Ki-young Shin

Kunil Lee, Ki-Young Shin, Jong-Hyeok Lee et al.

Multilingual knowledge editing (MKE) remains challenging because language-specific edits interfere with one another, even when locate-then-edit methods work well in monolingual settings. This paper focuses on three issues: the effectiveness of vector merging methods for MKE, the extent to which Task Singular Vectors for Merging (TSVM) can reduce multilingual interference, and the influence of the weight scaling factor and rank compression ratio on performance. We evaluate six merging variants with two popular backbone large language models, two base knowledge editing methods, and 12 languages on the MzsRE benchmark under a large-scale batch-editing setting. Our results show that vector summation with shared covariance is the most reliable overall strategy, whereas simple summation without shared covariance performs poorly. TSVM improves performance in some settings, but its ability to mitigate multilingual interference is limited. We also find that performance is sensitive to both weight scale and rank ratio, with larger-than-default scaling and relatively low rank often yielding better results. These findings clarify the practical strengths and limits of current vector merging methods for MKE and provide guidance for future multilingual knowledge editing research.

Jianri Li, Jae-whan Lee, Woo-sang Song et al.

This paper describes Designovel's systems which are submitted to the Fashion IQ Challenge 2019. Goal of the challenge is building an image retrieval system where input query is a candidate image plus two text phrases describe user's feedback about visual differences between the candidate image and the search target. We built the systems by combining methods from recent work on deep metric learning, multi-modal retrieval and natual language processing. First, we encode both candidate and target images with CNNs into high-level representations, and encode text descriptions to a single text vector using Transformer-based encoder. Then we compose candidate image vector and text representation into a single vector which is exptected to be biased toward target image vector. Finally, we compute cosine similarities between composed vector and encoded vectors of whole dataset, and rank them in desceding order to get ranked list. We experimented with Fashion IQ 2019 dataset in various settings of hyperparameters, achieved 39.12% average recall by a single model and 43.67% average recall by an ensemble of 16 models on test dataset.