Delavnica: Deep learning with Keras
Dogodek bo potekal prek videokonferenčnega sistema Zoom.
SLO:
Na delavnici Globoko učenje s Kerasom se boste seznanili s knjižnico Keras in njeno uporabo na praktičnih primerih iz globokega učenja. V zadnjih letih se je globoko učenje razširilo iz akademskih krogov in našlo svoje mesto tudi v industriji, oglaševalstvu, zdravstvu, bančništvu, varnosti in tako naprej. Na delavnici bomo predstavili najuporabnejše arhitekture nevronskih mrež in njihove sestavne dele. Modele bomo implementirali znotraj oblačne storitve Colaboratory, ki že vključuje vso potrebno programsko opremo in omogoča enostavno deljenje in izvajanje kode na Googlovi visoko zmogljivi infrastrukturi (GPU, TPU). Preko zgledov boste spoznali glavne tipe podatkov, ki jih srečujemo v strojnem učenju, in na kakšen način jih uporabimo kot vhod ali izhod pri nevronskih mrežah. Predstavili bomo smernice, ki nas vodijo pri zasnovi primernega modela in pogoste ovire, ki jih srečujemo v globokem učenju.
ENG:
On a Deep learning with Keras workshop you will get acquainted with the Keras library and its use through practical examples from deep learning. In recent years, deep learning has spread from academic circles and found its place in industry, advertising, healthcare, banking, security and so on. In the workshop, we will present the most useful neural network architectures and their components. We will implement the models within the Colaboratory cloud service, which already includes all the necessary software and enables easy code sharing and execution on Google's high-performance infrastructure (GPU, TPU). Through examples, you will learn about the main types of data we encounter in machine learning and how we use them as input or output with neural networks. We will present guidelines that lead us in designing an appropriate model and the common obstacles we encounter in deep learning.
Organizer
Instructors
- Janez Povh, PhD, Faculty for Mechanical Engineering, Univeristy of Ljubljana
- Primoz Godec, Faculty of Computer and Information Science, Univeristy of Ljubljana
- Rok Hribar, Jožef Stefan Institute
