Research Feed Page 43

Node4All introduces a general-purpose method for learning graph node representations that works across arbitrary datasets without needing dataset-specific optimization or hyperparameter tuning.

The paper introduces Epistemic Byzantine Fault Tolerance to prevent protocol compliant validators from reaching consensus on semantically invalid state transitions.

CADAQUES introduces a budget-aware architecture that treats query costs as first-class primitives to manage resource expenditure in autonomous discovery loops.

The paper introduces a compositional framework that splits robot tasks into standard free-space planning and learned policies for contact-rich segments to improve reliability and generalization.

The paper introduces a coupled egocentric control system that automates torso and base movements based on operator body motion, reducing manual input requirements and kinematic conflicts.

The paper introduces GeoChrono, a multi-modal large language model designed to track and reason about geographic evolution over time using a new cognitive hierarchy and dataset.

Orbis 2 uses a two level architecture to separate long term spatial reasoning from pixel level detail generation for better driving simulation.

JoyNexus is a service-oriented framework that enables multi-tenant post-training for Vision-Language-Action models by decoupling compute resources from model execution.

The paper introduces optimized computational methods for the R and Q steps in IKPLS algorithms to achieve significant speedups on modern multi-processor hardware.

The authors developed ShellFlow, a transformer based generative model that learns particle collision patterns directly from ATLAS data without relying on traditional Monte Carlo simulations.

The paper introduces a behaviour-conditioned neural process model that improves short-term residential energy demand forecasting by incorporating inferred consumption patterns.

The paper introduces a cluster-aware matching framework that uses Laplacian regularization to align point clouds and data clusters more consistently than standard independent methods.

Xiaomi-Robotics-1 introduces a large-scale vision-language-action model and a scalable auto-labeling pipeline to overcome data bottlenecks in robotics, achieving strong performance on robot benchmarks.

FVAttn accelerates video generation by dynamically balancing computational workloads across GPUs to fix inefficiencies caused by sparse attention mechanisms.

The paper investigates whether current multimodal large language models exercise active observation by introducing a benchmark called ActiveVision that makes this capability measurable.

The paper evaluates open-weight Large Language Models on converting unstructured connected and autonomous vehicle vulnerability descriptions into structured threat information expressions.

The paper presents a thermodynamic computing paradigm that utilizes stochastic analog superconducting circuits to execute probabilistic machine learning models directly in physical hardware.

The paper introduces Recursive Harness Self-Improvement to iteratively refine user-constructed harnesses using pairwise feedback over revision history, reducing inference cost for opus-4.8.

RecGPT-V3 is a recommender system architecture that improves recommendation performance while cutting serving resource consumption and user-modeling computation through structured behavior compression and latent intent reasoning.

The paper introduces S1-Omni, a unified multimodal reasoning model that addresses the fragmentation of existing AI for Science architectures across scientific understanding, prediction, and generation.