Computer Vision for video understanding and Interpretability of automated decisions

Course Description

Part I (Viktoriia): Intro to deep learning in video data: spatial and temporal dynamic streams, convolutional neural networks, recurrent neural networks (pros and cons). Examples of automatic action recognition and emotion recognition. Elements of neural style transfer in video data. Part II (Novi): Interpretability/explainability: generating decisions that take into account how well a human could understand the decisions in the given context. Explicitly explaining decisions to humans.

Course topics

Deep convolutional neural networks, object recognition, transparency, interpretability, trust.

Course tools

Python, Caffe, Tensorflow.

Prerequisites

Basic linear algebra, proficiency in Python, machine learning basics (understanding of different types of learning (supervised, unsupervised learning), classification, regression problems, generalization error, overfitting, train/test datasets split). Optional, but desirable: neural network (NN) basics, feed forward NN, different activation functions, backpropagation.

Lecturers

Dr. Viktoriia Sharmanska

Imperial College London, United Kingdom Viktoriia is passionate about designing intelligent systems that can learn concepts from visual data using machine learning models. She got her PhD in Computer Vision and Machine Learning from IST Austria, and MSc in Applied Mathematics from Taras Shevchenko National University of Kyiv, Ukraine. Since 2015, she’s a visiting research fellow at the University of Sussex, UK working on cross modal learning with privileged information. In October 2017, she joined Imperial College London as a research fellow, where she’s working on action and emotion recognition in video data using deep learning. Computer Vision and Machine Learning. Contacts[email protected] linkedin.com/in/viktoriiasharmanska/  

Dr. Novi Quadrianto

Senior Lecturer at the University of Sussex, United Kingdom Novi is a co-founder and co-director of Predictive Analytics Lab (PAL) at University of Sussex, Brighton, UK. In addition to undertaking high quality research and publishing in top machine learning and computer vision conferences/journals including NIPS, ICML, CVPR, ICCV/ECCV, JMLR, and TPAMI, the PAL group also creates significant impact by providing support, technology, and highly-trained specialists to a new generation of technology companies. Prior to Sussex, Novi was a Newton International Fellow of the Royal Society at University of Cambridge, UK. He is an Associate Editor of TPAMI and an Area Chair of NIPS 2017 and 2015. Ethical Machine Learning. Contacts[email protected]