다중 로봇 시스템에서 각 로봇이 자신의 데이터만 사용하여 NeRF 모델을 학습하고, 상대 자세를 최적화하면서 협력적으로 전체 장면을 재구성할 수 있다.
Di-NeRF enables a group of robots to collaboratively optimize the parameters of a Neural Radiance Field (NeRF) in a distributed manner, without explicitly sharing visual data. It also jointly optimizes the relative poses of the robots, allowing for accurate 3D reconstruction with less accurate initial relative camera poses.
Mercury는 비잔틴 합의 프로토콜의 복원력 임계값을 자율적으로 조정하여 작은 쿼럼을 형성할 수 있게 함으로써 거래 주문 속도를 크게 높일 수 있다.
Mercury, a self-optimizing protocol transformation, adapts the resilience threshold of a Byzantine Fault-Tolerant (BFT) consensus protocol to enable the emergence of smaller quorums for faster transaction ordering, while preserving the standard safety and liveness guarantees.
The authors propose a novel variance-reduced gradient estimator that combines the advantages of 2-point and 2d-point gradient estimators to address the trade-off between convergence rate and sampling cost in distributed zeroth-order optimization for smooth nonconvex functions.
분산 최적화 문제를 해결하기 위해 팽창 좌표계에서의 에너지 보존 관점을 도입하여 O(1/t^2-β) 수렴 속도를 달성하는 새로운 분산 가속 경사 흐름 알고리즘을 제안한다.
The proposed distributed accelerated gradient flow algorithm achieves a convergence rate of O(1/t^(2-β)) for smooth convex optimization problems, which is near-optimal in the distributed setting.
Multiple robotic agents can collaboratively learn a comprehensive neural radiance field (NeRF) representation of a scene by sharing only their learned network weights, without transferring raw sensor data, enabling efficient multi-agent perception.
An approach to assess and correct time synchronization errors between rigidly mounted sensors based on their rotational motion.
컨소시엄 블록체인에서 대용량 데이터를 효율적으로 저장하고 관리할 수 있는 분산 저장 시스템 DBNode를 제안한다.