Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Existing multi-DNN inference systems support only a single model or a few sparse variants per task, which impedes matching efficiency and results in high Service Level Objective violation rates.limitation 85%
• Model stitching creates model variants by recombining subgraphs from sparse models without requiring re-training.method 90%
• The SparseLoom demonstrator system shows that model stitching can be deployed to System-on-Chips (SoCs).result 90%
• SparseLoom reduces Service Level Objective (SLO) violation rates by up to 74% compared to state-of-the-art multi-DNN inference systems.result 90%
• SparseLoom improves throughput by up to 2.31x compared to state-of-the-art multi-DNN inference systems.result 90%
• SparseLoom lowers memory overhead by an average of 28% compared to state-of-the-art multi-DNN inference systems.result 90%
• Modern edge applications increasingly require multi-DNN inference systems to execute tasks on heterogeneous processors.market 80%
• Performance gains in multi-DNN systems come from both concurrent execution and from matching each model to the most suited accelerator.technical 85%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card