Skip to content

Commit 5407a7e

Browse files
Keyu-HeKeyu-He
committed
Add news items, update news layout & bibliography
1 parent 126e888 commit 5407a7e

7 files changed

Lines changed: 34 additions & 8 deletions

File tree

‎_bibliography/papers.bib‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -2,12 +2,12 @@
22
---
33
44
@article{keyu2025vlm,
5-
abbr={PREPRINT},
5+
abbr={ACL 2026},
66
title={Believing without Seeing: Quality Scores for Contextualizing Vision-Language Model Explanations},
77
author={Keyu He and Tejas Srinivasan and Brihi Joshi and Xiang Ren and Jesse Thomason and Swabha Swayamdipta},
8-
year={2025},
9-
journal={Under Review},
10-
abstract={When people query Vision-Language Models (VLMs) but cannot see the accompanying visual context (e.g. for blind and low-vision users), augmenting VLM predictions with natural language explanations can signal which model predictions are reliable. However, prior work has found that explanations can easily convince users that inaccurate VLM predictions are correct. To remedy undesirable overreliance on VLM predictions, we propose evaluating two complementary qualities of VLM-generated explanations via two quality scoring functions. We propose Visual Fidelity, which captures how faithful an explanation is to the visual context, and Contrastiveness, which captures how well the explanation identifies visual details that distinguish the model’s prediction from plausible alternatives. On the A-OKVQA and VizWiz tasks, these quality scoring functions are better calibrated with model correctness than existing explanation qualities. We conduct a user study in which participants have to decide whether a VLM prediction is accurate without viewing its visual context. We observe that showing our quality scores alongside VLM explanations improves participants’ accuracy at predicting VLM correctness by 11.1%, including a 15.4% reduction in the rate of falsely believing incorrect predictions. These findings highlight the utility of explanation quality scores in fostering appropriate reliance on VLM predictions.},
8+
year={2026},
9+
journal={ACL},
10+
abstract={When people query Vision-Language Models (VLMs) but cannot see the accompanying visual context (e.g. for blind and low-vision users), augmenting VLM predictions with natural language explanations can signal which model predictions are reliable. However, prior work has found that explanations can easily convince users that inaccurate VLM predictions are correct. To remedy undesirable overreliance on VLM predictions, we propose evaluating two complementary qualities of VLM-generated explanations via two quality scoring functions. We propose Visual Fidelity, which captures how faithful an explanation is to the visual context, and Contrastiveness, which captures how well the explanation identifies visual details that distinguish the model's prediction from plausible alternatives. On the A-OKVQA, VizWiz, and MMMU-Pro tasks, these quality scoring functions are better calibrated with model correctness than existing explanation qualities. We conduct a user study in which participants have to decide whether a VLM prediction is accurate without viewing its visual context. We observe that showing our quality scores alongside VLM explanations improves participants' accuracy at predicting VLM correctness by 11.1%, including a 15.4% reduction in the rate of falsely believing incorrect predictions. These findings highlight the utility of explanation quality scores in fostering appropriate reliance on VLM predictions.},
1111
selected={true},
1212
doi={10.48550/arXiv.2509.25844},
1313
pdf={https://arxiv.org/pdf/2509.25844},

‎_includes/news.liquid‎

Lines changed: 3 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -22,6 +22,9 @@
2222
{{ item.content | remove: '<p>' | remove: '</p>' | emojify }}
2323
{% else %}
2424
<a class="news-title" href="{{ item.url | relative_url }}">{{ item.title }}</a>
25+
{% unless include.limit %}
26+
<p class="news-preview" style="margin: 0.25em 0 0; font-size: 0.9em; color: var(--global-text-color-light);">{{ item.content | strip_html | truncatewords: 50 }}</p>
27+
{% endunless %}
2528
{% endif %}
2629
</td>
2730
</tr>

‎_news/2025-02-15-ELI-research.md‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,7 @@
11
---
22
layout: post
3-
title: Research on Pedagogical Utility of LLM Explanations
4-
date: 2025-02-15
3+
title: ELI-Why accepted at ACL Findings 2025
4+
date: 2025-05-15
55
---
66

7-
I am excited to share that we just submitted our research paper titled **"ELI-Why: Evaluating the Pedagogical Utility of LLM Explanations"** to ACL Rolling Review. In this work, we introduced **ELI-Why**, a benchmark to assess the pedagogical capabilities of LLMs, and we found that inference-time instructions alone is insufficient for LLMs to produce high-utility explanations tailored to users' informational needs.
7+
Our paper **"ELI-Why: Evaluating the Pedagogical Utility of LLM Explanations"** has been accepted at **ACL Findings 2025**! In this work, we introduced **ELI-Why**, a benchmark to assess the pedagogical capabilities of LLMs, and found that inference-time instructions alone are insufficient for LLMs to produce high-utility explanations tailored to users' informational needs.

‎_news/2025-08-25-cmu-start.md‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
---
2+
layout: post
3+
title: Started MIIS at Carnegie Mellon University
4+
date: 2025-08-25
5+
---
6+
7+
Excited to begin my **Masters of Science in Intelligent Information Systems** at **Carnegie Mellon University** School of Computer Science! Looking forward to diving deeper into NLP research and collaborating with brilliant minds in the field. Here's to a new chapter of learning and growth! 🚀📚

‎_news/2026-01-28-adobe-internship.md‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
---
2+
layout: post
3+
title: Joining Adobe as AI/ML Intern
4+
date: 2026-01-28
5+
---
6+
7+
Excited to share that I will be joining **Adobe** as an **AI/ML Intern** this summer in San Jose!

‎_news/2026-03-15-acl-acceptance.md‎

Lines changed: 7 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,7 @@
1+
---
2+
layout: post
3+
title: Paper accepted at ACL 2026
4+
date: 2026-03-15
5+
---
6+
7+
Our paper **"Believing without Seeing: Quality Scores for Contextualizing Vision-Language Model Explanations"** has been accepted at **ACL 2026**! We propose quality scoring functions for VLM-generated explanations that help users better assess model reliability without viewing the visual context. Excited to present this work in San Diego!

‎_pages/news.md‎

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,7 +1,9 @@
11
---
22
layout: page
3-
title: news
3+
title: News
44
permalink: /news/
5+
nav: true
6+
nav_order: 2
57
---
68

79
{% include news.liquid %}

0 commit comments

Comments
 (0)