#Talk 1
Speaker: Giovanni Vacanti, Data Scientist at Seldon Title: Machine learning model explainability and Alibi, an open source model inspection python library. Bio: Giovanni is a theoretical quantum physicist and data science expert with eight years of experience in academic research. He received his Ph.D from the National University of Singapore, and has also carried out work in Italy, France, and the United Kingdom. Giovanni is highly skilled in mathematical modeling, numerical analysis of complex systems, analytic calculations, and problem solving in general. Abstract: Machine learning models explanations, or “black box” explanations, has become in recent years a quickly evolving field of research. The importance of explainability in Machine learning and in AI, in general, is important for two reasons: 1) It helps scientists to understand better the behaviour of ML models by providing insight about the decision making process; 2) it contributes to the process of building trust in AI systems, which is a the core of any real-world deployment of such systems. In this talk, we will review some state of the art model explanation techniques and illustrate how such techniques can be applied to a real-world use case. We will also highlight Alibi, a new open source python library designed for ML models inspection and explainability which uses tensorflow as a backend.
#Talk 2
Speaker: Igor Gotlibovych, Head of Machine Learning at Octopus Energy Title: Beyond MSE: Tensorflow models for predicting and sampling random time series Bio: Igor used to think that everything is a filter, but now he thinks everything is a neural network. Following a PhD in quantum physics at the University of Cambridge, he has pursued a number of roles from sailing across oceans to developing algorithms for medical diagnostics. His main interests are in deep learning, time series data, and smart energy grids. Abstract: When dealing with forecasting energy consumption, and beyond predicting a single target variable, it is important to quantify errors (which may themselves be seasonal, skewed etc.), as well as to be able to generate (sample) realistic "alternative" scenarios. In this talk, Igor will cover background and applications to smart energy grids, variational maximum likelihood estimation using TensorFlow, implementing and fitting Mixture Density Networks (MDNs), sampling from MDNs and auto-regressive MDNs and correlated time series.
See you there!
TensorFlow London Meetup |
