arXiv subjects

Wonjong Rhee

Publications and source records attributed to Wonjong Rhee.

At least 19 recordsLinked to original sources

arXiv2026

Beyond On-Policy Exploration: Integrating External Policy Rollouts for Reinforcement Learning in Diffusion Language Models

Recent reinforcement learning methods for diffusion large language models (dLLMs) commonly rely on on-policy rollouts generated by the target dLLM itself. When successful on-policy rollouts are scarce, however, on-policy training may receive little positive reward and make only limited progress. To mitigate this problem, we explore incorporating higher-reward rollouts generated by a stronger external policy alongside on-policy rollouts from the target dLLM. However, directly incorporating these external rollouts introduces two practical challenges: differences in rollout length and instability when jointly processing rewards from on-policy and external rollouts. To address these challenges, we propose External Rollout Integration with Length Control and Source-Specific Processing (ERILS), which controls external-rollout length and processes the rewards of on-policy and external rollouts separately. Experiments on Sudoku, Countdown, and MATH500 under zero-shot evaluation show that ERILS improves multi-sample performance across all three tasks, with the largest gains on Sudoku. On Sudoku, ERILS achieves 98.4% best-of-4 completion accuracy, compared with 40.3% for the strongest baseline. ERILS also maintains approximately 90% deterministic single-completion accuracy on Sudoku across generation lengths of 128, 256, and 512 tokens. Our component analysis further shows that length-controlled external rollouts are more effective than uncontrolled external rollouts, and that source-specific reward processing avoids the training collapse observed with joint reward processing. These results show that rollout construction and reward processing are important design dimensions when integrating external rollouts into dLLM reinforcement learning.

cs.LG↗

arXiv2026

Personalized Deep Learning for Short-Term Forecasting of Impending Atrial Fibrillation from Continuous Wearable ECG Signals

Background and Objective: Continuous wearable electrocardiogram (ECG) monitoring is increasingly used for ambulatory arrhythmia surveillance, yet forecasting impending atrial fibrillation (AF) is challenged by inter-patient ECG variability. This study investigated whether personalizing a global model via fine-tuning on an individual's ECG signals improves short-term forecasting of impending AF. Methods: A global model trained on the ICENTIA11K dataset was compared against personalized models fine-tuned across three cohorts: ICENTIA11K, IRIDIA-AF, and MobiCARE. Following preprocessing, models processed 60-second ECG segments for a five-minute forecast horizon. We evaluated the impact of adaptation data volume and analyzed ECG features, such as heart rate and RMSSD. Results: Personalized models significantly outperformed the global model, achieving AUROCs of 0.711 vs. 0.614 in ICENTIA11K and 0.686 vs. 0.585 in MobiCARE. Personalization benefits increased with the amount of patient-specific fine-tuning data. While the global model's accuracy rose as AF onset approached, personalized models in the two external cohorts exhibited distinct temporal dynamics, which may indicate the capture of patient-specific characteristics less dependent on proximity to the AF event. Pre-AF episodes showed elevated heart rates and RMSSD. Feature attributions highlighted clinically relevant precursors, including frequent premature atrial complexes (PACs) and short supraventricular tachycardias (SVTs). Conclusions: Adapting deep learning models with patient-specific wearable ECG data significantly enhances short-term forecasting of impending AF. This personalized framework supports timely preventive interventions and improved AF management in ambulatory monitoring environments.

eess.SP↗

arXiv2026

Orthogonal Negative Guidance in Attention Feature Space for Text-to-Image Generation

Text-to-image (T2I) models have become increasingly capable of generating high-quality images. Yet, enforcing the explicit absence of a specified object or attribute remains a fundamentally challenging problem. Existing approaches, including prompt negation, post-hoc editing, and negative guidance, remain insufficient for explicit concept suppression, often failing to remove the target concept or degrading overall image quality. To this end, we propose Orthogonal Negative Guidance in attention feature space, a training-free method that operates in the attention output space of MM-DiT-based T2I transformers. Our method orthogonalizes negative-prompt attention features with respect to positive-prompt features and subtracts only the orthogonal component, suppressing unwanted concepts while preserving desired semantics. Experiments on FLUX-dev and FLUX-schnell show that our method achieves favorable trade-offs between concept suppression, prompt alignment, and image quality. In human evaluation, our method outperforms the second-best baseline by 18.78%. We further show that our method supports multi-concept suppression and adjustable concept suppression.

cs.CV↗

arXiv2026

When Confidence Misleads: Suffix Anchoring and Anchor-Proximity Confidence Modulation for Diffusion Language Models

Diffusion language models generate text by iteratively selecting and denoising masked positions, making position selection a central inference-time decision. Most training-free methods rely on model confidence, assuming that high-confidence positions are ready to be decoded. However, this assumption can fail in fully non-autoregressive (fully non-AR) decoding, where selection spans the entire masked response sequence. This can cause incomplete generation, partly due to EOT overconfidence. Existing remedies directly suppress EOT tokens, but provide limited gains or require additional fine-tuning. We show that placing even a semantically minimal suffix anchor near the end of the response region discourages premature EOT prediction and promotes complete generation. However, despite providing little supporting context, the anchor makes nearby positions highly confident and causes them to be decoded too early. We therefore propose Suffix-Anchored Confidence Modulation, a simple training-free method that uses suffix anchoring with progress-dependent confidence modulation near the anchor. Across text-only reasoning, vision-language reasoning, and code-generation benchmarks, our method outperforms diverse fully non-AR and semi-AR baselines. It also achieves particularly large gains under highly parallel decoding, demonstrating the benefit of improving the fully non-AR regime over block-wise semi-AR regime.

cs.CL↗

arXiv2026

Selective Aggregation of Attention Maps Improves Diffusion-Based Visual Interpretation

Numerous studies on text-to-image (T2I) generative models have utilized cross-attention maps to boost application performance and interpret model behavior. However, the distinct characteristics of attention maps from different attention heads remain relatively underexplored. In this study, we show that selectively aggregating cross-attention maps from heads most relevant to a target concept can improve visual interpretability. Compared to the diffusion-based segmentation method DAAM, our approach achieves higher mean IoU scores. We also find that the most relevant heads capture concept-specific features more accurately than the least relevant ones, and that selective aggregation helps diagnose prompt misinterpretations. These findings suggest that attention head selection offers a promising direction for improving the interpretability and controllability of T2I generation.

cs.CV↗

arXiv2025

DOS: Directional Object Separation in Text Embeddings for Multi-Object Image Generation

Recent progress in text-to-image (T2I) generative models has led to significant improvements in generating high-quality images aligned with text prompts. However, these models still struggle with prompts involving multiple objects, often resulting in object neglect or object mixing. Through extensive studies, we identify four problematic scenarios, Similar Shapes, Similar Textures, Dissimilar Background Biases, and Many Objects, where inter-object relationships frequently lead to such failures. Motivated by two key observations about CLIP embeddings, we propose DOS (Directional Object Separation), a method that modifies three types of CLIP text embeddings before passing them into text-to-image models. Experimental results show that DOS consistently improves the success rate of multi-object image generation and reduces object mixing. In human evaluations, DOS significantly outperforms four competing methods, receiving 26.24%-43.04% more votes across four benchmarks. These results highlight DOS as a practical and effective solution for improving multi-object image generation.

cs.CV↗

arXiv2025

Progressive Multimodal Search and Reasoning for Knowledge-Intensive Visual Question Answering

Knowledge-intensive visual question answering (VQA) requires external knowledge beyond image content, demanding precise visual grounding and coherent integration of visual and textual information. Although multimodal retrieval-augmented generation has achieved notable advances by incorporating external knowledge bases, existing approaches largely adopt single-pass frameworks that often fail to acquire sufficient knowledge and lack mechanisms to revise misdirected reasoning. We propose PMSR (Progressive Multimodal Search and Reasoning), a framework that progressively constructs a structured reasoning trajectory to enhance both knowledge acquisition and synthesis. PMSR uses dual-scope queries conditioned on both the latest record and the trajectory to retrieve diverse knowledge from heterogeneous knowledge bases. The retrieved evidence is then synthesized into compact records via compositional reasoning. This design facilitates controlled iterative refinement, which supports more stable reasoning trajectories with reduced error propagation. Extensive experiments across six diverse benchmarks (Encyclopedic-VQA, InfoSeek, MMSearch, LiveVQA, FVQA, and OK-VQA) demonstrate that PMSR consistently improves both retrieval recall and end-to-end answer accuracy.

cs.CV↗

arXiv2025

Soft Head Selection for Injecting ICL-Derived Task Embeddings

Large language models (LLMs) are commonly adapted to downstream tasks using parameter-efficient fine-tuning (PEFT) or in-context learning (ICL). Recently, ICL-driven embedding-based adaptation has been proposed as a distinct task adaptation paradigm. It derives task-specific embeddings from intermediate activations using few-shot prompts and injects them during inference. Despite its conceptual appeal, this approach has not demonstrated consistent performance gains over PEFT or ICL, and its empirical advantages have been limited in practice. We propose Soft head-selection for ICL-derived Task Embeddings (SITE), a gradient-based method that identifies task-relevant attention heads to enable effective task embedding injection. Across various types of open-ended generation, reasoning, and natural language understanding tasks, SITE significantly outperforms prior embedding-based adaptation methods and few-shot ICL, while using substantially fewer trainable parameters than PEFT. Experiments on 12 LLMs ranging from 4B to 70B parameters demonstrate the generality of our approach, and intra-task and inter-task activation patching analyses further provide new mechanistic insights by revealing strong task dependence in attention head functionality.

cs.CL↗

arXiv2025

ReFlex: Text-Guided Editing of Real Images in Rectified Flow via Mid-Step Feature Extraction and Attention Adaptation

Rectified Flow text-to-image models surpass diffusion models in image quality and text alignment, but adapting ReFlow for real-image editing remains challenging. We propose a new real-image editing method for ReFlow by analyzing the intermediate representations of multimodal transformer blocks and identifying three key features. To extract these features from real images with sufficient structural preservation, we leverage mid-step latent, which is inverted only up to the mid-step. We then adapt attention during injection to improve editability and enhance alignment to the target text. Our method is training-free, requires no user-provided mask, and can be applied even without a source prompt. Extensive experiments on two benchmarks with nine baselines demonstrate its superior performance over prior methods, further validated by human evaluations confirming a strong user preference for our approach.

cs.CV↗

arXiv2024

Task-Specific Preconditioner for Cross-Domain Few-Shot Learning

Cross-Domain Few-Shot Learning~(CDFSL) methods typically parameterize models with task-agnostic and task-specific parameters. To adapt task-specific parameters, recent approaches have utilized fixed optimization strategies, despite their potential sub-optimality across varying domains or target tasks. To address this issue, we propose a novel adaptation mechanism called Task-Specific Preconditioned gradient descent~(TSP). Our method first meta-learns Domain-Specific Preconditioners~(DSPs) that capture the characteristics of each meta-training domain, which are then linearly combined using task-coefficients to form the Task-Specific Preconditioner. The preconditioner is applied to gradient descent, making the optimization adaptive to the target task. We constrain our preconditioners to be positive definite, guiding the preconditioned gradient toward the direction of steepest descent. Empirical evaluations on the Meta-Dataset show that TSP achieves state-of-the-art performance across diverse experimental scenarios.

cs.LG↗

arXiv2024

Cross-Attention Head Position Patterns Can Align with Human Visual Concepts in Text-to-Image Generative Models

Recent text-to-image diffusion models leverage cross-attention layers, which have been effectively utilized to enhance a range of visual generative tasks. However, our understanding of cross-attention layers remains somewhat limited. In this study, we introduce a mechanistic interpretability approach for diffusion models by constructing Head Relevance Vectors (HRVs) that align with human-specified visual concepts. An HRV for a given visual concept has a length equal to the total number of cross-attention heads, with each element representing the importance of the corresponding head for the given visual concept. To validate HRVs as interpretable features, we develop an ordered weakening analysis that demonstrates their effectiveness. Furthermore, we propose concept strengthening and concept adjusting methods and apply them to enhance three visual generative tasks. Our results show that HRVs can reduce misinterpretations of polysemous words in image generation, successfully modify five challenging attributes in image editing, and mitigate catastrophic neglect in multi-concept generation. Overall, our work provides an advancement in understanding cross-attention layers and introduces new approaches for fine-controlling these layers at the head level.

cs.CV↗

arXiv2024

A Benchmark Suite for Evaluating Neural Mutual Information Estimators on Unstructured Datasets

Mutual Information (MI) is a fundamental metric for quantifying dependency between two random variables. When we can access only the samples, but not the underlying distribution functions, we can evaluate MI using sample-based estimators. Assessment of such MI estimators, however, has almost always relied on analytical datasets including Gaussian multivariates. Such datasets allow analytical calculations of the true MI values, but they are limited in that they do not reflect the complexities of real-world datasets. This study introduces a comprehensive benchmark suite for evaluating neural MI estimators on unstructured datasets, specifically focusing on images and texts. By leveraging same-class sampling for positive pairing and introducing a binary symmetric channel trick, we show that we can accurately manipulate true MI values of real-world datasets. Using the benchmark suite, we investigate seven challenging scenarios, shedding light on the reliability of neural MI estimators for unstructured datasets.

stat.ML↗

arXiv2024

Towards a Better Evaluation of Out-of-Domain Generalization

The objective of Domain Generalization (DG) is to devise algorithms and models capable of achieving high performance on previously unseen test distributions. In the pursuit of this objective, average measure has been employed as the prevalent measure for evaluating models and comparing algorithms in the existing DG studies. Despite its significance, a comprehensive exploration of the average measure has been lacking and its suitability in approximating the true domain generalization performance has been questionable. In this study, we carefully investigate the limitations inherent in the average measure and propose worst+gap measure as a robust alternative. We establish theoretical grounds of the proposed measure by deriving two theorems starting from two different assumptions. We conduct extensive experimental investigations to compare the proposed worst+gap measure with the conventional average measure. Given the indispensable need to access the true DG performance for studying measures, we modify five existing datasets to come up with SR-CMNIST, C-Cats&Dogs, L-CIFAR10, PACS-corrupted, and VLCS-corrupted datasets. The experiment results unveil an inferior performance of the average measure in approximating the true DG performance and confirm the robustness of the theoretically supported worst+gap measure.

cs.LG↗

arXiv2024

Unveiling Key Aspects of Fine-Tuning in Sentence Embeddings: A Representation Rank Analysis

The latest advancements in unsupervised learning of sentence embeddings predominantly involve employing contrastive learning-based (CL-based) fine-tuning over pre-trained language models. In this study, we analyze the latest sentence embedding methods by adopting representation rank as the primary tool of analysis. We first define Phase 1 and Phase 2 of fine-tuning based on when representation rank peaks. Utilizing these phases, we conduct a thorough analysis and obtain essential findings across key aspects, including alignment and uniformity, linguistic abilities, and correlation between performance and rank. For instance, we find that the dynamics of the key aspects can undergo significant changes as fine-tuning transitions from Phase 1 to Phase 2. Based on these findings, we experiment with a rank reduction (RR) strategy that facilitates rapid and stable fine-tuning of the latest CL-based methods. Through empirical investigations, we showcase the efficacy of RR in enhancing the performance and stability of five state-of-the-art sentence embedding methods.

cs.CL↗

arXiv2024

An Image Grid Can Be Worth a Video: Zero-shot Video Question Answering Using a VLM

Stimulated by the sophisticated reasoning capabilities of recent Large Language Models (LLMs), a variety of strategies for bridging video modality have been devised. A prominent strategy involves Video Language Models (VideoLMs), which train a learnable interface with video data to connect advanced vision encoders with LLMs. Recently, an alternative strategy has surfaced, employing readily available foundation models, such as VideoLMs and LLMs, across multiple stages for modality bridging. In this study, we introduce a simple yet novel strategy where only a single Vision Language Model (VLM) is utilized. Our starting point is the plain insight that a video comprises a series of images, or frames, interwoven with temporal information. The essence of video comprehension lies in adeptly managing the temporal aspects along with the spatial details of each frame. Initially, we transform a video into a single composite image by arranging multiple frames in a grid layout. The resulting single image is termed as an image grid. This format, while maintaining the appearance of a solitary image, effectively retains temporal information within the grid structure. Therefore, the image grid approach enables direct application of a single high-performance VLM without necessitating any video-data training. Our extensive experimental analysis across ten zero-shot video question answering benchmarks, including five open-ended and five multiple-choice benchmarks, reveals that the proposed Image Grid Vision Language Model (IG-VLM) surpasses the existing methods in nine out of ten benchmarks.

cs.CV↗

arXiv2024

Selectively Informative Description can Reduce Undesired Embedding Entanglements in Text-to-Image Personalization

In text-to-image personalization, a timely and crucial challenge is the tendency of generated images overfitting to the biases present in the reference images. We initiate our study with a comprehensive categorization of the biases into background, nearby-object, tied-object, substance (in style re-contextualization), and pose biases. These biases manifest in the generated images due to their entanglement into the subject embedding. This undesired embedding entanglement not only results in the reflection of biases from the reference images into the generated images but also notably diminishes the alignment of the generated images with the given generation prompt. To address this challenge, we propose SID~(Selectively Informative Description), a text description strategy that deviates from the prevalent approach of only characterizing the subject's class identification. SID is generated utilizing multimodal GPT-4 and can be seamlessly integrated into optimization-based models. We present comprehensive experimental results along with analyses of cross-attention maps, subject-alignment, non-subject-disentanglement, and text-alignment.

cs.CV↗

arXiv2024

Improving Forward Compatibility in Class Incremental Learning by Increasing Representation Rank and Feature Richness

Class Incremental Learning (CIL) constitutes a pivotal subfield within continual learning, aimed at enabling models to progressively learn new classification tasks while retaining knowledge obtained from prior tasks. Although previous studies have predominantly focused on backward compatible approaches to mitigate catastrophic forgetting, recent investigations have introduced forward compatible methods to enhance performance on novel tasks and complement existing backward compatible methods. In this study, we introduce an effective-Rank based Feature Richness enhancement (RFR) method, designed for improving forward compatibility. Specifically, this method increases the effective rank of representations during the base session, thereby facilitating the incorporation of more informative features pertinent to unseen novel tasks. Consequently, RFR achieves dual objectives in backward and forward compatibility: minimizing feature extractor modifications and enhancing novel task performance, respectively. To validate the efficacy of our approach, we establish a theoretical connection between effective rank and the Shannon entropy of representations. Subsequently, we conduct comprehensive experiments by integrating RFR into eleven well-known CIL methods. Our results demonstrate the effectiveness of our approach in enhancing novel-task performance while mitigating catastrophic forgetting. Furthermore, our method notably improves the average incremental accuracy across all eleven cases examined.

cs.LG↗

arXiv2024

Harmonizing Visual and Textual Embeddings for Zero-Shot Text-to-Image Customization

In a surge of text-to-image (T2I) models and their customization methods that generate new images of a user-provided subject, current works focus on alleviating the costs incurred by a lengthy per-subject optimization. These zero-shot customization methods encode the image of a specified subject into a visual embedding which is then utilized alongside the textual embedding for diffusion guidance. The visual embedding incorporates intrinsic information about the subject, while the textual embedding provides a new, transient context. However, the existing methods often 1) are significantly affected by the input images, eg., generating images with the same pose, and 2) exhibit deterioration in the subject's identity. We first pin down the problem and show that redundant pose information in the visual embedding interferes with the textual embedding containing the desired pose information. To address this issue, we propose orthogonal visual embedding which effectively harmonizes with the given textual embedding. We also adopt the visual-only embedding and inject the subject's clear features utilizing a self-attention swap. Our results demonstrate the effectiveness and robustness of our method, which offers highly flexible zero-shot generation while effectively maintaining the subject's identity.

cs.CV↗