Alessandro Chessa

https://lnkd.in/dvrt3AaT via @researchgate https://lnkd.in/dvrt3AaT Securing and validating AI-ML systems are major concerns for the industry and the society. Our paper on the state of the art reached 2500 reads in a few weeks.. check it out for the notion and a survey of the #security4ai issues Chan Yeob Yeun

19

To stay sane, a computer scientist must occasionally write something unrelated to AI. With Iacopo Colonnelli, Doriana Medić, Alberto Mulone, Viviana Bono, Luca Padovani, Viviana Bono (Marco Aldinucci), we designed a novel intermediate representation language for distributed scientific workflows named SWIRL. Being designed for compilers, SWIRL would be considered a very boring language by a human (by an AI ... who knows, they refuse to make moral judgments; otherwise, they get suited), but SWIRL has its own nice features; it has formal semantics that is consistent according with Church-Rosser property, which is, in this case, mean that the semantics (effect of distributed execution) is not sensible to the execution order of independent steps of the workflow. This is the groundwork for defining a canonical distributed execution and also an effective way to design static workflow execution optimizations. And this is really new. To our knowledge, most (if not all) scientific workflow management systems (WMS) are interpreted, not compiled, and therefore, they hardly support automatic optimization. Once you succeed in evaluating if the result is the same, you can play with extra-functional features: performance (for #hpc lovers), data locality and privacy (for #multicloud lovers), resource optimization (for #energy lovers), and much more. You can also build your HPC federation protocols, hoping to not get random numbers as a result. The paper itself will appear at Formal Methods 2024 (13th Sept 2024 https://lnkd.in/dwaez7PU ), but it is already available as TR on arxiv https://lnkd.in/dY777upr This work has been supported by ICSC - Centro Nazionale di Ricerca in HPC, Big Data e Quantum Computing EUPEX Pilot EuroHPC Joint Undertaking (EuroHPC JU) HPC4AI Dipartimento di Informatica - Università degli Studi di Torino Università degli Studi di Torino CINECA #hpc #highperformancecomputing #ai #cloud #formalmethod

39

1 commento

😊 Very happy to announce our new publication in the #UMUAI journal, together with Giuseppe Spillo, Pasquale Lops, Marco de Gemmis, Gianni Semeraro 📄 "Recommender systems based on neuro-symbolic knowledge graph embeddings encoding first-order logic rules" 👩🏻‍💻It was a quite long process, with many review iterations and several constructive comments, but at the end we made it :-) ✍🏻💡 In this work, we have proposed a new #knowledge-aware #recsys that exploits a #neuro-#symbolic method to learn graph embeddings. We showed that the introduction of #logical rules in the recommendation process increases the accuracy of the model w.r.t. several state-of-the-art approach Paper available here:

59

1 commento

I am very happy to report that my paper (with #CaterinaDAssergio, Elisa Montaguti & Sara Valentini) on firm usage of EU's GDPR opt-in emails is now forthcoming at Journal of Marketing (link to paper below)! We would like to thank the Editor (Shrihari (Hari) Sridhar) and the rest of the review team for a very thoughtful and constructive review process. GDPR mandates explicit user opt-in consent for data access. This paper uses theory, multiple datasets and multiple methods to investigate firms’ communication formats for opt-in requests, addressing three questions: 1) how do firms design their opt-in requests? 2) does the chosen format affect consumer response? 3) what drives firms' choices of formats? The analysis of 1,506 re-permission emails from 1,396 firms post-GDPR shows that 26% use only persuasive cues (e.g., monetary incentives) to improve opt-in, while 24% blend persuasive and informative cues. A field experiment rationalizes this behavior showing that a mix of persuasive and informative cues was the most successful at generating opt-in. Firms that depend more on personal data utilize persuasive cues more often than firms concerned with the reputational risk of GDPR non-compliance. Overall, a significant proportion of firms technically comply with GDPR but fail to capture its underlying intent. This paper offers valuable insights for regulators and consumers, highlighting the diversity in how firms obtain consent and the subsequent impact on user behavior. #Privacy #GDPR #EmailOptin #MachineLearning #FieldExperiments #CrossIndustryAnalysis https://lnkd.in/gh5cTrMU

212

27 commenti

Temporal graphs can evolve and be observed at very irregular time instants. Our recently accepted #IJCAI2024 paper shows how to effectively handle learning on this class of graph streams. See the original post by Alessio Gravina below, who is the lead author of the work together with Daniele Zambon. Joint work by University of Pisa and IDSIA-USI. https://lnkd.in/dPKcYH48

135

🚀 New Open Access Article 🚀 Emanuele Cavenaghi published his work on TORS! 👇 His paper, "Towards a Causal Decision-Making Framework for Recommender Systems", addresses the limitations of learning from offline observed data and the challenges of current debiasing strategies. To advance the field, he proposes integrating causality frameworks such as potential outcomes and structural causal models into recommender systems. This approach, developed in collaboration with Free University of Bozen-Bolzano, aims to enhance future research by applying causal discovery strategies on offline data, enabling the computation of counterfactuals and improving debiasing methods. You can find the full paper at: https://lnkd.in/dVgcJbMq Congratulations to Emanuele and his team for this amazing contribution! Dive into our research to explore these innovative ideas and their potential to transform recommender systems at: https://lnkd.in/dtpqzmX4 #Research #ArtificialIntelligence #MachineLearning #RecommenderSystems #Causality #MADLab

9

Call for Papers - #ECCV2024 Workshop: At the second CVF/ECCV More Exploration, Less Exploitation (MELEX) workshop (in conjunction with ECCV 2024 in Milan), we will publish papers in all areas of computer vision that propose innovative models, algorithms, and ideas that show competitive results, but not necessarily the SOTA results in benchmarks. We call for unpublished papers that have a history of being rejected by the main conferences because they did not achieve SOTA results. We are looking for unpublished papers that have already received citations on arXiv. More information at: https://lnkd.in/dBCHuReb Organisation: Vasileios Belagiannis Azade Farshad Angela Yao Gustavo Carneiro. #MELEX #ECCV2024 #ECCV #ComputerVision #DeepLearning 

40

🎉We are thrilled to announce that we are presenting four articles at the #ACL2024 NLP conference this week! Check out the list of accepted papers here: https://lnkd.in/dDG5qX5q Joint work with Sapienza NLP and other research groups! #NLP #NLProc #AI #ComputationalLinguistics #MachineLearning

25

Paper on statistical analysis of the history of the Republic of Venice officially published https://lnkd.in/drFs_Ph9 Check it out if you're insterested in Venice, #rstats or just out of general curiosity.

1

I just presented our research at ACM RecSys where we examined how the topology of user-item graphs affects the performance of #GNN-based recommender systems, including models like LightGCN, DGCF, and SVD-GCN. Using datasets such as Yelp, Gowalla, and Amazon-Book, we uncovered #topologicalcharacteristics that can significantly enhance recommendation #accuracy. This new evaluation perspective could lead to important innovations in how recommender systems are developed and applied. #RecommenderSystems #GraphCF #DatasetTopology #recsys2024

78

2 commenti

> sharing resources < Great paper today: "A Study on Quantum Graph Neural Networks Applied to Molecular Physics" by  Simone Piperno, Andrea Ceschini, Su Yeon Chang, Michele Grossi, Sofia Vallecorsa, Massimo Panella Abstract: This paper introduces a novel architecture for Quantum Graph Neural Networks, which is significantly different from previous approaches found in the literature. The proposed approach produces similar outcomes with respect to previous models but with fewer parameters, resulting in an extremely interpretable architecture rooted in the underlying physics of the problem. The architectural novelties arise from three pivotal aspects. Firstly, we employ an embedding updating method that is analogous to classical Graph Neural Networks, therefore bridging the classical-quantum gap. Secondly, each layer is devoted to capturing interactions of distinct orders, aligning with the physical properties of the system. Lastly, we harness SWAP gates to emulate the problem's inherent symmetry, a novel strategy not found currently in the literature. The obtained results in the considered experiments are encouraging to lay the foundation for continued research in this field. Link: https://lnkd.in/eW3BCvEP #quantummachinelearning #quantumcomputing #HEP

48

2 commenti

🌟 Excited about human-inspired AI? Submit your papers to our HCV Workshop at #ECCV2024 in Milan! Dive into the future of computer vision, inspired by human cognition. 📅 Paper deadline: July 19th Details: https://lnkd.in/gtiHzeG5 #ComputerVision #AI #Neuroscience #CognitiveScience

29

[EDIT] Deadline extended to Jul 17th! See you in Milan! We are organizing a workshop on Uncertainty Quantification for Computer Vision at #ECCV2024 #computervision #deeplearning with Martin Trapp Arno Solin Rui Li Gianni Franchi Andrei Bursuc Neill Campbell Ivor Simpson Angela Yao Marcus Klasson Gaëtan Brison Detailed information: https://lnkd.in/ddcETMKa Topics include, but not limited to • Applications of uncertainty quantification • Failure prediction (e.g., OOD detection) • Robustness in CV • Safety critical applications in CV • Domain-shift in CV • Probabilistic deep models • Deep probabilistic models • Deep ensemble uncertainty • Connections between NNs and GPs • Incorporating explicit prior knowledge in deep learning • Computational aspects and real-time probabilistic inference • Output ambiguity, multi-modality and diversity

124

1 commento

Happy to share our latest publication in the field of statistical learning with Donato Riccio and Elvira Romano, reaching one of the top journals I’ve been aiming for! It’s a great reward for all the passion and deep focus we’ve dedicated to the fascinating topic of supervised classification over the years. Check out the paper: https://lnkd.in/ddqZA3-e Free access to the paper: rdcu.be/dVrbL #StatisticalLearning #MachineLearning #FunctionalDataAnalysis #EnsembleLearning #HighDimensionalData #SupervisedClassification #PredictiveModeling #DataScience #Research #FunctionalVotingClassifier

27

We'll present our paper at the IEEE International Conference on Image Processing on Trustworthy Visual Data Processing! It is titled "Interpreting the Fraudulence Level of Different Finger Photo Presentation Attack Instruments" by Anudeep Vurity, Emanuela Marasco, Raghavendra Ramachandra, and Duoduo Liao. See you in Abu Dhabi! hashtag #AI hashtag #ICIP2024 hashtag #ExplainableAI

23

great job, Erica Coppolillo, Simona Perri, Francesco Ricca, Giuseppe Manco! https://lnkd.in/dtdweMMN === 🎉 Great News! 🎉 We are thrilled to announce that our paper "LLASP: Fine-tuning Large Language Models for Answer Set Programming" has been accepted at KR2024! In the work, we perform a systematic evaluation of several state-of-the-art LLMs, showing inadequate performances in generating correct programs in Answer Set Programming (ASP). Therefore, we propose LLASP, a lightweight model fine-tuned to encode fundamental ASP program patterns. To this aim, we create and release an ad-hoc dataset covering a wide variety of fundamental problem specifications that can be encoded in ASP. Thank you to Francesco Calimeri, Giuseppe Manco, Simona Perri, and Francesco Ricca, who supervised me on the paper. Stay tuned for more updates! 🚀 #KR2024 #AI #MachineLearning #AnswerSetProgramming #LLMs #logic #KRR #KR #NMR #ArtificialIntelligence #ComputerScience

27

📚 Excited to share our latest article on the methodological framework and tools used in the #PromptEngineering course at Fondazione Bruno Kessler - FBK UniTrento DIPSCO In this course, we empower students to effectively interact with #generativeAI models. Key highlights include: 🔹 The "Card Model": A structured approach to defining and organizing generative AI tasks. 🔹 The "Flow" concept: This structures the relationships between tasks, facilitating the creation of complex workflows. Students have been using the #Azure AI Proxy Playground, an open-source GUI, to interact with OpenAI models like GPT-3.5 Turbo. This allows them to apply their theoretical knowledge in practical scenarios without needing extensive technical skills. Stay tuned for the second part of our blog series, where we'll dive deeper into these teaching methods for prompt engineering! #AI #GenerativeAI #PromptEngineering #Education #Innovation #TechSkills Nadia Mana

8