Papers by Yingjin Song
Burn After Reading: Do Multimodal Large Language Models Truly Capture Order of Events in Image Sequences? (2025.findings-acl)
Copied to clipboard
| Challenge: | Existing benchmarks focus on single image settings, but some focus on multi-image settings. |
| Approach: | They introduce the TempVS benchmark which focuses on temporal grounding and reasoning capabilities of Multimodal Large Language Models in image sequences. |
| Outcome: | The proposed model performs poorly compared to human models in vision and language tasks. |
Disentangling the Roles of Representation and Selection in Data Pruning (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for data pruning involve many different design choices, which have not been systematically studied. |
| Approach: | They decompose data pruning into two key components: data representation and selection algorithm . theoretical and empirical results highlight crucial role of representations . |
| Outcome: | The proposed method can be used to train models with less data. |
Modeling Emotion Dynamics in Song Lyrics with State Space Models (2023.tacl-1)
Copied to clipboard
| Challenge: | Existing studies in music emotion recognition assume a single label for the whole song, but annotated data is scarce and difficult to obtain. |
| Approach: | They propose a method to predict emotion dynamics in song lyrics without annotation . they frame each song as a time series and use a State Space Model to generate the full emotion dynamics. |
| Outcome: | The proposed method improves performance of sentence-level baselines without annotating songs, making it ideal for limited training scenarios. |