Deep Learning Meetup #5
Détails
Dear DeepLearners,
Our next Meetup, is scheduled on Thursday 25/02 at 7:00 pm !
We'll be pleased to welcome you in our new offices !
Speakers :
• Matthieu Cord : Deep CNN and weak supervision for visual recognitionDeep learning and Convolutional Neural Networks (CNN) are state-of-the-art methods for various visual recognition tasks, e.g. image classification or object detection. To better identify or localize objects, bounding box annotations are often used. These rich annotations quickly become too costly to get, making the development of Weakly Supervised Learning (WSL) models appealing.
We discuss several strategies (recently published in CVPR and ICCV) to automatically select relevant image regions from weak annotations (e.g. image-level labels) in deep CNN.
LIP6 (http://webia.lip6.fr/~cord/M._Cord.html)
• Tristan Deleu : Learning algorithms with Neural Turing Machines
Among the recent innovations in Deep Learning, the idea of adding en external memory to existing architectures has shown promising results, especially in NLP. In this talk, we will have a quick overview of these networks before diving deeper in the particular example of Neural Turing Machines. We will see how Neural Turing Machines work and how they can learn complete algorithms only from examples.
Apprendre des algorithmes avec les Neural Turing Machines
Parmi les récentes innovations en Deep Learning, l’idée d’ajouter une mémoire externe à certaines architectures existantes a montré des résultats prometteurs, notamment en NLP. Dans cette présentation, nous aurons une rapide vue d’ensemble de ces réseaux avant de se plonger plus en profondeur dans l’exemple des Neural Turing Machines. Nous verrons comment les Neural Turing Machines fonctionnent et comment ils sont capables d’apprendre des algorithmes complets uniquement à partir d’exemples.
• Vincent Gire : Oscar, a tool for hyperparameters research.
oscar.sensout.com (http://oscar.sensout.com/)
• Aloïs Gruson : Deep Learning for Music Recommendation
We present the recent advances we’ve made at niland using deep learning for automatic music similarity estimation and how we used social data to mimic human perception of similarity. We also describe how we used deep convolutional networks to account for the specifics of music. We finally show how we could push forward our approach to extract relevant informations directly from the raw audio.
Food and drinks will be provided.
Deep Learning Meetup #5