Seungjun Baek

Dongik Shin, Beomsuk Kim, Seungjun Baek

Automated segmentation of ultrasound images can assist medical experts with diagnostic and therapeutic procedures. Although using the common modality of ultrasound, one typically needs separate datasets in order to segment, for example, different anatomical structures or lesions with different levels of malignancy. In this paper, we consider the problem of jointly learning from heterogeneous datasets so that the model can improve generalization abilities by leveraging the inherent variability among datasets. We merge the heterogeneous datasets into one dataset and refer to each component dataset as a subgroup. We propose to train a single segmentation model so that the model can adapt to each sub-group. For robust segmentation, we leverage recently proposed Segment Anything model (SAM) in order to incorporate sub-group information into the model. We propose SAM with Condition Embedding block (CEmb-SAM) which encodes sub-group conditions and combines them with image embeddings from SAM. The conditional embedding block effectively adapts SAM to each image sub-group by incorporating dataset properties through learnable parameters for normalization. Experiments show that CEmb-SAM outperforms the baseline methods on ultrasound image segmentation for peripheral nerves and breast cancer. The experiments highlight the effectiveness of Cemb-SAM in learning from heterogeneous datasets in medical image segmentation tasks.

Minsang Kim, Seungjun Baek

Real-world knowledge graphs (KG) are mostly incomplete. The problem of recovering missing relations, called KG completion, has recently become an active research area. Knowledge graph (KG) embedding, a low-dimensional representation of entities and relations, is the crucial technique for KG completion. Convolutional neural networks in models such as ConvE, SACN, InteractE, and RGCN achieve recent successes. This paper takes a different architectural view and proposes ComDensE which combines relation-aware and common features using dense neural networks. In the relation-aware feature extraction, we attempt to create relational inductive bias by applying an encoding function specific to each relation. In the common feature extraction, we apply the common encoding function to all input embeddings. These encoding functions are implemented using dense layers in ComDensE. ComDensE achieves the state-of-the-art performance in the link prediction in terms of MRR, HIT@1 on FB15k-237 and HIT@1 on WN18RR compared to the previous baseline approaches. We conduct an extensive ablation study to examine the effects of the relation-aware layer and the common layer of the ComDensE. Experimental results illustrate that the combined dense architecture as implemented in ComDensE achieves the best performance.

Minsang Kim, Seungjun Baek

Large language models (LLMs) closely interact with humans, and thus need an intimate understanding of the cultural values of human society. In this paper, we explore how open-source LLMs make judgments on diverse categories of cultural values across countries, and its relation to training methodology such as model sizes, training corpus, alignment, etc. Our analysis shows that LLMs can judge socio-cultural norms similar to humans but less so on social systems and progress. In addition, LLMs tend to judge cultural values biased toward Western culture, which can be improved with training on the multilingual corpus. We also find that increasing model size helps a better understanding of social values, but smaller models can be enhanced by using synthetic data. Our analysis reveals valuable insights into the design methodology of LLMs in connection with their understanding of cultural values.

Minsang Kim, Seungjun Baek

Compute-efficient training of language models has become an important issue. We consider data pruning for data-efficient training of LLMs. In this work, we consider a data pruning method based on information entropy. We propose that the samples in the training corpus be ranked in terms of their informativeness which we estimate through entropy functions. The key idea is that, less informative samples are likely to contain redundant information, and thus should be pruned first. We use the entropy functions based on the negative log-likelihood and the average inverse word frequency of a sample as a surrogate to measure its informativeness. Experiments reveal that the proposed information-based pruning can improve upon various language modeling and downstream tasks, and enhance the generalization capability of language models.