Understanding Neural Networks
I am currently spending most of my time trying to understand how neural networks work. As a subscriber to “What I cannot create, I do not understand,” I believe that if we better understand the behaviours of neural networks we should be able to use this understanding to provide algorithmic improvements. This means thinking about topics such as generalisation, modularity, compositionality, unit-level understanding, and interpretability.
How Does The World Affect Representations & How Should We Represent The World?
I am also thinking a lot about how the structure of data that humans receive should inform our models. Almost all of the sensory input humans receive is highly structured multi-modal time series data - how can we use this structure to learn better models? This means looking into things like self-prediction with unshuffled data, independent causal mechanims (and causal structure in general), and multi-modality.
Exploring Career Options
My current position will be ending in March so I am now looking for the next opportunity. I plan to stay in AI research and am considering both industry and academic options.
What is this page? The now page is a place to let people know what I am focusing my time on right now. That is, a combination of my current focus at work and in personal projects. The Now Page.
Last updated: 03/01/2022. Last reviewed: 03/01/2022.