Abstract

Speaker 1: Sian Gooding

Title: Predicting Text Readability and Reading Comprehension from Reading Interactions

Abstract: Judging the readability of text has many important applications, for instance in text simplification and when sourcing reading material for language learners. Additionally, being able to infer if a reader has understood text can be crucial when delivering critical information. In this paper, we present a 600 participant study which investigates how the implicit user interactions of readers relate to (1) the complexity of a text and (2) the user’s comprehension of the text. We make our dataset publicly available and show that implicit measures of interaction not only correlate with text readability and reading comprehension but are also able to predict them.

Speaker 2: Siana Zhekova

Title: Project Silica: Encoding and Storing Data into Glass

Abstract: Over the last decade, the potential applications of the Cloud in the form of a long-term data storage medium have been evolving and significantly expanding into the zettabytes. Nonetheless, current technologies do not possess an efficient solution to a perennial data storage framework. In order for us to establish such a large-scale system, we would need to re-design its operational specifications as well as the underlying storage platforms.

The teams at Project Silica at Microsoft Research have been developing an innovative technology incorporating recent applications of ultra-fast laser optics to encode data in quartz glass by using femtosecond lasers, and decode it efficiently via imaging processed by machine learning algorithms.

In the summer of 2020, over the course of 12 weeks I was virtually interning at Microsoft Research, Cambridge, where I was a part of the Cloud Computing team at Project Silica. My project revolved around developing a framework for detecting different types of error patterns that could be attributed to errors occurring at various stages of the decoding pipeline of optically encoded data, ranging from polarization state errors to inaccuracies occurring in the machine learning algorithms that had been deployed.