| Proj No. | A3222-251 |
| Title | Detecting Hallucination in Large Vision Language Models |
| Summary | Recent advancements in Generative Artificial Intelligence (GAI) techniques have enabled numerous practical applications, including the creation of artwork and advancements in healthcare. However, recent studies have revealed that GAI techniques suffer hallucination issues, resulting in pressing security concerns to the public at large. Hallucination in GAI outputs poses a poisonous and challenging problem for GAI techniques. Hallucination refers to a phenomenon where the AI generated content seems plausible and coherent with the input but is factually inaccurate, nonsensical, or false. Hallucination can lead to a wide variety of unintentional harms, among which can be very problematic in some critical applications, such as medical diagnosis and financial analysis. Therefore, it is of utmost importance to detect hallucinations in Large Vision Language Models (LVLM). This project's objective is to first collect a dataset for LVLM hallucination detection. Secondly, based on the dataset, this project will build a comprehensive benchmark to rigorously evaluate the detection performance of state-of-the-art models. In this project, the students should complete the following tasks: 1. Algorithm implementation: The students are required to implement both the baseline methods and recent state-of-the-art methods in hallucination detection. Additionally, they are encouraged to develop innovative approaches. 2. Data understanding: The students should gain a comprehensive understanding of the entire process, become well-versed with the evaluation metrics, and evaluate the outcomes of the implemented algorithms. 3. The required knowledge and skill for this project can be summarized as follows: (a) Python; (b) Pytorch. (c) OpenCV. Interested candidates can email their CVs to me (bihan.wen@ntu.edu.sg). Only qualified candidates will be notified. |
| Supervisor | A/P Wen Bihan (Loc:S2 > S2 B2B > S2 B2B 54, Ext: +65 67904708) |
| Co-Supervisor | - |
| RI Co-Supervisor | - |
| Lab | Centre for Information Sciences & System (CISS) (Loc: S2-B4b-05) |
| Single/Group: | Single |
| Area: | Digital Media Processing and Computer Engineering |
| ISP/RI/SMP/SCP?: |