Taki Hasan Rafi

Bio

I am a Ph.D. candidate in the Data Intelligence Lab at Hanyang University, working with Prof. Dong-Kyu Chae. Before joining HYU in 2022, I received a bachelor's degree in Electrical Engineering from AUST in 2021. My Ph.D. is supported by Brain Korea 21 (NRF)! I occasionally collaborate with researchers from Oracle Inc. and TD Securities.

Research Interests

My research interests lie in two directions at the intersection of vision and language AI.

• Data-Efficient Robust Learning: In a world where models are often deployed in unseen environments, the ability to perform well with limited or no access to target domain data is challenging. My work explores methods to build robust models that can generalize effectively to new, potentially very different domains. Furthermore, I investigate methods for test-time adaptation, allowing models to dynamically adjust to the data stream at inference time to maximize model performance in real-world scenarios. Another scope of my research is identifying accurate and robust training data for model training that can influence model performance at inference. Hence, I aim to develop algorithms that are both accurate and adaptable, enabling reliable and efficient machine-learning solutions across diverse and evolving data distributions.
• Safety and Reliability of AI: LLMs and VLMs are increasingly used in everyday life, they must be dependable with responses. My work involves building benchmarks and evaluating these AI systems to see how well they handle different prompts. I focus on areas to improve the robustness of the model under bias, and cultural deviation. I also explore how well these models work in low-resource language settings and how humans can best collaborate with AI to improve results. A key area is understanding how LLMs perform across different languages, especially when data is scarce and diverse. Eventually, I aim to help build safer, more trustworthy AI that benefits everyone, regardless of language or resource availability.

Recent News

Invited as a Reviewer at ACM MM 2025!

Invited as a Reviewer at ACL 2025 (Industry Track)!

Invited as a Reviewer at ACL ARR (February Cycle)!

One paper gets accepted at NAACL 2025!

Three papers get accepted at DASFAA 2025!

One paper gets accepted at WWW 2025!

Selected Recent Papers

[NAACL 2025]
SweEval: Do LLMs Really Swear? A Safety Benchmark for Testing Limits for Enterprise Use
Hitesh Laxmichand Patel, Amit Agarwal, Arion Das, Bhargava Kumar, Srikant Panda, Priyaranjan Pattnayak, Taki Hasan Rafi, Tejaswini Kumar, Dong-Kyu Chae
Annual Conference of the North American Chapter of the Association for Computational Linguistics (Industry Track), 2025

[DASFAA 2025]
Instance-Aware Test-Time Adaptation for Domain Generalization
Taki Hasan Rafi, Serbeter Karlo, Amit Agarwal, Hitesh Patel, Bhargava Kumar, and Dong-Kyu Chae
30th International Conference on Database Systems for Advanced Applications, 2025
(Acceptance rate ≈ 32%)

[ACCV 2024]
GReFEL: Geometry-Aware Reliable Facial Expression Learning under Bias and Imbalanced Data Distribution
Azmine Toushik Wasi*, Taki Hasan Rafi*, Serbeter Karlo, Raima Islam, and Dong-Kyu Chae (*=Equal contributions)
17th Asian Conference on Computer Vision, 2024
(Acceptance rate ≈ 32%)