Pohang University of Science and Technology

The work proposes Dynamic Expert Pruning (DEP): its analysis shows that in multi-agent systems different tasks and roles recruit different experts while static pruning applies one fixed mask to every request; DEP uses an agent's system and task prompts to produce a per-request expert mask in a single forward pass with no per-configuration calibration, achieving better overall accuracy than static pruning and merging baselines across diverse tasks, roles, model scales, and MoE architectures, generalizing to workflows unseen in training, with the largest margin when few experts are retained.

The work proposes Dynamic Expert Pruning (DEP): its analysis shows that in multi-agent systems different tasks and roles recruit different experts while static pruning applies one fixed mask to every request; DEP uses an agent's system and task prompts to produce a per-request expert mask in a single forward pass with no per-configuration calibration, achieving better overall accuracy than static pruning and merging baselines across diverse tasks, roles, model scales, and MoE architectures, generalizing to workflows unseen in training, with the largest margin when few experts are retained.

The work proposes causality-enforced evolutional networks (CEEN), which divide the time domain into nonoverlapping subintervals, assign a unique neural network to each subinterval, build the loss from the integral form of the PDE (trapezoidal rule), and train sequentially from the initial time step, achieving higher accuracy than the original PINN on long-time integration of the KdV, Schrödinger, convection, and Allen-Cahn equations while requiring less computational cost and memory, with a parallelization algorithm for further speedup.

The work proposes causality-enforced evolutional networks (CEEN), which divide the time domain into nonoverlapping subintervals, assign a unique neural network to each subinterval, build the loss from the integral form of the PDE (trapezoidal rule), and train sequentially from the initial time step, achieving higher accuracy than the original PINN on long-time integration of the KdV, Schrödinger, convection, and Allen-Cahn equations while requiring less computational cost and memory, with a parallelization algorithm for further speedup.

The authors revisit a watermark-distillation protocol for LLM fingerprinting, show that its utility evaluation understates text-quality degradation in open-ended generation, and propose near-tie restriction, which uses top-1-relative logit gaps to confine the watermark bias to tokens close to the base model's top prediction; across Llama-3.2-3B, Qwen-2.5-3B, Llama-3.1-8B, and Gemma-9B, this improves detection-quality frontiers under sampling, system prompts, quantization, pruning, and fine-tuning, preserves higher text quality across query budgets, and further improves existing schemes when combined with them.

The authors revisit a watermark-distillation protocol for LLM fingerprinting, show that its utility evaluation understates text-quality degradation in open-ended generation, and propose near-tie restriction, which uses top-1-relative logit gaps to confine the watermark bias to tokens close to the base model's top prediction; across Llama-3.2-3B, Qwen-2.5-3B, Llama-3.1-8B, and Gemma-9B, this improves detection-quality frontiers under sampling, system prompts, quantization, pruning, and fine-tuning, preserves higher text quality across query budgets, and further improves existing schemes when combined with them.

OpenBox introduces a two-stage automatic annotation pipeline: in the first stage it uses a 2D vision foundation model and cross-modal instance alignment to associate instance-level cues from 2D images with the corresponding 3D point clouds; in the second stage it categorizes instances by rigidity and motion state and generates adaptive bounding boxes with class-specific size statistics, thereby producing high-quality 3D bounding box annotations without self-training and improving accuracy and efficiency over baselines on the Waymo Open Dataset, the Lyft Level 5 Perception dataset, and the nuScenes dataset.

OpenBox introduces a two-stage automatic annotation pipeline: in the first stage it uses a 2D vision foundation model and cross-modal instance alignment to associate instance-level cues from 2D images with the corresponding 3D point clouds; in the second stage it categorizes instances by rigidity and motion state and generates adaptive bounding boxes with class-specific size statistics, thereby producing high-quality 3D bounding box annotations without self-training and improving accuracy and efficiency over baselines on the Waymo Open Dataset, the Lyft Level 5 Perception dataset, and the nuScenes dataset.

This study develops the P2F method, a node-assigned hybrid framework that couples a parameterized Node-Assigned physics-informed neural network (NA-PINN) with a finite difference method (FDM) solver: the parameterized NA-PINN takes the water-level difference, initial velocity, and time t as inputs and learns a solution manifold so that a single trained network serves as a data-free surrogate for the momentum conservation equation across all flow paths, while the FDM solver advances the mass conservation equation at each time step to ensure exact discrete mass conservation; verification on a six-tank gravity-driven draining scenario yields a water level mean absolute error of 7.85×10⁻⁵ m and a velocity mean absolute error of 3.21×10⁻³ m/s under the nominal condition with Δt = 1.

This study develops the P2F method, a node-assigned hybrid framework that couples a parameterized Node-Assigned physics-informed neural network (NA-PINN) with a finite difference method (FDM) solver: the parameterized NA-PINN takes the water-level difference, initial velocity, and time t as inputs and learns a solution manifold so that a single trained network serves as a data-free surrogate for the momentum conservation equation across all flow paths, while the FDM solver advances the mass conservation equation at each time step to ensure exact discrete mass conservation; verification on a six-tank gravity-driven draining scenario yields a water level mean absolute error of 7.85×10⁻⁵ m and a velocity mean absolute error of 3.21×10⁻³ m/s under the nominal condition with Δt = 1.

The work proposes SIUM, which models each modality subset's task representation as a Gaussian (mean carrying task information, variance measuring uncertainty from missing evidence), aligns subset means toward a full-modality anchor via an uncertainty-aware alignment loss that scales variance with their discrepancy, and adds an uncertainty ordering loss keeping subset variance above its supersets, achieving better segmentation than RFNet, mmFormer, M3AE, and DC-Seg across diverse missing-modality configurations on BraTS 2018 and 2020.

The work proposes SIUM, which models each modality subset's task representation as a Gaussian (mean carrying task information, variance measuring uncertainty from missing evidence), aligns subset means toward a full-modality anchor via an uncertainty-aware alignment loss that scales variance with their discrepancy, and adds an uncertainty ordering loss keeping subset variance above its supersets, achieving better segmentation than RFNet, mmFormer, M3AE, and DC-Seg across diverse missing-modality configurations on BraTS 2018 and 2020.

Through empirical analysis, this work identifies a "similarity bias" in dot-product self-attention that systematically overlooks heterogeneous item relations carrying causal signal for target prediction, and proposes PRISM, a module that calibrates attention through K Perspective Lenses operating in two complementary views, an Affinity View and a Contrast View, consistently outperforming state-of-the-art baselines on seven real-world benchmarks.

Through empirical analysis, this work identifies a "similarity bias" in dot-product self-attention that systematically overlooks heterogeneous item relations carrying causal signal for target prediction, and proposes PRISM, a module that calibrates attention through K Perspective Lenses operating in two complementary views, an Affinity View and a Contrast View, consistently outperforming state-of-the-art baselines on seven real-world benchmarks.