We also conduct a fine-grained error analysis and show that the underlying ground truth data in popular emotion datasets is inherently unreliable, adding to the challenges of this highly subjective task! Check out the preprint for more details! 🧐

We show that: 1️⃣ foundation models lag behind architectures designed specifically for emotion recognition, 2️⃣ show biases towards specific emotions and response formats, and 3️⃣ fail at predicting emotions under assumed personas.

Evaluating Vision-Language Models for Emotion Recognition Large Vision-Language Models (VLMs) have achieved unprecedented success in several objective multimodal reasoning tasks. However, to further enhance their capabilities of empathetic and effective comm...

Excited to share my recent work on evaluating VLMs for ✨evoked emotion recognition ✨: arxiv.org/abs/2502.05660 (will be in NAACL Findings)!

The Worlds I See - Dr. Fei-Fei Li’s autobiography.

Instead of 'What can AI teach every other field?', we should ask 'What can AI learn FROM every other field?', e.g. sociology, education, indigenous knowledge systems. After all, what kind of study of general intelligence does not try to build upon the amassed collective wisdom of all humanity?

Will be eagerly looking forward to your course nonetheless!

And can such representations be compounded?

In the context of multimodal foundation models, can we create models that can process such composite entities as a single modality of input?

Not sure if this is too naive an idea or more of an engineering than a research problem 😅

The possibilities of what a “modality” can be is endless. Considering composite entities (like a user or a complete human), can we say that a single representation of a complex entity is as good as/better/worse than some combination of its subparts (language/speech by the human, appearance, etc.)?

Okay genius idea to improve quality of #nlp #arr reviews. Literally give gold stars to the best reviewers, visible on open review next to your anonymously ID during review process.

Here’s why it would work, and why would you should RT this fab idea: