📅 Date: Thursday, July 17, 2025

🕘 Time: 9:00 AM – 10:00 AM

📍 Venue: VISHC Open Space, VinUniversity

🔗 Zoom Link: Join here 

🎙 Speakers:

1️⃣ Nguyen Lam Duy Khang

🗣 TOPIC: MACHINE LEARNING APPLICATION FOR INDOOR AIR QUALITY MULTISENSOR TRACKING AND PREDICTION

💡 Abstract: Indoor air quality (IAQ) poses a serious global challenge, especially as people spend over 90% of their time indoors. While existing ML/DL models for IAQ prediction show promise, they often prioritize univariate performance and ignore deployment limitations. This talk introduces a novel multi-sensor framework that integrates data from CO₂, PM₂.₅, temperature, humidity, and more, aiming to enhance prediction robustness and enable real-world applicability in IAQ tracking.

2️⃣ Oanh Tran, Research Assistant, VISHC

🗣 TOPIC: A NORM-BASED APPROACH TO IDENTIFYING IRRELEVANT TOKENS IN LVLMS
💡 Abstract: Despite the impressive performance of Large Vision Language Models (LVLMs), recent studies reveal that irrelevant or “background” tokens — often overlooked — can absorb excessive attention. This talk investigates these attention sink tokens and proposes a norm-based method to detect them in a single run. Findings suggest these tokens, though seemingly unrelated to specific prompts, still carry subtle semantic signals that affect downstream performance. The study contributes to improving LVLM interpretability and robustness.

📌 Speaker Short Bio: Oanh Tran graduated with a Bachelor of Science majoring in Computer Science from Ho Chi Minh City University of Technology (HCMUT). She is currently a Research Assistant at VinUniversity, working under the guidance of Prof. Doan Dang Khoa and Prof. Heng Ji (UIUC). Her research interests include trustworthy AI and multimodal AI in NLP, particularly in the context of Large Language Models (LLMs).

#VISHC #VinUniversity #VISHCSeminar

_________

VinUni-Illinois Smart Health Center (VISHC) 

✉️ shc@vinuni.edu.vn

🌐 https://smarthealth.vinuni.edu.vn