Juan Yeo

Ijun Jang, Jewon Yeom, Juan Yeo et al.

Knowledge distillation (KD) is a widely adopted technique for transferring knowledge from large language models to smaller student models; however, conventional supervised KD often suffers from a distribution mismatch between training and inference. While on-policy KD approaches attempt to mitigate this issue by learning directly from student-generated outputs, they frequently encounter training instabilities because the distributional gap between the novice student and the expert teacher is often too wide to bridge directly. These challenges manifest as pathological gradients in forward KL objectives or diversity collapse in reverse KL regimes. To address these limitations, we propose Veto, an objective-level reformulation that constructs a geometric bridge in the logit space. Unlike prior methods that mix data samples, Veto creates an intermediate target distribution that promotes alignment between the teacher and the student. By introducing a tunable parameter beta, Veto serves as an Adaptive Gradient Veto that stabilizes optimization by suppressing harmful gradients on low-confidence tokens, while simultaneously acting as a Decisiveness Knob to balance reward-driven performance with output diversity. Extensive experiments across various reasoning and generation tasks demonstrate that Veto consistently outperforms supervised fine-tuning and existing on-policy baselines.

Juan Yeo, Soonwoo Cha, Jiwoo Song et al.

Vision-language models such as CLIP have recently propelled open-vocabulary dense prediction tasks by enabling recognition of a broad range of visual concepts. However, CLIP still struggles with fine-grained, region-level understanding, hindering its effectiveness on these dense prediction tasks. We identify two pivotal factors required to address this limitation: semantic coherence and fine-grained vision-language alignment. Current adaptation methods often improve fine-grained alignment at the expense of semantic coherence, and often rely on extra modules or supervised fine-tuning. To overcome these issues, we propose Any-to-Any Self-Distillation (ATAS), a novel approach that simultaneously enhances semantic coherence and fine-grained alignment by leveraging own knowledge of a model across all representation levels. Unlike prior methods, ATAS uses only unlabeled images and an internal self-distillation process to refine representations of CLIP vision encoders, preserving local semantic consistency while sharpening local detail recognition. On open-vocabulary object detection and semantic segmentation benchmarks, ATAS achieves substantial performance gains, outperforming baseline CLIP models. These results validate the effectiveness of our approach and underscore the importance of jointly maintaining semantic coherence and fine-grained alignment for advanced open-vocabulary dense prediction.

Juan Yeo, Ijun Jang, Taesup Kim

Dense self-supervised learning (SSL) methods showed its effectiveness in enhancing the fine-grained semantic understandings of vision models. However, existing approaches often rely on parametric assumptions or complex post-processing (e.g., clustering, sorting), limiting their flexibility and stability. To overcome these limitations, we introduce Patch-level Kernel Alignment (PaKA), a non-parametric, kernel-based approach that improves the dense representations of pretrained vision encoders with a post-(pre)training. Our method propose a robust and effective alignment objective that captures statistical dependencies which matches the intrinsic structure of high-dimensional dense feature distributions. In addition, we revisit the augmentation strategies inherited from image-level SSL and propose a refined augmentation strategy for dense SSL. Our framework improves dense representations by conducting a lightweight post-training stage on top of a pretrained model. With only 14 hours of additional training on a single GPU, our method achieves state-of-the-art performance across a range of dense vision benchmarks, demonstrating both efficiency and effectiveness.

Juan Yeo, Jinkwan Jang, Kyubyung Chae et al.

Recent studies show that pretrained vision models can boost performance in audio downstream tasks. To enhance the performance further, an additional pretraining stage with large scale audio data is typically required to infuse audio specific knowledge into the vision model. However, such approaches require extensive audio data and a carefully designed objective function. In this work, we propose bypassing the pretraining stage by directly fine-tuning the vision model with our Look Aside Adapter (LoAA) designed for efficient audio understanding. Audio spectrum data is represented across two heterogeneous dimensions time and frequency and we refine adapters to facilitate interactions between tokens across these dimensions. Our experiments demonstrate that our adapters allow vision models to reach or surpass the performance of pretrained audio models in various audio and speech tasks, offering a resource efficient and effective solution for leveraging vision models in audio applications.