@shanestorks.bsky.social
NLP researcher and postdoc at University of Michigan Weinberg Institute for Cognitive Science. Grounded language understanding and reasoning. (he/him) 🏳️🌈
Adding more details. Space is (very) limited. Please contact me by next Wednesday 1/15/2025 for full consideration. Proposal doesn’t have to be formal.
January 10, 2025 at 1:23 PM
Adding more details. Space is (very) limited. Please contact me by next Wednesday 1/15/2025 for full consideration. Proposal doesn’t have to be formal.
So happy to finally share this last piece of my dissertation (and my first post on Bluesky)!
Obligatory photo after my recent hooding attached 🧑🎓
Obligatory photo after my recent hooding attached 🧑🎓
December 17, 2024 at 1:59 PM
So happy to finally share this last piece of my dissertation (and my first post on Bluesky)!
Obligatory photo after my recent hooding attached 🧑🎓
Obligatory photo after my recent hooding attached 🧑🎓
Compared to vanilla VLMs, our interventions improve the accuracy of mistake detection and the relevance, coherence, and efficiency of explanations.
We also show that patterns in metrics can indicate common issues in VLMs, such as visual hallucination! 😵💫
We also show that patterns in metrics can indicate common issues in VLMs, such as visual hallucination! 😵💫
December 17, 2024 at 1:59 PM
Compared to vanilla VLMs, our interventions improve the accuracy of mistake detection and the relevance, coherence, and efficiency of explanations.
We also show that patterns in metrics can indicate common issues in VLMs, such as visual hallucination! 😵💫
We also show that patterns in metrics can indicate common issues in VLMs, such as visual hallucination! 😵💫
In this work, we expand the recently studied problem of procedural mistake detection in images to require explanations through self-Q&A. 👁🗨🤖💬
We define automated metrics for explanation coherence, and incorporate them into VLMs with various inference and fine-tuning methods.
We define automated metrics for explanation coherence, and incorporate them into VLMs with various inference and fine-tuning methods.
December 17, 2024 at 1:59 PM
In this work, we expand the recently studied problem of procedural mistake detection in images to require explanations through self-Q&A. 👁🗨🤖💬
We define automated metrics for explanation coherence, and incorporate them into VLMs with various inference and fine-tuning methods.
We define automated metrics for explanation coherence, and incorporate them into VLMs with various inference and fine-tuning methods.