DynaFlow: Transparent and Flexible Intra-Device Parallelism via Programmable Operator Scheduling

2026-05-22T08:52:21Zc5a25ad7379ff5d8956e4378f4fe45c3ff2595940e15ee86ef928bb77d275655
collective-communicationcustom-cryptographydata-leakageedge-computingintegrity-verificationmachine-learning-infrastructuremodel-confidentialitynetwork-multicastprivilege-escalationsecure-computationside-channel

What happened

The document is an arXiv feed of systems and ML-infrastructure papers. Several submissions have direct security/privacy implications: (1) LiveR enables live streaming of model state across high-bandwidth interconnects to perform topology reconfiguration without stop-and-restart, which reduces downtime but increases risk of confidentiality and integrity exposure of model weights and optimizer state if interconnects or transfer protocols are compromised. (2) The Secure Parallel Determinant Computation (SPDC) framework proposes a custom lightweight obfuscation/encryption scheme (Composite Element

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_dc
Record identifier
c5a25ad7379ff5d8956e4378f4fe45c3ff2595940e15ee86ef928bb77d275655
Enrichment time
2026-05-22T08:52:21Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.