Andriy Drozdyuk

About

I was born in Ukraine. I finished high school in Toronto, Canada after which I got my bachelors in Computer Science at University of Toronto. I received my masters in Computer Science at University of New Brunswick, Fredericton, Canada. Presently I’m enrolled in PhD at Carleton University in Ottawa, Canada.

Projects

Upside Down Reinforcement Learning implementation based on 2019 Juergen Schmidhuber paper: https://github.com/drozzy/upsidedown

Publications

Lioutas, Vasileios and Andriy Drozdyuk. “Copy this Sentence.” ArXiv abs/1905.09856 (2019)
Attention is an operation that selects some largest element from some set, where the notion of largest is defined elsewhere. Applying this operation to sequence to sequence mapping results in significant improvements to the task at hand. In this paper we provide the mathematical definition of attention and examine its application to sequence to sequence models. We highlight the exact correspondences between machine learning implementations of attention and our mathematical definition. We provide clear evidence of effectiveness of attention mechanisms evaluating models with varying degrees of attention on a very simple task: copying a sentence. We find that models that make greater use of attention perform much better on sequence to sequence mapping tasks, converge faster and are more stable.

Scott Buffett, Michael Fleming and Andriy Drozdyuk “Incremental Sequential Rule Mining with Streaming Input Traces”, Canadian AI (2020)

Data mining studies how to separate useful data from the rest. Recent work in the field of Sequential Rule Mining has led to mining new types of rules, partially ordered sequential rules. However, existing approaches do not consider the case of unbounded data. With unbounded data, conventional approaches to data mining are not sufficient because they become slow as the dataset increases in size. We propose an algorithm that makes use of previous intelligence when processing new information. This requires a slight reformulation of the problem and the solution. In the end, we show that our algorithm, on average, outperforms the existing approaches when applied to unbounded data.