Meetup #AperiTech della Community di Deep Learning Italia
Deep Learning Italia” torna a Roma con il suo Meetup . Giovedì 23 Aprile 2019, dalle ore 18.30 alle ore 20.00 il primo meet up ONLINE di DLI.
Si affronteranno temi legati esclusivamente alla progettazione e implementazione di Reti Neurali e il Deep Learning nell'Intelligenza Artificiale di oggi.
Il Meetup di DLI nasce dall'esigenza di sopperire alla carenza di meetup tecnici sul Deep Learning in Italia mentre ormai se ne contano molteplici in altre città del mondo come Londra, Amsterdam e San Francisco. Il Meetup sarà orientato sia alla ricerca accademica e che al mondo delle startup.
Per la registration ecco il link
Lorenzo Giusti Data Scientist @ CERN
"Extreme Rare events detection"
CERNs accelerators facilities require adequate tools to operate, maintain and guide the consolidation of the machines, to fulfil the physics program and deliver the required luminosity for the experiments. One of the main problems in the accelerators complex is the downtime due to a fault in the technical infrastructure which can cause up to hours of inactivity. However, the technical infrastructure is such a complex system that failure rate of a device is highly unlikely that have widespread impact and might destabilize the overall system, bringing the detection of an upcoming fault as a high priority task. In order to detect these extreme events, we propose a novel approach which exploits the strengths of the Denoising Autoencoders and the Bidirectional LSTM Neural Networks in order to learn the normal behaviour of a device analyzing its sensors and predict if it is being anomalous within the time necessary to maintain the infrastructure.
Luca Pedrelli Director Academy @DLI
"Deep Neural Networks for Time Series"
In this talk, we present Deep Echo State Networks as a tool to analyze and design efficient deep recurrent architectures for real-world tasks concerning time-series and sequence modelling.
Roberto Leuzzi - Codin S.p.A.
Energy-Based Models (EBMs) discover dependencies between variables by applying scalar energy to each configuration of variables. Inference consists of assign the value of observed variables and finding values to the remaining variables that minimize the energy. Machine learning consists of discovering an energy function that assigns low energy to the correct values of the remaining variables and higher energies to the incorrect values. A loss functional, minimized during training, is used to measure the quality of the energy functions. The EBM approach provides a common framework for probabilistic and non-probabilistic approaches.