arXiv subjects

Seohyun Lee

Publications and source records attributed to Seohyun Lee.

9 recordsLinked to original sources

arXiv2026

Reason in the Words You Speak: Idiolectal Paraphrasing Off-Policy Traces for Reasoning Distillation in VideoLLMs

Recent large language models achieve strong performance on complex reasoning tasks, where reinforcement learning with Group Relative Policy Optimization (GRPO) has emerged as a leading paradigm for optimizing models on self-generated trajectories. However, the on-policy nature of GRPO bounds the model to the reasoning skills it can already produce, restricting to learn more advanced capabilities. Prior works inject privileged reasoning traces from a stronger teacher policy to guide training, yet these traces are inherently out of distribution with respect to the student policy. We observe that this mismatch between on-policy and off-policy causes gradient clipping on semantically critical reasoning tokens, ultimately rewarding correct answers while leaving the reasoning that justifies them unlearned. Hence, we propose \textbf{Echo-GRPO}, a framework that lets the model reason in the words it speaks. Rather than imitating low-probability privileged traces from the teacher model, Echo-GRPO rewrites them into the student policy's own \textit{idiolect}, that is, its own characteristic vocabulary and expression patterns, while preserving their semantics via Dual-Reference Decoding. We instantiate this framework as \textbf{VideoEcho-R1} for video reasoning distillation, achieving consistent improvements across three multimodal LLM backbones and five benchmarks. Finally, we show that our idiolectal paraphrasing is a plug-in module that consistently improves both RL and supervised fine-tuning frameworks for reasoning distillation, demonstrating that policy-aligned supervision extends beyond GRPO.

cs.CV↗

arXiv2026

VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement

As video corpora continue to expand in both scale and task complexity, there is increasing demand for approaches that retrieve relevant videos from large-scale corpora (inter-video reasoning) and subsequently perform fine-grained, query-conditioned tasks (intra-video reasoning) within the retrieved content, such as temporal grounding. However, existing approaches typically treat retrieval as a preprocessing step, and consequently, when the initial retrieval fails, there is no mechanism to refine the search, leading to the failure of subsequent fine-grained intra-video reasoning. Moreover, while recent agentic frameworks have advanced video understanding, they typically assume that the query-relevant video is already given, focusing exclusively on intra-video reasoning tasks. To address these limitations, we propose VideoSearch-R1, an agentic framework for iterative video retrieval and reasoning through multi-turn interaction with a video search engine. Specifically, we introduce Soft Query Refinement (SQR) to refine search query tokens in a continuous latent space rather than rewriting queries in the discrete text space, enabling more efficient and fine-grained adjustments. SQR and its reasoning process are trained using Group Relative Policy Optimization (GRPO), guided by task-level reward signals derived from retrieval and downstream tasks. Building upon this, VideoSearch-R1 achieves state-of-the-art performance across three datasets on Video Corpus Moment Retrieval (VCMR), iteratively retrieving videos from large-scale corpora, refining search queries, and performing precise query-conditioned temporal grounding within the retrieved content. Our analyses show that SQR effectively refines the original query, requiring significantly fewer generated tokens than explicit text-level query refinement. Code and model checkpoints are publicly available at mlvlab.github.io/VideoSearch-R1.

cs.CV↗

arXiv2026

Detecting and Mitigating Backdoor Attacks in OTA-FL Systems: A Two-Stage Robust Aggregation Scheme

Over-the-air federated learning (OTA-FL) improves communication efficiency by exploiting the superposition property of wireless channels, but this same property also creates a critical security vulnerability: the parameter server (PS) cannot access individual local updates, making it difficult to identify and exclude poisoned gradients. The challenge is further exacerbated under non-independent and identically distributed (Non-IID) training data, where benign gradient drift can closely resemble malicious updates. In this paper, we propose a two-stage robust aggregation framework for defending against backdoor attacks in OTA-FL. Under our scheme, each client is first assigned a modality-aware multi-indicator trust score, where the specific indicators are selected according to the data modality (e.g., waveform, text, image) and model architecture to capture the most discriminative footprint of backdoor updates. Based on this score, the PS then performs trust-based multiple access (TBMA) to separate clients into trusted, suspicious, and malicious categories. Suspicious clients are further examined through PS-side layer-wise inspection and a longitudinal reputation mechanism. Experimental results on several datasets demonstrate that the proposed methodology effectively suppresses stealthy backdoor attacks, including bounded-scaling attacks, Euclidean-constrained attacks, Cosine-constrained attacks, and Neurotoxin, while maintaining competitive main-task accuracy.

cs.CR↗

arXiv2026

Self-Play Enhancement via Advantage-Weighted Refinement in Online Federated LLM Fine-Tuning with Real-Time Feedback

Recent works have advanced feedback-based learning systems, whereby a foundation model is able to intake incoming feedback (e.g., a user) to self-improve, creating a self-loop system of training. However, existing works are limited in needing to consider an offline setup to allow for such feedback-based methods, and are further limited in the need of requiring privileged ground-truth contexts for training. Moreover, there is limited consideration of federated learning (FL), which is particularly well-suited for incorporating external feedback across large networks of end users, for example, but requires methods to be efficient for training on resource-constrained edge devices. Therefore, we introduce SPEAR (Self-Play Enhancement via Advantage-Weighted Refinement), an efficient online learning algorithm for federated LLM fine-tuning. SPEAR utilizes a feedback-guided self-play loop to construct naturally contrastive pairs per prompt which are utilized to be trained on (i) standard maximum likelihood on correct completions and (ii) confidence-weighted unlikelihood on tail tokens of incorrect completions. Without the need of expensive group generations and ground-truth contexts for training (i.e., only partial, non-answer feedback), in contrast with existing works, SPEAR can be trained both online and in a resource-efficient manner. We validate SPEAR across various benchmark datasets, demonstrating its superior performance in comparison to state-of-the-art baselines. The implementation code is publicly available at https://github.com/lee3296/SPEAR.

cs.LG↗

arXiv2026

MoE-GRPO: Optimizing Mixture-of-Experts via Reinforcement Learning in Vision-Language Models

Mixture-of-Experts (MoE) has emerged as an effective approach to reduce the computational overhead of Transformer architectures by sparsely activating a subset of parameters for each token while preserving high model capacity. This paradigm has recently been extended to Vision-Language Models (VLMs), enabling scalable multi-modal understanding with reduced computational cost. However, the widely adopted deterministic top-K routing mechanism may overlook more optimal expert combinations and lead to expert overfitting. To address this limitation and improve the diversity of expert selection, we propose MoE-GRPO, a reinforcement learning (RL)-based framework for optimizing expert routing in MoE-based VLMs. Specifically, we formulate expert selection as a sequential decision-making problem and optimize it using Group Relative Policy Optimization (GRPO), allowing the model to learn adaptive expert routing policies through exploration and reward-based feedback. Furthermore, we introduce a modality-aware router guidance that enhances training stability and efficiency by discouraging the router from exploring experts that are infrequently activated for a given modality. Extensive experiments on multi-modal image and video benchmarks show that MoE-GRPO consistently outperforms standard top-K routing and its variants by promoting more diverse expert selection, thereby mitigating expert overfitting and enabling a task-level expert specialization.

cs.CV↗

arXiv2025

TAP: Two-Stage Adaptive Personalization of Multi-Task and Multi-Modal Foundation Models in Federated Learning

In federated learning (FL), local personalization of models has received significant attention, yet personalized fine-tuning of foundation models remains underexplored. In particular, there is a lack of understanding in the literature on how to personalize foundation models in settings where there exist heterogeneity not only in data, but also in tasks and modalities across the clients. To address this gap, we propose Two-Stage Adaptive Personalization (TAP). In the first stage, TAP leverages mismatched model architectures between clients and the server to selectively replace personalized parameters with global updates, explicitly limiting cross-task and cross-modality interference. In the second stage, TAP conducts post-FL distillation on the global model to recover a beneficial shared structure. By reintroducing generalizable knowledge only after the global model has stabilized, TAP enhances generalization without compromising personalization. In developing our methodology, we introduce the first convergence analysis of federated foundation model training at the server under modality-task pair heterogeneity across clients, and demonstrate the impact of the number of modality-task pairs on model fine-tuning. Through extensive experiments, we demonstrate the effectiveness of TAP across a variety of datasets and tasks in comparison to state-of-the-art baselines. The implementation code is publicly available at https://github.com/lee3296/TAP.

cs.LG↗

arXiv2025

Cooperative Decentralized Backdoor Attacks on Vertical Federated Learning

Federated learning (FL) is vulnerable to backdoor attacks, where adversaries alter model behavior on target classification labels by embedding triggers into data samples. While these attacks have received considerable attention in horizontal FL, they are less understood for vertical FL (VFL), where devices hold different features of the samples, and only the server holds the labels. In this work, we propose a novel backdoor attack on VFL which (i) does not rely on gradient information from the server and (ii) considers potential collusion among multiple adversaries for sample selection and trigger embedding. Our label inference model augments variational autoencoders with metric learning, which adversaries can train locally. A consensus process over the adversary graph topology determines which datapoints to poison. We further propose methods for trigger splitting across the adversaries, with an intensity-based implantation scheme skewing the server towards the trigger. Our convergence analysis reveals the impact of backdoor perturbations on VFL indicated by a stationarity gap for the trained model, which we verify empirically as well. We conduct experiments comparing our attack with recent backdoor VFL approaches, finding that ours obtains significantly higher success rates for the same main task performance despite not using server information. Additionally, our results verify the impact of collusion on attack performance.

cs.LG↗

arXiv2025

ODPG: Outfitting Diffusion with Pose Guided Condition

Virtual Try-On (VTON) technology allows users to visualize how clothes would look on them without physically trying them on, gaining traction with the rise of digitalization and online shopping. Traditional VTON methods, often using Generative Adversarial Networks (GANs) and Diffusion models, face challenges in achieving high realism and handling dynamic poses. This paper introduces Outfitting Diffusion with Pose Guided Condition (ODPG), a novel approach that leverages a latent diffusion model with multiple conditioning inputs during the denoising process. By transforming garment, pose, and appearance images into latent features and integrating these features in a UNet-based denoising model, ODPG achieves non-explicit synthesis of garments on dynamically posed human images. Our experiments on the FashionTryOn and a subset of the DeepFashion dataset demonstrate that ODPG generates realistic VTON images with fine-grained texture details across various poses, utilizing an end-to-end architecture without the need for explicit garment warping processes. Future work will focus on generating VTON outputs in video format and on applying our attention mechanism, as detailed in the Method section, to other domains with limited data.

cs.CV↗

arXiv2024

Smart Information Exchange for Unsupervised Federated Learning via Reinforcement Learning

One of the main challenges of decentralized machine learning paradigms such as Federated Learning (FL) is the presence of local non-i.i.d. datasets. Device-to-device transfers (D2D) between distributed devices has been shown to be an effective tool for dealing with this problem and robust to stragglers. In an unsupervised case, however, it is not obvious how data exchanges should take place due to the absence of labels. In this paper, we propose an approach to create an optimal graph for data transfer using Reinforcement Learning. The goal is to form links that will provide the most benefit considering the environment's constraints and improve convergence speed in an unsupervised FL environment. Numerical analysis shows the advantages in terms of convergence speed and straggler resilience of the proposed method to different available FL schemes and benchmark datasets.

cs.LG↗