What question did this study set out to answer?

To design and evaluate LIMA, a system framework that performs fine-grained lineage tracing and intermediate computation reuse to eliminate redundancy within exploratory machine learning workflows.

June 9, 2021

LIMA: Fine-grained Lineage Tracing and Reuse in Machine Learning Systems

View Full Paper

Ask AI

Bookmark

Authors

APArnab Phani BRBenjamin Rath MBMatthias Böehm

Discussion

Member takes

Overview

Benchmarking evaluation demonstrates up to 12.4x speedups across exploratory machine learning pipelines, highlighting the efficiency of fine-grained lineage tracing and computational reuse.

Key Points

To design and evaluate LIMA, a system framework that performs fine-grained lineage tracing and intermediate computation reuse to eliminate redundancy within exploratory machine learning workflows.
Designed multi-level lineage tracing and reuse architectures with dedicated deduplication for program loops and functions inside machine learning systems.
Integrated fine-grained operational reuse with task parallelism, operator fusion, and versioning mechanisms to manage internal non-determinism.
Evaluated execution runtime, system overhead, and reuse efficiency across diverse exploratory machine learning pipelines featuring data cleaning, feature engineering, and hyper-parameter tuning.
Achieved runtime performance improvements of up to 12.4x across a variety of evaluated machine learning pipelines.
Maintained low tracing overhead while successfully eliminating fine-grained computational redundancy across full and partial program hierarchies.
Provided native versioning and experiment reproducibility across exploratory iterative cycles.

Cite This Study

Phani et al. (2021) studied this question.

synapsesocial.com/papers/6a11cf321d1aaf855556342a https://doi.org/10.1145/3448016.3452788

View Full Paper

Ask AI

Bookmark

June 9, 2021

LIMA: Fine-grained Lineage Tracing and Reuse in Machine Learning Systems

View Full Paper

Ask AI

Bookmark

Authors

APArnab Phani BRBenjamin Rath MBMatthias Böehm

Discussion

Member takes

Overview

Benchmarking evaluation demonstrates up to 12.4x speedups across exploratory machine learning pipelines, highlighting the efficiency of fine-grained lineage tracing and computational reuse.

Key Points

To design and evaluate LIMA, a system framework that performs fine-grained lineage tracing and intermediate computation reuse to eliminate redundancy within exploratory machine learning workflows.
Designed multi-level lineage tracing and reuse architectures with dedicated deduplication for program loops and functions inside machine learning systems.
Integrated fine-grained operational reuse with task parallelism, operator fusion, and versioning mechanisms to manage internal non-determinism.
Evaluated execution runtime, system overhead, and reuse efficiency across diverse exploratory machine learning pipelines featuring data cleaning, feature engineering, and hyper-parameter tuning.
Achieved runtime performance improvements of up to 12.4x across a variety of evaluated machine learning pipelines.
Maintained low tracing overhead while successfully eliminating fine-grained computational redundancy across full and partial program hierarchies.
Provided native versioning and experiment reproducibility across exploratory iterative cycles.

Cite This Study

Phani et al. (2021) studied this question.

synapsesocial.com/papers/6a11cf321d1aaf855556342a https://doi.org/10.1145/3448016.3452788

View Full Paper

Ask AI

Bookmark