.png){kind=icon}
Return
Distributed Variational Autoencoder Using Actor-Based Architecture
.png){kind=icon}
PRE
AI
DOI：10.1007/978-3-032-03711-4_15.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study presents a novel Variational Autoencoder (VAE) implementation using an actor-based distributed architecture with builtin fault tolerance mechanisms. The key innovation lies in transforming a traditional autoencoder into a variational autoencoder by implementing a specialized bottleneck actor that learns deviations from the normal distribution. The developed system utilizes a distributed actor model in which each component (encoder, decoder, and variational bottleneck) operates as an independent actor, enabling robust parallel processing and fault recovery. The implementation incorporates TensorFlow's distributed strategy for GPU utilization and includes a comprehensive fault tolerance system that can handle actor failures with a 0.5% failure rate. The results demonstrated successful training on the utilized dataset, achieving a reconstruction loss of 0.089 and a Kullback-Leibler (KL) divergence loss of 0.076 after 50 epochs. The system maintained 99.5% uptime during training, automatically recovering from simulated actor failures without significant performance degradation.
Keywords:
Variational Autoencoder
Actor Model
Distributed Systems
Fault Tolerance
Journal
A
IF:
0
Papers:
27
Citations:
0
Organization
Cited Papers
ResNet Autoencoders for Unsupervised Feature Learning From High-Dimensional Data: Deep Models Resistant to Performance Degradation.png)
IEEE ACCESS
OAAI
IEEE ACCESSIF3.6
OAAIShareSaveSelf-Supervised Remote Sensing Feature Learning: Learning Paradigms, Challenges, and Future WorksOAAI
OAAIShareSaveShareSaveShareSave
View more