Italiano
Il server monta 6 schede GPU Tesla V100 SXM2 32GB ed un HDD da 2 TB configurato in RAID 10. Il server è ottimizzato per l’utilizzo di Python (con un'installazione Miniconda preconfigurata con CUDA 10.1) e pensato per eseguire algoritmi di Deep Learning con molti strati di elaborazione che necessitano di un elevato numero di risorse computazionali per calcolo tensoriale.
Fonte di Finanziamento:
Media o grande attrezzatura acquisita/cofinanziata con fondi di Ateneo
anno del bando:
2018
anno di collaudo:
2020
Nome e acronimo del laboratorio o locale che ospita l'attrezzatura:
Intelligent Signal Processing and Multi Media (ISPAMM) laboratory
Department or host center:
Edificio:
RM032 - S. Pietro in Vincoli - Edificio B
Pagina web laboratorio/attrezzatura:
Contatti:
| cognome | nome | |
|---|---|---|
Scarpiniti | Michele | |
Scardapane | Simone |
Elenco Imprese utenti:
Elenco altri utenti:
Ricavi - trasferimenti interni:
Anno:
2020
fatture emesse:
| data |
|---|
29/10/2020 |
spese manutenzione:
| anno |
|---|
2020 |
Description of research activity:
The research activities mainly concern the investigation of new deep learning algorithms with many processing layers in order to improve the performance of current state-of-the-art solutions. This improvement concerns the number of units on the one hand, and on the other hand, more importantly, the choice of particular flexible activation functions, optimization and regularization techniques. A particular emphasis will be given to the applications of these algorithms to the processing of audio signals recorded by microphone arrays as well as to the new frontier of the Explainability, i.e., the meaning and interpretation of the values assumed by the units of the internal layers of a deep neural architectures.
Description of Third Mission activity:
The Third Mission activities mainly concern the implementation and execution of high-performance deep learning algorithms, which require very powerful computing architectures. In fact, in the applications requested by external companies, there is a constant demand for algorithms that involve a very high number of inputs and processing layers, as well as the possibility to manage a large amount of data to be used as the training set of such algorithms.
Description of educational/training activity:
The educational/training activities concern the usage of the GPU server during practice of Machine Learning and Deep Learning courses held by the instructors of the DIET department. The available equipment, in fact, makes it possible to simply demonstrate to the students how deep learning works. This is made by taking practical lessons oriented towards the interactive implementation of complex applications that are really functional and operating on large amounts of real-world data.
Responsabile dell'Attrezzatura:
michele.scarpiniti@uniroma1.it
Settore ERC:
PE6_11
Ambiti tecnologici trasversali - Key Enabling Technologies:
Big data & computing
Keyword iris:
Calcolo parallelo e distribuito
Apprendimento automatico
deep learning
Elaborazione dei segnali
Modelli matematici dei sistemi complessi
Stato dell'attrezzatura:
In funzione