arXiv subjects

Soojeong Lee

Publications and source records attributed to Soojeong Lee.

6 recordsLinked to original sources

arXiv2026

BaRA: Budget-constrained and Reliable Web Data Collection Agent

Large language model (LLM)-based web agents automate web navigation and data collection. However, live web data collection demands capabilities beyond task completion: agents must discover site-internal pages and retrieve text, image, and video artifacts in an accessible form within a fixed interaction budget. We formulate this setting as budget-constrained, site-level multimodal web data collection and propose Budget-constrained and Reliable Agent (BaRA). BaRA performs breadth-first search (BFS)-based link discovery with liveness verification to filter hallucinated and dead links, then validates extracted multimodal artifacts using rule-based provenance and accessibility checks. A history-based self-reflection module recovers from execution failures and incomplete outputs. On controlled synthetic and real-world websites, BaRA consistently improves valid-link discovery and download-valid multimodal extraction over existing agents. Our code is available at https://github.com/MLAI-Yonsei/BaRA-Agent.

cs.IR↗

arXiv2026

Deployable Prototype Testing and Control Allocation of the CABLESSail Concept for Solar Sail Shape Control and Momentum Management

This paper presents prototype testing and a control allocation algorithm for the Cable-Actuated Bio-inspired Lightweight Elastic Solar Sail (CABLESSail) concept aimed at performing momentum management of a solar sail. CABLESSail uses actuated cables routed along the structural booms of the solar sail to control the shape of the solar sail and changes the solar radiation pressure disturbance torques acting on it. Small-scale prototype tests of CABLESSail are presented in this paper, which demonstrate the effectiveness of cable actuation on deployable booms. A novel control allocation method is also presented in this paper that provides a computationally-efficient manner to determine the deformations required in each of the structural booms to impart the desired momentum management torque on the solar sail. Numerical simulation results with the proposed algorithm demonstrate robustness to uncertainty in the shape of the sail membrane, resulting in reliable generation of momentum management torques that exceed or meet the capabilities of state-of-the-art solar sail actuators. Both the prototype tests and control allocation methods presented in this paper represent key steps in raising the technology readiness level of the CABLESSail concept.

physics.space-ph↗

arXiv2025

Passivity-Based Robust Shape Control of a Cable-Driven Solar Sail Boom for the CABLESSail Concept

Solar sails provide a means of propulsion using solar radiation pressure, which offers the possibility of exciting new spacecraft capabilities. However, solar sails have attitude control challenges because of the significant disturbance torques that they encounter due to imperfections in the sail and its supporting structure, as well as limited actuation capabilities. The Cable-Actuated Bio-inspired Lightweight Elastic Solar Sail (CABLESSail) concept was previously proposed to overcome these challenges by controlling the shape of the sail through cable actuation. The structural flexibility of CABLESSail introduces control challenges, which necessitate the design of a robust feedback controller for this system. The purpose of the proposed research here is to design a robust controller to ensure precise and reliable control of CABLESSail's boom. Taking into account the system dynamics and the dynamic properties of the CABLESSail concept, a passivity-based proportional-derivative (PD) controller for a single boom on the CABLESSail system is designed. To reach the nonzero desired setpoints, a feedforward input is additionally applied to the control law and a time-varying feedforward input is used instead of the constant one to effectively track a time-varying desired boom tip deflection. This control law is assessed by numerical simulations and by tests using a smaller-scale prototype of Solar Cruiser. Both the simulation and the test results show that this PD control with the time-varying feedforward input robustly controls the flexible cable-actuated solar sail.

eess.SY↗

arXiv2025

Adaptive Task Vectors for Large Language Models

In-Context Learning (ICL) enables Large Language Models (LLMs) to perform tasks without parameter updates by conditioning on a few demonstrations provided in the prompt. Despite its success, ICL suffers from several limitations, including sensitivity to demonstration order, context length constraints, and computational inefficiency. To address these challenges, task vector-based approaches compress task information into a single vector. However, these methods typically construct task vectors from fixed sets of demonstrations and reuse them across input queries, without conditioning on the specific input. This limitation can lead models to struggle with effective adaptation when the input query is not well aligned with the underlying demonstrations, consequently degrading their generalization performance on unseen tasks. To overcome this limitation, we propose Adaptive Task Vectors (ATV), a simple and effective framework that dynamically generates task vectors conditioned on each input query. ATV employs a small language model to generate task vectors, which are then transformed to match the target LLM's architecture and applied to guide its output generation. In contrast to ICL and previous vector-based approaches, which rely on fixed demonstration sets and their corresponding vectors, ATV dynamically generates task vectors tailored to each specific input query and task. Consequently, ATV demonstrates strong performance and generalization capabilities, even for unseen tasks. Furthermore, we provide a theoretical analysis indicating that ATV is expressively equivalent to LoRA under equal rank budgets and more expressive than Prefix-Tuning, thereby offering formal support for its representational advantage.

cs.LG↗

arXiv2023

Slot-Mixup with Subsampling: A Simple Regularization for WSI Classification

Whole slide image (WSI) classification requires repetitive zoom-in and out for pathologists, as only small portions of the slide may be relevant to detecting cancer. Due to the lack of patch-level labels, multiple instance learning (MIL) is a common practice for training a WSI classifier. One of the challenges in MIL for WSIs is the weak supervision coming only from the slide-level labels, often resulting in severe overfitting. In response, researchers have considered adopting patch-level augmentation or applying mixup augmentation, but their applicability remains unverified. Our approach augments the training dataset by sampling a subset of patches in the WSI without significantly altering the underlying semantics of the original slides. Additionally, we introduce an efficient model (Slot-MIL) that organizes patches into a fixed number of slots, the abstract representation of patches, using an attention mechanism. We empirically demonstrate that the subsampling augmentation helps to make more informative slots by restricting the over-concentration of attention and to improve interpretability. Finally, we illustrate that combining our attention-based aggregation model with subsampling and mixup, which has shown limited compatibility in existing MIL methods, can enhance both generalization and calibration. Our proposed methods achieve the state-of-the-art performance across various benchmark datasets including class imbalance and distribution shifts.

cs.CV↗

arXiv2023

Learning Missing Modal Electronic Health Records with Unified Multi-modal Data Embedding and Modality-Aware Attention

Electronic Health Record (EHR) provides abundant information through various modalities. However, learning multi-modal EHR is currently facing two major challenges, namely, 1) data embedding and 2) cases with missing modality. A lack of shared embedding function across modalities can discard the temporal relationship between different EHR modalities. On the other hand, most EHR studies are limited to relying only on EHR Times-series, and therefore, missing modality in EHR has not been well-explored. Therefore, in this study, we introduce a Unified Multi-modal Set Embedding (UMSE) and Modality-Aware Attention (MAA) with Skip Bottleneck (SB). UMSE treats all EHR modalities without a separate imputation module or error-prone carry-forward, whereas MAA with SB learns missing modal EHR with effective modality-aware attention. Our model outperforms other baseline models in mortality, vasopressor need, and intubation need prediction with the MIMIC-IV dataset.

cs.LG↗