TALP Talk: My Recurrent Neighbour is Now PyTorched: a Long story Shortened in Time with deep Messages

My Recurrent Neighbour is Now PyTorched: a Long story Shortened in Time with deep Messages


date: Wednesday 27-09-2017 time: From 11:00 to 13:00

room: Campus NORD UPC, Building D5, Classroom MERIT D5-010, Barcelona

by: Santi Pascual, PhD Researcher at TALP



PyTorch is a Python Deep Learning framework, where we can mainly build Artificial Neural Networks (ANNs or NNs), similarly to what we achieve with TensorFlow, Keras, Theano, Caffe2, and many others. There is, however, a hidden beauty behind PyTorch, and a nice level of simplicity to achieve our Deep Learning goals quickly and effectively.

This talk will cover the basic PyTorch structures to build a neural network, BUT going straight into the deeply recursive world of Recurrent Neural Networks. Thus we will play with an RNN in a way as interpretable as possible (thus understanding what's going on inside), and delve into its evolved Long Short Term Memory (LSTM) version with some toy example. The idea is to demonstrate how fun and easy and every-good-adjective it is to build the so called dynamic graphs (like RNNs) in PyTorch.


(For other questions, contact This email address is being protected from spambots. You need JavaScript enabled to view it. )

1) What prior knowledge do I need for this tutorial/meetup/handson?
- I know what is a Neural Network (at least the typical MLP structure with sigmoid/tanh/relu units), and I absolutely know what a Loss function is, what does it mean to train a Machine Learning model, etc.. (Machine Learning basics in the end of the day, as we will move on to RNNs quickly).

2) Why RNNs?
- Because they are (1) challenging, (2) a key component in many nowadays Deep Learning applications, and (3) PyTorch is a very interesting framework to implement these dynamic models (e.g. models dealing with varying length data).

3) Do I have to bring my laptop with any installation?
- Yes, you should definetly bring a laptop if you want to actually learn about this (as it is a coding hands-on session). Bring the following requirements installed:

* Python 3.5+ (you can work with 2.7, but you'd be able to solve any syntax mismatch b/w both Python versions if they appear, something I don't expect to happen often)
* Latest NumPy
* Latest PyTorch for your Python version (install whatever fits your distro from http://pytorch.org/ )

-Tip: easiest thing to do is downloading an Anaconda package (https://www.anaconda.com/download/) that works for your system, and install PyTorch there, this way you won't have to worry about any extra packages required.

4) Are we working with CUDA devices? (i.e. Do we need a GPU?)
-No, we'll work with proofs of concept that don't need GPUs on our laptops, so we'll fatigue our CPUs. Nonetheless if you have a CUDA device you're absolutely free to install the PyTorch version compiled with CUDA, it'll work on CPU anyway.

20/09/2017 - Abraham Woubie Zewoudie PhD dissertation

PhD THESIS DEFENSE Mr. Abraham Woubie Zewoudie

Date: Sep 20, 2017, time: 11:00

Room: Aula de Seminaris D5-007 
Edificio D5, Campus Nord, UPC.

Discriminative features for GMM and i-vector based speaker diarization

Advisors: Javier Hernando, Jordi Luque

Antonio Bonafonte parla al suplement de catalunya radio sobre assistents virtuals i cerques per veu

A “Tema suplement” parlem sobre màquines controlades per veu humana: telèfons intel·ligents, GPS, robots... Els assistents virtuals i les cerques per veu han canviat la nostra comunicació i plantegen problemes ètics, de privacitat i lingüístics. En parlem amb Iratxe Gómez, veu de l’assistent Siri de l’iPhone, Antonio Bonafonte al minut 22'09'', investigador del centre de tecnologies i aplicacions del llenguatge i la parla de la Universitat Politècnica de Catalunya, l’actor i doblador Abel Folk i l’escriptor Màrius Serra. 


  • Join us

    Job Offers and Research Positions

  • Awards

    Scientific awards Best: Publications, Thesis, Demos, ...

  • Seminars

    Framework to promote research related with Natural Language Processing or Speech Processing.

  • Thesis

    TALP PhD dissertation

  • Events

    Information about Courses, Conferences, workshops, talks, ...

  • Call For Papers

    Relevant Conference, workshops or special issues call for papers announcement

  • Relevant Publications
  • Members

    Professors, students and colaborators visiting TALP.

    New members incorporation and former members carreer.

  • Projects

    TALP research and innovation projects

  • Press

    Selected articles and reports about the TALP Research Center taken from the international press

  • Formation

Additional information