Supervision

Thesis Supervision

Ongoing

• Jiaao Li (with Nils Feldhus) - Master Thesis - Title: Code Verification of Thought: Natural Language Explanation Generation through Code Completion

Student Research Assistant Supervision

Ongoing

• Yoana Tsoneva (2025/01/16-): intent recognition in conversational XAI systems
• Fedor Splitt (2025/02/01): cross-lingual and multilingual intent recognition in conversational XAI systems

Open Topics

If you are interested in any topics mentioned below, please send me your CV, transcript of record, and a short motivation letter.

1. Development of new approaches for generating alteractuals

One of popular explainability method is called counterfactual, which refers to the edited input whose prediction after edition is different than that before edition. However, alterfactual is kind of the opposition, which refers to altered version and the prediction does not change. Since alterfactual generation is less explored by the NLP communities, the goal of the thesis is to create a new approach to generate alterfactual.

References: Mertes et al. (2024); Nguyen et al. (2024)

2. Refining and Aligning Natural Language Explanations through Human-Guided Feedback

References: Li et al. (2022); Hong et al. (2024); Chen et al. (2024)

3. How different quantization approaches affect feature importance

Quantization approaches can be divided into two categories: quantization-aware training and post-training quantization. In this thesis, we want to explore how different approaches affect feature importance, e.g. faithfulness.

Reference: Gholami et al. (2022); Wan et al. (2024); Zhu et al. (2024);