3 days / 40+ speakers
12 workshops

May 17-19, 2017 | Vilnius, Lithuania
http://devdays.lt/wp-content/uploads/2017/02/Sébastien-Bratières2-320x320.jpg
University of Cambridge, Italy

Sébastien Bratières


Sebastien is a machine intelligence specialist with experience in speech interfaces and machine learning. He has spent 15 years in the speech and language industry in different European ventures, starting from the EU branch of Tellme Networks (now Microsoft) to startups in speech recognition and virtual conversational agents. His last stint was with dawin gmbh, working on voice products for the maintenance and inspection industry.

A few years ago, he decided to expand his expertise from language technologies to machine learning, and undertook a PhD with the machine learning group at the University of Cambridge. Just now, he is writing up his thesis on””Non parametric Bayesian models for structured output prediction”.

Sebastien graduated with an Ingenieur degree from Ecole Centrale Paris and an MPhil in Speech and Language Processing from the University of Cambridge. He is French and lives in Rome, Italy, with his Italian wife, and has fun raising his bilingual children.


Topic

What every developer should know about machine learning

Whether it comes up in a casual conversation at work, or as a concrete project opportunity, there are things every developer at heart should know about machine learning. In fact there are several such things. Many. Too many. So in this talk, we want to tie together loose ends, bits and pieces that you might have heard or read about, and get some important concepts nailed down. This talk aims at blowing away the hype fog, and focus on important, clear-cut notions, with just the right technicality.

Topics:

  • a whirlwind tour: prediction, reinforcement learing, unsupervised learning and dimensionality reduction, probabilistic approaches, optimization
  • comments on Five tribes in artificial intelligence, P Domingos
  • basic concepts: features, model parameters, performance measurement, training, testing, inductive bias
  • why is my accuracy low? bias-variance decomposition, model capacity, regularization, overfitting, data peeking, no free lunch theorem
  • important approaches to ML, libraries: deep learning, linear models, kernel machines, decision tree models & random forests, probabilistic models
  • changing faces of ML in industry: shipping products, cloud services, portal user analytics, data science, big data


Make

Connection

Follow