Parameter Database : Data-centric Synchronization for Scalable Machine Learning
We propose a new data-centric synchronization framework for carrying out of machine learning (ML) tasks in a distributed environment. Our framework exploits the iterative nature of ML algorithms and relaxes the application agnostic bulk synchronization parallel (BSP) paradigm that has previously been used for distributed machine learning. Data-centric synchronization complements function-centric synchronization based on using stale updates to increase the throughput of distributed ML computations. Experiments to validate our framework suggest that we can attain substantial improvement over BSP while guaranteeing sequential correctness of ML tasks.
Code (0)
등록된 구현이 없습니다.
Tasks
BIG-bench Machine LearningSimilar Papers 제목 키워드 기반
Dual Audio-Centric Modality Coupling for Talking Head Generation
The generation of audio-driven talking head videos is a key challenge in computer vision and graphics, with applications in virtual avatars and digital media. Traditional approaches often struggle with capturing the comp…
NeRFTalking Head Generationtext-to-speechText to SpeechScalable global state synchronization of discrete-time double integrator multi-agent systems with input saturation via linear protocol (Completed Version)
This paper studies scalable global state synchronization of discrete-time double integrator multi-agent systems in presence of input saturation based on localized information exchange. A scale-free collaborative linear d…
Scalable Resetting Algorithms for Synchronization of Pulse-Coupled Oscillators over Rooted Directed Graphs
We study the problem of robust global synchronization of pulse-coupled oscillators (PCOs) over directed graphs. It is known that when the digraphs are strongly connected, global synchronization can be achieved by using a…
WorldWander: Bridging Egocentric and Exocentric Worlds in Video Generation
Recent advances in video world models enable interactive environments with free navigation, making translation between first-person (egocentric) and third-person (exocentric) perspectives increasingly important. However,…
Video GenerationA Database-Driven Framework for 3D Level Generation with LLMs
Procedural Content Generation for 3D game levels faces challenges in balancing spatial coherence, navigational functionality, and adaptable gameplay progression across multi-floor environments. This paper introduces a no…