inference.DOI: 10.48550/arxiv.2603.11244
Metrics:
2026
Other
Open Access
Resilienza verde
Rapisarda Beatrice
Le piante vivono esposte a ferite, attacchi, stress ambientali e malattie per tutta la loro esistenza. Un ramo spezzato dal vento, una foglia mangiata da un insetto, una radice danneggiata da un fungo o dalla siccità: per una pianta, il danno è una condizione quotidiana. Proprio per questo, nel corso dell’evoluzione, hanno sviluppato una sorprendente capacità di riparazione, come ci racconta Alberto Santini dell’Istituto per la protezione sostenibile delle piante del Cnr
Rapisarda Beatrice
Le piante vivono esposte a ferite, attacchi, stress ambientali e malattie per tutta la loro esistenza. Un ramo spezzato dal vento, una foglia mangiata da un insetto, una radice danneggiata da un fungo o dalla siccità: per una pianta, il danno è una condizione quotidiana. Proprio per questo, nel corso dell’evoluzione, hanno sviluppato una sorprendente capacità di riparazione, come ci racconta Alberto Santini dell’Istituto per la protezione sostenibile delle piante del Cnr
See at:
almanacco.cnr.it | CNR IRIS | CNR IRIS
2026
Other
Open Access
Custodi della biodiversità
Rapisarda Beatrice
La varietà della vita sulla Terra è uno dei patrimoni più straordinari e, al tempo stesso, fragili che abbiamo. Le banche della biodiversità, come ci racconta Mario Sprovieri, direttore dell’Istituto di scienze marine del Cnr, sono una risposta concreta che scienziati, governi e organizzazioni internazionali hanno ideato per preservarla
Rapisarda Beatrice
La varietà della vita sulla Terra è uno dei patrimoni più straordinari e, al tempo stesso, fragili che abbiamo. Le banche della biodiversità, come ci racconta Mario Sprovieri, direttore dell’Istituto di scienze marine del Cnr, sono una risposta concreta che scienziati, governi e organizzazioni internazionali hanno ideato per preservarla
See at:
almanacco.cnr.it | CNR IRIS | CNR IRIS
2026
Conference article
Open Access
Forward index compression for learned sparse retrieval
Bruch Sebastian, Fontana Martino, Nardini Franco Maria, Rulli Cosimo, Venturini Rossano
Text retrieval using learned sparse representations of queries and documents has, over the years, evolved into a highly effective approach to search. It is thanks to recent advances in approximate nearest neighbor search—with the emergence of highly efficient algorithms such as the inverted index-based Seismic and the graph-based Hnsw—that retrieval with sparse representations became viable in practice. In this work, we scrutinize the efficiency of sparse retrieval algorithms and focus particularly on the size of a data structure that is common to all algorithmic flavors and that constitutes a substantial fraction of the overall index size: the forward index. In particular, we seek compression techniques to reduce the storage footprint of the forward index without compromising search quality or inner product computation latency. In our examination with various integer compression techniques, we report that StreamVByte achieves the best trade-off between memory footprint, retrieval accuracy, and latency. We then improve StreamVByte by introducing DotVByte, a new algorithm tailored to inner product computation. Experiments on MsMarco show that our improvements lead to significant space savings while maintaining retrieval efficiency.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16484, pp. 444-451. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21300-6_35
DOI: 10.48550/arxiv.2602.05445
Metrics:
Bruch Sebastian, Fontana Martino, Nardini Franco Maria, Rulli Cosimo, Venturini Rossano
Text retrieval using learned sparse representations of queries and documents has, over the years, evolved into a highly effective approach to search. It is thanks to recent advances in approximate nearest neighbor search—with the emergence of highly efficient algorithms such as the inverted index-based Seismic and the graph-based Hnsw—that retrieval with sparse representations became viable in practice. In this work, we scrutinize the efficiency of sparse retrieval algorithms and focus particularly on the size of a data structure that is common to all algorithmic flavors and that constitutes a substantial fraction of the overall index size: the forward index. In particular, we seek compression techniques to reduce the storage footprint of the forward index without compromising search quality or inner product computation latency. In our examination with various integer compression techniques, we report that StreamVByte achieves the best trade-off between memory footprint, retrieval accuracy, and latency. We then improve StreamVByte by introducing DotVByte, a new algorithm tailored to inner product computation. Experiments on MsMarco show that our improvements lead to significant space savings while maintaining retrieval efficiency.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16484, pp. 444-451. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21300-6_35
DOI: 10.48550/arxiv.2602.05445
Metrics:
See at:
arXiv.org e-Print Archive | CNR IRIS | link.springer.com | doi.org | doi.org | CNR IRIS | CNR IRIS
2026
Journal article
Open Access
Leveraging Topic Specificity and Social Relationships for Expert Finding in Community Question Answering Platforms
Amendola Maddalena, Passarella Andrea, Perego Raffaele
Online Community Question Answering (CQA) platforms have become indispensable tools for users seeking expert solutions to their technical queries. The effectiveness of these platforms relies on their ability to identify and direct questions to the most knowledgeable users within the community, a process known as Expert Finding (EF). EF accuracy is crucial for increasing user engagement and the reliability of provided answers. We present TUEF, a Topic-oriented User-Interaction model for EF, which aims to fully and transparently leverage the heterogeneous information available within online CQA platforms. TUEF integrates content and social data by constructing a multi-layer graph that maps user relationships based on their answering patterns on specific topics. By combining these sources of information, TUEF identifies the most relevant users for any given question and ranks them using learning-to-rank techniques. Our findings indicate that TUEF’s topic-oriented model significantly enhances performance, particularly in large communities discussing well-defined topics. Additionally, we show that the interpretable learning-to-rank algorithm integrated into TUEF offers transparency and explainability with minimal performance trade-offs. The exhaustive experiments conducted across six CQA communities show that TUEF outperforms all competitors, achieving a minimum performance boost of 42.42% in P@1, 32.73% in NDCG@3, 21.76% in R@5, and 29.81% in MRR.Source: ACM TRANSACTIONS ON INFORMATION SYSTEMS
DOI: 10.1145/3793531
Project(s): EFRA
Metrics:
Amendola Maddalena, Passarella Andrea, Perego Raffaele
Online Community Question Answering (CQA) platforms have become indispensable tools for users seeking expert solutions to their technical queries. The effectiveness of these platforms relies on their ability to identify and direct questions to the most knowledgeable users within the community, a process known as Expert Finding (EF). EF accuracy is crucial for increasing user engagement and the reliability of provided answers. We present TUEF, a Topic-oriented User-Interaction model for EF, which aims to fully and transparently leverage the heterogeneous information available within online CQA platforms. TUEF integrates content and social data by constructing a multi-layer graph that maps user relationships based on their answering patterns on specific topics. By combining these sources of information, TUEF identifies the most relevant users for any given question and ranks them using learning-to-rank techniques. Our findings indicate that TUEF’s topic-oriented model significantly enhances performance, particularly in large communities discussing well-defined topics. Additionally, we show that the interpretable learning-to-rank algorithm integrated into TUEF offers transparency and explainability with minimal performance trade-offs. The exhaustive experiments conducted across six CQA communities show that TUEF outperforms all competitors, achieving a minimum performance boost of 42.42% in P@1, 32.73% in NDCG@3, 21.76% in R@5, and 29.81% in MRR.Source: ACM TRANSACTIONS ON INFORMATION SYSTEMS
DOI: 10.1145/3793531
Project(s): EFRA
Metrics:
2026
Conference article
Restricted
Decentralized and self-adaptive core maintenance on temporal graphs
Rucci D., Carlini E., Dazzi P., Kavalionak H., Mordacchini M.
Key graph-based problems play a central role in understanding network topology and uncovering patterns of similarity in homogeneous and temporal data. Such patterns can be revealed by analyzing communities formed by nodes, which in turn can be effectively modeled through temporal k-cores. This paper introduces a novel decentralized and incremental algorithm for computing the core decomposition of temporal networks. Decentralized solutions leverage the ability of network nodes to communicate and coordinate locally, addressing complex problems in a scalable, adaptive, and timely manner. By leveraging previously computed coreness values, our approach significantly reduces the activation of nodes and the volume of message exchanges when the network changes over time. This enables scalability with only a minimal trade-off in precision. Experimental evaluations on large real-world networks under varying levels of dynamism demonstrate the efficiency of our solution compared to a state-of-the-art approach, particularly in terms of active nodes, communication overhead, and convergence speed.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16322, pp. 346-361. Niagara Falls, ON, Canada, 2025
DOI: 10.1007/978-3-032-13513-1_28
Metrics:
Rucci D., Carlini E., Dazzi P., Kavalionak H., Mordacchini M.
Key graph-based problems play a central role in understanding network topology and uncovering patterns of similarity in homogeneous and temporal data. Such patterns can be revealed by analyzing communities formed by nodes, which in turn can be effectively modeled through temporal k-cores. This paper introduces a novel decentralized and incremental algorithm for computing the core decomposition of temporal networks. Decentralized solutions leverage the ability of network nodes to communicate and coordinate locally, addressing complex problems in a scalable, adaptive, and timely manner. By leveraging previously computed coreness values, our approach significantly reduces the activation of nodes and the volume of message exchanges when the network changes over time. This enables scalability with only a minimal trade-off in precision. Experimental evaluations on large real-world networks under varying levels of dynamism demonstrate the efficiency of our solution compared to a state-of-the-art approach, particularly in terms of active nodes, communication overhead, and convergence speed.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16322, pp. 346-361. Niagara Falls, ON, Canada, 2025
DOI: 10.1007/978-3-032-13513-1_28
Metrics:
See at:
CNR IRIS | CNR IRIS | link.springer.com
2026
Conference article
Restricted
Evaluating the efficiency and effectiveness of learned sparse retrieval with the lsr_benchmark
Frobe Maik, Schlatt Ferdinand, Rulli Cosimo, Hagen Tim, Merker Jan Heinrich, Hendriksen Gijs, Lassance Carlos, Nardini Franco Maria, Venturini Rossano, Potthast Martin
Learned sparse retrieval (LSR) models exhibit varying trade-offs between effectiveness and efficiency. But while standard tools exist for evaluating LSR effectiveness, there is none for evaluating efficiency. Also, datasets with high-quality relevance judgments are too large for repeated efficiency experiments, e.g., on different hardware configurations. To promote the evaluation of LSR models in terms of their effectiveness and efficiency, we introduce the lsr_benchmark, which measures retrieval efficiency at each step of an LSR pipeline (document embedding, indexing, query embedding, and retrieval) as well as its overall effectiveness. To ensure tractability and extensibility, we apply current corpus subsampling methods to eleven TREC tasks, precompute embeddings with eleven LSR models per task, and evaluate eight retrieval engines as baselines. For the benchmark’s hosted version, a modular API, along with tools for evaluating effectiveness and efficiency, facilitates the submission of new approaches. Our experiments show that the chosen embedding model significantly affects the efficiency of a retrieval engine and that LSR is more effective but less efficient than BM25—an efficiency gap that our benchmark now tracks as new LSR models are published.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16486, pp. 528-543. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21321-1_57
Metrics:
Frobe Maik, Schlatt Ferdinand, Rulli Cosimo, Hagen Tim, Merker Jan Heinrich, Hendriksen Gijs, Lassance Carlos, Nardini Franco Maria, Venturini Rossano, Potthast Martin
Learned sparse retrieval (LSR) models exhibit varying trade-offs between effectiveness and efficiency. But while standard tools exist for evaluating LSR effectiveness, there is none for evaluating efficiency. Also, datasets with high-quality relevance judgments are too large for repeated efficiency experiments, e.g., on different hardware configurations. To promote the evaluation of LSR models in terms of their effectiveness and efficiency, we introduce the lsr_benchmark, which measures retrieval efficiency at each step of an LSR pipeline (document embedding, indexing, query embedding, and retrieval) as well as its overall effectiveness. To ensure tractability and extensibility, we apply current corpus subsampling methods to eleven TREC tasks, precompute embeddings with eleven LSR models per task, and evaluate eight retrieval engines as baselines. For the benchmark’s hosted version, a modular API, along with tools for evaluating effectiveness and efficiency, facilitates the submission of new approaches. Our experiments show that the chosen embedding model significantly affects the efficiency of a retrieval engine and that LSR is more effective but less efficient than BM25—an efficiency gap that our benchmark now tracks as new LSR models are published.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16486, pp. 528-543. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21321-1_57
Metrics:
See at:
doi.org | CNR IRIS | CNR IRIS | link.springer.com
2026
Conference article
Open Access
Neural lexical search with learned sparse retrieval
Yates Andrew, Lassance Carlos, Rulli Cosimo, Yang Eugene, Macavaney Sean, Singh Siddharth A. K., Nguyen Thong, Lei Yibin
Learned Sparse Retrieval (LSR) techniques use neural machinery to represent queries and documents as learned bags of words. In contrast with other neural retrieval techniques, such as generative retrieval and dense retrieval, LSR has been shown to be a remarkably robust, transferable, and efficient family of methods for retrieving high-quality search results. This half-day tutorial aims to provide an extensive overview of LSR, ranging from its fundamentals to the latest emerging techniques. By the end of the tutorial, attendees will be familiar with the important design decisions of an LSR model, know how to apply them to text and other modalities, and understand the latest techniques for retrieving with them efficiently. Website: https://lsr-tutorial.github.io.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16486, pp. 35-43. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21321-1_5
DOI: 10.1145/3726302.3731693
Project(s): HEFPA
Metrics:
Yates Andrew, Lassance Carlos, Rulli Cosimo, Yang Eugene, Macavaney Sean, Singh Siddharth A. K., Nguyen Thong, Lei Yibin
Learned Sparse Retrieval (LSR) techniques use neural machinery to represent queries and documents as learned bags of words. In contrast with other neural retrieval techniques, such as generative retrieval and dense retrieval, LSR has been shown to be a remarkably robust, transferable, and efficient family of methods for retrieving high-quality search results. This half-day tutorial aims to provide an extensive overview of LSR, ranging from its fundamentals to the latest emerging techniques. By the end of the tutorial, attendees will be familiar with the important design decisions of an LSR model, know how to apply them to text and other modalities, and understand the latest techniques for retrieving with them efficiently. Website: https://lsr-tutorial.github.io.Source: LECTURE NOTES IN COMPUTER SCIENCE, vol. 16486, pp. 35-43. Delft, The Netherlands, 29/03-02/04/2026
DOI: 10.1007/978-3-032-21321-1_5
DOI: 10.1145/3726302.3731693
Project(s): HEFPA
Metrics:
See at:
IRIS Cnr | IRIS Cnr | Universiteit van Amsterdam (UvA) Institutional Repository UvA-DARE | Universiteit van Amsterdam (UvA) Institutional Repository UvA-DARE | IRIS Cnr | doi.org | DBLP | CNR IRIS | CNR IRIS | CNR IRIS | link.springer.com
2026
Journal article
Open Access
Decentralized edge learning: a comparative study of distillation strategies and dissimilarity measures
Molo Mbasa J., Vadicamo Lucia, Gennaro Claudio, Carlini Emanuele
Decentralized learning is emerging as a scalable and privacy-preserving alternative to centralized machine learning, particularly in distributed systems where data cannot be centrally shared among multiple nodes or clients. While Federated Learning is widely adopted in this context, Knowledge Distillation (KD) is emerging as a flexible and scalable alternative where model output is used to share knowledge among distributed clients. However, existing studies often overlook the efficiency and effectiveness of various knowledge transfer strategies in KD, especially in decentralized environments where data is non-IID. This study provides key insights by examining the impact of network topology and distillation strategies in KD-based decentralized learning approaches. Our evaluation spans several dissimilarity measures, including Cross-Entropy, Kullback-Leibler divergence, Triangular Divergence, Jensen-Shannon divergence, Structural Entropic Distance, and Multi-way SED, assessed under both pairwise and holistic distillation schemes. In the pairwise approach, distillation is performed by summing the client-wise dissimilarities between a client's output and each neighbor's prediction individually, while the holistic approach computes dissimilarity with respect to the average of the output predictions received from neighboring clients. We also analyze performance across client connectivity levels to explore the trade-off between convergence speed and model accuracy. The results indicate that the holistic distillation approach, which averages client predictions, outperforms the sum of pairwise distillation, especially when employing alternative measures like TD, SED, and JS. These measures offer improved performance over conventional metrics such as CE and KL divergence.Source: FUTURE GENERATION COMPUTER SYSTEMS, vol. 176
DOI: 10.1016/j.future.2025.108171
Project(s): National Centre for HPC, Big Data and Quantum Computing, Sustainable Mobility Center
Metrics:
Molo Mbasa J., Vadicamo Lucia, Gennaro Claudio, Carlini Emanuele
Decentralized learning is emerging as a scalable and privacy-preserving alternative to centralized machine learning, particularly in distributed systems where data cannot be centrally shared among multiple nodes or clients. While Federated Learning is widely adopted in this context, Knowledge Distillation (KD) is emerging as a flexible and scalable alternative where model output is used to share knowledge among distributed clients. However, existing studies often overlook the efficiency and effectiveness of various knowledge transfer strategies in KD, especially in decentralized environments where data is non-IID. This study provides key insights by examining the impact of network topology and distillation strategies in KD-based decentralized learning approaches. Our evaluation spans several dissimilarity measures, including Cross-Entropy, Kullback-Leibler divergence, Triangular Divergence, Jensen-Shannon divergence, Structural Entropic Distance, and Multi-way SED, assessed under both pairwise and holistic distillation schemes. In the pairwise approach, distillation is performed by summing the client-wise dissimilarities between a client's output and each neighbor's prediction individually, while the holistic approach computes dissimilarity with respect to the average of the output predictions received from neighboring clients. We also analyze performance across client connectivity levels to explore the trade-off between convergence speed and model accuracy. The results indicate that the holistic distillation approach, which averages client predictions, outperforms the sum of pairwise distillation, especially when employing alternative measures like TD, SED, and JS. These measures offer improved performance over conventional metrics such as CE and KL divergence.Source: FUTURE GENERATION COMPUTER SYSTEMS, vol. 176
DOI: 10.1016/j.future.2025.108171
Project(s): National Centre for HPC, Big Data and Quantum Computing, Sustainable Mobility Center
Metrics:
See at:
CNR IRIS | www.sciencedirect.com | Future Generation Computer Systems | CNR IRIS | CNR IRIS
2026
Journal article
Open Access
ECLYPSE: a Python framework for simulation and emulation of the cloud-edge continuum
Massa Jacopo, Decaro Valerio, Forti Stefano, Dazzi Patrizio, Bacciu Davide, Brogi Antonio
The Cloud-Edge continuum enhances application performance by bringing computation closer to data sources. However, it presents considerable challenges in managing resources and determining application service placement, as these tasks require analyzing diverse, dynamic environments characterized by fluctuating network conditions. Addressing these challenges calls for tools combining simulation and emulation of Cloud-Edge systems to rigorously assess novel application and resource management strategies. In this paper, we introduce ECLYPSE, a Python-based framework that enables the simulation and emulation of the Cloud-Edge continuum via adaptable resource allocation and service placement models. ECLYPSE features an event-driven architecture for dynamically adapting network configurations and resources. It also supports seamless transitions between simulated and emulated setups, thus enabling the execution of experiments in simulated, emulated, and hybrid settings. In this work, we illustrate and assess ECLYPSE capabilities over three use cases, demonstrating the framework's effectiveness in rapid prototyping across diverse scenarios.Source: JOURNAL OF SOFTWARE, vol. 38 (issue 1)
DOI: 10.1002/smr.70081
Metrics:
Massa Jacopo, Decaro Valerio, Forti Stefano, Dazzi Patrizio, Bacciu Davide, Brogi Antonio
The Cloud-Edge continuum enhances application performance by bringing computation closer to data sources. However, it presents considerable challenges in managing resources and determining application service placement, as these tasks require analyzing diverse, dynamic environments characterized by fluctuating network conditions. Addressing these challenges calls for tools combining simulation and emulation of Cloud-Edge systems to rigorously assess novel application and resource management strategies. In this paper, we introduce ECLYPSE, a Python-based framework that enables the simulation and emulation of the Cloud-Edge continuum via adaptable resource allocation and service placement models. ECLYPSE features an event-driven architecture for dynamically adapting network configurations and resources. It also supports seamless transitions between simulated and emulated setups, thus enabling the execution of experiments in simulated, emulated, and hybrid settings. In this work, we illustrate and assess ECLYPSE capabilities over three use cases, demonstrating the framework's effectiveness in rapid prototyping across diverse scenarios.Source: JOURNAL OF SOFTWARE, vol. 38 (issue 1)
DOI: 10.1002/smr.70081
Metrics:
See at:
CNR IRIS | onlinelibrary.wiley.com | CNR IRIS
2025
Other
Open Access
Viaggio fantastico dal Dna alle proteine
Rapisarda B.
Come avvicinare il pubblico alla conoscenza scientifica e rendere accessibili, e persino appassionanti, concetti complessi come i meccanismi alla base della vita? È la domanda da cui sono partiti i ricercatori dell’Istituto di farmacologia traslazionale (Ift) e dell’Istituto di fotonica e nanotecnologie del Cnr, che hanno scelto la via della gamification per raccontare in modo coinvolgente i principi della biologia.
Rapisarda B.
Come avvicinare il pubblico alla conoscenza scientifica e rendere accessibili, e persino appassionanti, concetti complessi come i meccanismi alla base della vita? È la domanda da cui sono partiti i ricercatori dell’Istituto di farmacologia traslazionale (Ift) e dell’Istituto di fotonica e nanotecnologie del Cnr, che hanno scelto la via della gamification per raccontare in modo coinvolgente i principi della biologia.
See at:
almanacco.cnr.it | CNR IRIS | CNR IRIS