Hi to all Deep Learning interested people,
Our 11th meetup is coming up, and we have two exciting topics this time again: Distributed Deep Learning (+ Benchmarking) and a Bidirectional LSTM-HMM approach.
Our agenda is as follows:
• Welcome (Tom Lidy)
• Introduction by our host: Casinos Austria (Isabell Brandenberger)
• A Comparison of Deep Learning Frameworks for Distributed Training (Peter Ruch)
Deep Networks are computationally very expensive and training of state of the art models can take from a few hours up to several days. By distributing the training process across multiple devices/nodes it's possible to significantly speed up one's experiments or to train larger models. Peter will give a short introduction to distributed deep learning, present the common problems and pitfalls, and discuss performance benchmarks of the most popular DL-Frameworks when training across multiple devices/nodes.
• An Introduction to Bidirectional LSTM-HMM for Sound Event Detection (Ana Jalali)
This talk is about the hybrid system BLSTM-HMM (Bidirectional Long Short Term Memory - Hidden Markov Model) and is used for tasks such as Sound Event detection and Speech recognition. BLSTMS are extensions of Recurrent Neural Networks (RNNs) and they are a solution to the vanishing gradient problem. Bidirectional LSTMs not only have a connection to their previous steps, but also to the future steps. Hidden Markov Models are extended to this system in order to improve the accuracy of the mentioned tasks.
• Latest News and Hot Topics (Jan Schlüter and Tom Lidy)
• Open discussions
This meetup takes place in the Casinos Austria Innovation Hub and we will kindly be supported by drinks & snacks.
IMPORTANT NOTE: We will start at 18:30 sharp and entry into the building is only possible until 19:00 (no later entry possible!).
Looking forward to a fruitful meetup,
Tom, Jan and Alex