Today, Intel unveiled its 4th Gen Intel Xeon Scalable processors. TruEra is a proud Gold partner of Intel, as well as a member of the Intel Disruptor Initiative.
Why processor architecture matters in ML
Training and inference machine learning models are notoriously compute intensive. Testing and analyzing those models, as TruEra does, are also highly compute intensive. In particular, the state of the art techniques for explaining models, critical to evaluating models, can be orders of magnitude more expensive than inference. For these reasons, accelerating the performance of AI explainability can significantly improve the velocity with which data scientists evaluate and debug their models. As a result, the underlying processor architecture can be critical.
What we like about the new 4th Gen Intel Xeon Scalable processors:
- Faster query throughput. Accelerators boost query throughput for more responsive analytics. Also, Intel Xeon Scalable processors dynamically balance loads across multiple cores.
- Higher energy efficiency, to reduce environmental impact. Ok, this one is not about AI performance, but about more efficient processor utilization and reduced energy footprint. As organizations embrace AI and digital transformation, we also want to make sure that we are doing so while minimizing harm to the environment.
TruEra is working closely with Intel through the Intel Disruptor Initiative to optimize its solutions for Intel processors and technologies, including 4th Gen Intel Xeon Scalable processors. For example, the collaboration team recently validated TruEra’s ability to integrate with Intel architecture-optimized frameworks such as Intel Extension for Sci-kit Learn, Intel Extension for PyTorch, Intel Distribution of Modin and the Intel optimizations included in XGBoost—all of which help to accelerate the end-to-end AI pipeline.
1.45x performance improvement on 4th Gen Intel Xeon Scalable processors
TruEra provides ML testing, evaluation, and monitoring solutions, in what we call “AI Quality management.” Data scientists and ML engineers use TruEra to rapidly assess and debug their models, which allow them to create high performance ML models more quickly, as well as keep models in production longer. To assess the impact of the new Intel processors on TruEra’s ability to rapidly evaluate ML models, we set up benchmarking tests on the standard adult census datasets.
For a model that predicted an individual’s income based on 1980 census data about the individual’s household information, we used TruEra to test the performance of the model, identify model issues including issues with fairness and performance, and to explain how the income prediction model worked (i.e., conduct a feature influence analysis and perform root cause analysis). We evaluated the performance of TruEra on both 3rd Gen Intel Xeon Scalable processors and the new 4th Gen Intel Xeon Scalable processors.
We discovered major performance improvements when using the 4th Gen processors. The 4th Gen Intel Xeon Scalable processors provide 1.45x improvement in performance against the benchmark. While the performance on the 3rd Gen processors was already impressive, this allows users on the 4th Gen Intel Xeon Scalable processors to explain millions of predictions accurately in a matter of seconds, and allows them to debug issues with their models faster and more accurately than ever before.
Beyond this benchmarking initiative, the collaboration between TruEra and Intel continues, with plans to develop several joint pipelines that showcase optimized, end-to-end model training, deployment and explainability workflows.
AI Quality management – better testing, debugging, and monitoring
TruEra’s AI Quality Management solutions provide automated testing across a broad range of model quality areas, such as accuracy, reliability, stability and fairness/bias. This accelerates model development times, improves model performance and makes it easy for models to get approved and be monitored in production. Using TruEra’s solutions, resolving a perplexing performance problem may take only days instead of weeks.
Contact TruEra to learn how you can get started testing, analyzing, and monitoring your ML models, to drive better ML performance.
Interested in learning more about how to make machine learning models fairer? Check out the whitepaper – Fairness in Machine Learning, from Concept to Reality
Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries.
