Stefano Fiorucci
San Giustino, Umbria, Italia
11.326 follower
Oltre 500 collegamenti
Informazioni
📬 stefanofiorucci (at) gmail (dot) com
💫 Structural Engineer turned Software Engineer with a passion for exploring the realms of Machine Learning and Natural Language Processing.
🔍 Formerly at 01S for 5 years, I specialized in information extraction and retrieval from unstructured documents, making valuable information accessible to Italian citizens.
💙 Currently, I am proud to be part of deepset, contributing to Haystack, the open-source LLM Framework. I enjoy engaging with a vibrant community of users and contributors.
---
Languages: Python, Java
Machine Learning: Pandas, scikit-learn, H2O, fundamentals of Keras, TensorFlow and PyTorch, Ludwig, streamlit
NLP: NLTK, fastText, Hugging Face Transformers/TRL/PEFT, Haystack, BERTopic, Argilla
Information extraction/retrieval: Scrapy, Tika, Tesseract, Camelot, Solr
API: Flask, FastAPI
Bash, Docker
Attività
-
Announcing Our Series A https://sakana.ai/series-a
Announcing Our Series A https://sakana.ai/series-a
Consigliato da Stefano Fiorucci
-
Improving Language models (LM) across multiple capabilities is often a non-trivial task, specially with limited compute availability. This blog post…
Improving Language models (LM) across multiple capabilities is often a non-trivial task, specially with limited compute availability. This blog post…
Consigliato da Stefano Fiorucci
-
Today is my first day at LaunchDarkly! Excited to team up with Matthew Makai, Erin Mikail 🏄🏼♀️ Staples and the rest of the incredible crew over…
Today is my first day at LaunchDarkly! Excited to team up with Matthew Makai, Erin Mikail 🏄🏼♀️ Staples and the rest of the incredible crew over…
Consigliato da Stefano Fiorucci
Esperienza
Formazione
-
-
Received a two-months full scholarship to participate at the School of Artificial Intelligence of Pi School. Selected among some of the brightest Engineers in the field, as a scholarship winner.
I worked on a project presented by a real client, developing a suite of NLP and information extraction tools for the healthcare domain. -
-
The school is organized by the AI Education Foundation, a charity founded by some Google DeepMind researchers.
The goal is to provide the latest information on the state of the art in the field of Machine Learning and Artificial Intelligence, bringing together MSc students, early-stage PhDs, industry professionals, and world-class researchers. -
-
Università degli Studi di Perugia, Edison Data Science Framework
▪ Statistics
▪ Python programming; basics of R programming
▪ Machine learning (Simone Scardapane)
▪ Neural networks and Deep learning (Elisa Ricci)
▪ Big data tools (Todor Ivanov): Apache Hadoop (HDFS, MapReduce, Sqoop,
Flume, Hive, Impala), Apache Spark (RDD, Spark SQL, Spark Streaming,
Dataframes, MLlib)
▪ Network and data security; Cloud computing; Visualization Analytics -
Thesis: Equivalent beam models for the dynamic analysis of tall buildings: estimation of modes through methods based on sub-structures and applications to dynamic analysis in the time domain (Supervisor: Prof. Federico Cluni)
Main subjects: computational mechanics; dynamics of structures and anti-seismic design; design of prestressed concrete structures; bridge design; design of wooden and glass structures; structural rehabilitation; foundations
Licenze e certificazioni
-
-
-
-
-
Ecdl Specialised IT Security
AICA - Associazione Italiana per l'Informatica ed il Calcolo Automatico
Emissione: -
Abilitazione all'esercizio della professione di Ingegnere civile sez. A, conseguita con Esame di Stato
Università degli Studi di Perugia
Emissione: -
DELF scolaire niveau 1 (A2)
République Française - Ministère de l'éducation nationale - Commission Nationale du DELF et du DALF
Emissione: -
-
Esperienze di volontariato
-
Volontario
Operazione Mato Grosso
- Presente 23 anni e 4 mesi
Alleviamento povertà
Free work to support humanitarian missions in Peru, Ecuador, Brazil and Bolivia.
6 months experience as a volunteer in Ecuador in 2008.
Pubblicazioni
-
Estimation of the Mechanical Parameters for a Reduced Coupled Flexural–Torsional Beam Model of a Tall Building by a Sub-Structure Approach
Applied Sciences. 2021; 11(10):4655.
The use of equivalent beam models to estimate the dynamical characteristics of complex tall buildings has been investigated by several authors. The main reason is the structural response estimation to stochastic loads, such as wind and earthquake, using a reduced number of degrees of freedom, which reduces the computational costs and therefore gives the designer an effective tool to explore a number of possible structural solutions. In this paper, a novel approach to calibrate the mechanical…
The use of equivalent beam models to estimate the dynamical characteristics of complex tall buildings has been investigated by several authors. The main reason is the structural response estimation to stochastic loads, such as wind and earthquake, using a reduced number of degrees of freedom, which reduces the computational costs and therefore gives the designer an effective tool to explore a number of possible structural solutions. In this paper, a novel approach to calibrate the mechanical and dynamical features of a complete 3D Timoshenko beam, i.e., describing bending, shear and torsional behavior, is proposed. This approach is based on explicitly considering the sub-structures of the tall building. In particular, the frames, shear walls and lattice sub-systems are modeled as equivalent beams, constrained by means of rigid diaphragms at different floors. The overall dynamic features of the tall building are obtained by equating the deformation energy of an equivalent sandwich beam with that of the selected sub-structures. Finally, the 3D Timoshenko equivalent beam parameters are calibrated by minimizing a suitable function of modal natural frequencies and static displacements. The closed form modal solution of the equivalent beam model is used to obtain the response to stochastic loads.
Altri autoriVedi pubblicazione -
SNK @ DANKMEMES: Leveraging Pretrained Embeddings for Multimodal Meme Detection
Proceedings of the Seventh Evaluation Campaign of Natural Language Processing and Speech Tools for Italian. Final Workshop (EVALITA 2020)
In this paper, we describe and present the results of meme detection system, specifically developed and submitted for our participation to the first subtask of DANKMEMES (EVALITA 2020).
We built simple classifiers, consisting in feed forward neural networks.
They leverage existing pretrained embeddings, both for text and image representation.
Our best system (SNK1) achieves good results in meme detection (F1 = 0.8473), ranking 2nd in the competition, at a distance of 0.0028 from the…In this paper, we describe and present the results of meme detection system, specifically developed and submitted for our participation to the first subtask of DANKMEMES (EVALITA 2020).
We built simple classifiers, consisting in feed forward neural networks.
They leverage existing pretrained embeddings, both for text and image representation.
Our best system (SNK1) achieves good results in meme detection (F1 = 0.8473), ranking 2nd in the competition, at a distance of 0.0028 from the first classified. -
An approach based on substructures for the estimation of the response of tall buildings under wind loads using an equivalent beam
XIV conference of the Italian association for wind engineering (IN-VENTO 2016), Terni, 25-28 settembre 2016
The overall dynamic behaviour of tall buildings can be estimated by means of equivalent beam models, and several ways of assessing this equivalence are presented in papers by different authors.
The most appealing benefit of this approach is that the response of the structure to stochastic
actions, such as wind loads, can be estimated using a model with a reduced number of degrees of
freedom, consequently decreasing the computational costs and therefore allowing the designer…The overall dynamic behaviour of tall buildings can be estimated by means of equivalent beam models, and several ways of assessing this equivalence are presented in papers by different authors.
The most appealing benefit of this approach is that the response of the structure to stochastic
actions, such as wind loads, can be estimated using a model with a reduced number of degrees of
freedom, consequently decreasing the computational costs and therefore allowing the designer to
explore a greater variety of possible structural solutions (Cluni et al, 2013).
In this paper, the dynamic mechanical properties of an equivalent beam able to describe the threedimensional response of a tall building are estimated with an approach which extends that proposed in Cluni et al. (2014). The resulting model can fully describe the interaction between bending, shear and torsional behaviour.
In particular, the approach here proposed explicitly uses the information on the sub-structures of the building to calibrate the mechanical characteristics of the equivalent beam. More in details, the frames, the shear walls and the lattice structures are identified and modelled as equivalent beams connected by rigid diaphragms at different floors. Thereafter, using an energetic approach the dynamic characteristics of the tall building are estimated (as shown in Potzta and Kollar, 2003) and used to calibrate the mechanic characteristics of the equivalent beam by means of the minimization of a suitable function of modal periods and static displacements. The modal shapes of the equivalent beam can be evaluated in closed form, and thereafter the response of the structure to wind actions is estimated with modal analysis and/or stochastic analysis.
The results obtained with the proposed approach are finally compared both to standard Finite Element approach and to the modal approach with approximated modal shapes presented in Potzta and Kollar (2003).Altri autori
Riconoscimenti e premi
-
Randstad AI challenge - VGEN (1st place)
-
The aim of the competition was to train an ML model that would divide IT job offers into specific categories. I did a thorough study of the dataset, which allowed me to effectively clean up the original data. The model that has obtained the best performances is given by the simple combination of TF-IDF and a linear SVM, trained with stochastic gradient descent. I ranked 1st!
Topics: Text classification; Naive Bayes; SVM; TF-IDF
Libraries: scikit-learn -
Multimodal meme detection (2nd place)
-
As part of the DANKMEMES competition (EVALITA 2020), I developed a model for multimodal detection of political memes. The proposed solution is a feed-forward neural network that leverages pretrained embeddings, both for text and image representation. I ranked 2nd, at a small distance from the winner (Tor Vergata University team).
Topics: Multimodal classification; Word embeddings, Transformers, Image embeddings
Libraries: Ludwig, fastText, Huggingface’s Transformers
Lingue
-
Italiano
Conoscenza madrelingua o bilingue
-
Inglese
Conoscenza professionale
-
Spagnolo
Conoscenza lavorativa limitata
-
Francese
Conoscenza base
Altre attività di Stefano
-
Who never had problems with text encoding in pyton ? Ever wondered why two identical-looking strings sometimes don't match in your code? 🤔 The…
Who never had problems with text encoding in pyton ? Ever wondered why two identical-looking strings sometimes don't match in your code? 🤔 The…
Consigliato da Stefano Fiorucci
-
📈 Boost Your IE Performance with Our New Larger Reader Models! We’re thrilled to introduce our latest, larger Reader models for ReLiK that bring a…
📈 Boost Your IE Performance with Our New Larger Reader Models! We’re thrilled to introduce our latest, larger Reader models for ReLiK that bring a…
Consigliato da Stefano Fiorucci
-
🚀 Just completed an incredible short course from DeepLearning.AI on building LLM applications with Haystack! 🧠💻 A heartfelt thank you to Tuana…
🚀 Just completed an incredible short course from DeepLearning.AI on building LLM applications with Haystack! 🧠💻 A heartfelt thank you to Tuana…
Consigliato da Stefano Fiorucci
-
Continuing on from last week's article on Retrieval Augmented Generation (RAG for short), Bruce Nielson continues to add to his series of AI and ML…
Continuing on from last week's article on Retrieval Augmented Generation (RAG for short), Bruce Nielson continues to add to his series of AI and ML…
Consigliato da Stefano Fiorucci
-
Combining Long Context and Retrieval Paradigms ➿ The ability to comprehend and generate coherent text over extended passages is crucial for…
Combining Long Context and Retrieval Paradigms ➿ The ability to comprehend and generate coherent text over extended passages is crucial for…
Consigliato da Stefano Fiorucci
-
🎁 Just received this awesome gift commemorating the original Gemma launch! It even includes a pendrive with the original model weights – talk…
🎁 Just received this awesome gift commemorating the original Gemma launch! It even includes a pendrive with the original model weights – talk…
Consigliato da Stefano Fiorucci
-
Wouldn't it be great to generate embeddings directly from a dataset on the @huggingface Hub? This would enable us to find similar datasets or search…
Wouldn't it be great to generate embeddings directly from a dataset on the @huggingface Hub? This would enable us to find similar datasets or search…
Consigliato da Stefano Fiorucci
-
90% of all use cases in GenAI revolve around search (helping users find the right information, quickly). Knowledge-based chat, call transcript…
90% of all use cases in GenAI revolve around search (helping users find the right information, quickly). Knowledge-based chat, call transcript…
Consigliato da Stefano Fiorucci
-
🐘 We're excited to share that Mobius Labs will be at IBC 2024, debuting our cutting-edge Aana SDK! This powerful open-source AI toolkit is designed…
🐘 We're excited to share that Mobius Labs will be at IBC 2024, debuting our cutting-edge Aana SDK! This powerful open-source AI toolkit is designed…
Consigliato da Stefano Fiorucci
-
Je suis heureux de vous présenter "Chocolatine", mon tout nouveau modèle d'IA générative. Ce modèle marque une avancée majeure dans mon projet de…
Je suis heureux de vous présenter "Chocolatine", mon tout nouveau modèle d'IA générative. Ce modèle marque une avancée majeure dans mon projet de…
Consigliato da Stefano Fiorucci
-
This afternoon is my first day of work! After three months of sabbatical.. Time flies. Some months ago I decided I would slow down, work less…
This afternoon is my first day of work! After three months of sabbatical.. Time flies. Some months ago I decided I would slow down, work less…
Consigliato da Stefano Fiorucci
Altri profili simili
Altre persone che si chiamano Stefano Fiorucci in Italia
Su LinkedIn ci sono altre 22 persone che si chiamano Stefano Fiorucci in Italia
Vedi altre persone che si chiamano Stefano Fiorucci