PhD Research Seminar: Meta-learning in neural networks, Data Envelopment Analysis for the assessment of COVID-19 quarantine measures

First talk: Meta-Learning in Neural Networks    
Speaker: Alexander Grishin, third-year PhD student, Faculty of Computer Science

I will present an overview of meta-learning in deep learning. First, we will look at a high-level classification of modern meta-learning algorithms. In particular, we will discuss what, how, and why could be meta-learned. Then I will speak on meta-learning in reinforcement learning (RL) in more detail and on the key difference between various approaches in this area. In the end, I will share our idea on how to improve current results in meta-RL. It turns out that all approaches potentially suffer from bias in gradients. I will explain why this bias appears and how to alleviate it.


Second talk: New Methods of Data Envelopment Analysis and Its Application for the Efficiency Assessment of COVID-19 Quarantine Measures      
Speaker: Sergey Demin, third-year PhD student, Faculty of Economic Sciences

The majority of Data Envelopment Analysis (DEA) models need precise data for the efficiency assessment. However, there are many examples when the evaluation of input and output parameters used in the model cannot be made accurately. We propose new modifications of DEA for the case of interval data.

Using the Oxford COVID-19 Government Response Tracker’s data, we also provide a systematic way to measure and compare government responses to COVID-19 across countries by the evaluation of quarantine measures’ efficiency using constructed DEA methods.