Homepage

Act as if what you do makes a difference. It does. - William James

tl;dr

I’m Yiorgos Chochlakis, 5th-year CS PhD student at USC with Shrikanth Narayanan in SAIL. I work on complex subjective language tasks (i.e., problems where we can agree to disagree) and analyse how LLMs perform on them.

Welcome 👋! My name is Georgios (but please call me Yiorgos) Chochlakis. I am a Computer Science Ph.D. fellow at the University of Southern California and Signal Analysis and Interpretation Laboratory, supervised by prof. Shrikanth Narayanan. I am currently interested and actively publishing on (what I call) complex subjective tasks, such as emotion recognition, with LLMs. To define terms, complex refers to inherently difficult, usually multilabel, problems with interrelated labels, and subjective refers to settings where people care reasonably disagree about their interpretation of a stimulus. Why are these problems important, you ask? Plenty of reasons actually, you’re gonna have to read my papers for that, but let me try to summarize that before you Cmd/Ctrl+W out of here. First, what I would call the resurgence of Aristotelian logic: how do verifiable rewards (and veriable rewards only) influence the behavior of LLMs in domains where there is no right and wrong? I am collecting a multimodal dataset to figure that out at the moment. Second, how are you gonna claim AGI when you test your model on simplified problems? Really, single-label settings? Your model cannot do proper, multi-label classification and you think it’s gonna take over the world? The real world is murky and fuzzy, not “single-label”.

Before my academic enlightenment, and during my internships as ordered, I was publishing SOTA results. I have mostly focused on multimodal and/or multilingual emotion recognition, and mostly from social media. I have developed emotion recognition methods that can be deployed in dynamic environments, such as multilingual settings, different desired emotions, different annotation formats (e.g., transitioning from individual emotions to clusters of emotions), and multimodal yet modality-agnostic settings. Before I came to the US, I studied Electrical & Computer Engineering at the National Technical University of Athens (NTUA) under the supervision of prof. Alexandros Potamianos.

I also pretend to care about the philosophy of science, scientific rigor and reproducibility, having replicated (or failed to do so) results from published work (see here where I corrected an error in the paper, here, here, etc.), and doing my best to speedrun through books I should have read as an undergrad.

In my free time, I sleep, work out, and experience existential angst from the ghost of Ancient Greece and what could (should) have been.

I must not [procrastinate]. [Procrastination] is the mind-killer. [Procrastination] is the little death that brings total obliteration. I will face my [procrastination]. I will permit it to pass over me and through me. And when it has gone past, I will turn the inner eye to see its path. Where the [procrastination] has gone there will be nothing. Only I will remain. - Dune (Frank Herbert)