Hi! My name is Yuriel (or Ryan if your prefer) and I am a M.Eng student at the Singapore University of Technology and Design, kindly supported by the AI Singapore Accelerated Masters Program and the DSO-AISG Research Award. I go by “Yuriel Ryan” in my publications.

Broadly speaking, my interests lie in understanding what is relevant to a large model, how do these sources of information produce that task-relevant information, and where can we identify them. This means using theoretical frameworks, such as those from information theory, to construct my hypotheses and experiments.

Sources of Information could range from raw (multimodal) data to representations found within the hidden layers of a large model.

1) What is relevant?: training or evaluating models on difficult and abstract (human-specific) concepts (see Humor in Pixels).

2) How is it relevant?: applying or developing frameworks (See Multimodal Interaction Tuning) to illuminate how different sources of information come together in deep networks to produce the necessary information for a task.

3) Where can we find them?: designing solutions to train and search for the weights that contribute to a task (Currently working on this!) — I was particularly inspired by this Neural Thickets paper.

I managed to publish a few works on either (1) or (2) during my undergraduate and M.Eng days at SUTD. However, I am not quite satisfied just yet! I am continuously working on these three questions with the eventual goal of unifying them to produce useful Collaborative Artificial Intelligence that learn efficiently, and possess true understanding – as opposed to “memorizing/rote learning” — about the world.

I try my best to aim for rigorous “good science” to achieve actionable findings instead of chasing whatever that is popular in the AI space—this seems to change every couple of weeks anyways. If this research philosophy resonates with you, let’s work on something together! :)

Updates!