Data reduction through optimized scalar quantization for more compact neural networks
Raw data generation for several existing and planned large physics experiments now exceeds TB/s rates, generating untenable data sets in very little time. Those data often demonstrate high dimensionality while containing limited information. Meanwhile, Machine Learning algorithms are now becoming an...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2022-09-01
|
| Series: | Frontiers in Physics |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fphy.2022.957128/full |
| _version_ | 1797995359707856896 |
|---|---|
| author | Berthié Gouin-Ferland Ryan Coffee Audrey C. Therrien |
| author_facet | Berthié Gouin-Ferland Ryan Coffee Audrey C. Therrien |
| author_sort | Berthié Gouin-Ferland |
| collection | DOAJ |
| description | Raw data generation for several existing and planned large physics experiments now exceeds TB/s rates, generating untenable data sets in very little time. Those data often demonstrate high dimensionality while containing limited information. Meanwhile, Machine Learning algorithms are now becoming an essential part of data processing and data analysis. Those algorithms can be used offline for post processing and post data analysis, or they can be used online for real time processing providing ultra low latency experiment monitoring. Both use cases would benefit from data throughput reduction while preserving relevant information: one by reducing the offline storage requirements by several orders of magnitude and the other by allowing ultra fast online inferencing with low complexity Machine Learning models. Moreover, reducing the data source throughput also reduces material cost, power and data management requirements. In this work we demonstrate optimized nonuniform scalar quantization for data source reduction. This data reduction allows lower dimensional representations while preserving the relevant information of the data, thus enabling high accuracy Tiny Machine Learning classifier models for online fast inferences. We demonstrate this approach with an initial proof of concept targeting the CookieBox, an array of electron spectrometers used for angular streaking, that was developed for LCLS-II as an online beam diagnostic tool. We used the Lloyd-Max algorithm with the CookieBox dataset to design an optimized nonuniform scalar quantizer. Optimized quantization lets us reduce input data volume by 69% with no significant impact on inference accuracy. When we tolerate a 2% loss on inference accuracy, we achieved 81% of input data reduction. Finally, the change from a 7-bit to a 3-bit input data quantization reduces our neural network size by 38%. |
| first_indexed | 2024-04-11T09:59:13Z |
| format | Article |
| id | doaj.art-34b80e065ef04a2e88dac9add8a975e9 |
| institution | Directory Open Access Journal |
| issn | 2296-424X |
| language | English |
| last_indexed | 2024-04-11T09:59:13Z |
| publishDate | 2022-09-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Physics |
| spelling | doaj.art-34b80e065ef04a2e88dac9add8a975e92022-12-22T04:30:27ZengFrontiers Media S.A.Frontiers in Physics2296-424X2022-09-011010.3389/fphy.2022.957128957128Data reduction through optimized scalar quantization for more compact neural networksBerthié Gouin-Ferland0Ryan Coffee1Audrey C. Therrien2Interdisciplinary Institute for Technological Innovation - 3IT, Sherbrooke, QC, CanadaSLAC National Accelerator Laboratory, Menlo Park, CA, United StatesInterdisciplinary Institute for Technological Innovation - 3IT, Sherbrooke, QC, CanadaRaw data generation for several existing and planned large physics experiments now exceeds TB/s rates, generating untenable data sets in very little time. Those data often demonstrate high dimensionality while containing limited information. Meanwhile, Machine Learning algorithms are now becoming an essential part of data processing and data analysis. Those algorithms can be used offline for post processing and post data analysis, or they can be used online for real time processing providing ultra low latency experiment monitoring. Both use cases would benefit from data throughput reduction while preserving relevant information: one by reducing the offline storage requirements by several orders of magnitude and the other by allowing ultra fast online inferencing with low complexity Machine Learning models. Moreover, reducing the data source throughput also reduces material cost, power and data management requirements. In this work we demonstrate optimized nonuniform scalar quantization for data source reduction. This data reduction allows lower dimensional representations while preserving the relevant information of the data, thus enabling high accuracy Tiny Machine Learning classifier models for online fast inferences. We demonstrate this approach with an initial proof of concept targeting the CookieBox, an array of electron spectrometers used for angular streaking, that was developed for LCLS-II as an online beam diagnostic tool. We used the Lloyd-Max algorithm with the CookieBox dataset to design an optimized nonuniform scalar quantizer. Optimized quantization lets us reduce input data volume by 69% with no significant impact on inference accuracy. When we tolerate a 2% loss on inference accuracy, we achieved 81% of input data reduction. Finally, the change from a 7-bit to a 3-bit input data quantization reduces our neural network size by 38%.https://www.frontiersin.org/articles/10.3389/fphy.2022.957128/fullmachine learning - MLneural networkquantizationclassificationfree electron lasersdata acquisition |
| spellingShingle | Berthié Gouin-Ferland Ryan Coffee Audrey C. Therrien Data reduction through optimized scalar quantization for more compact neural networks Frontiers in Physics machine learning - ML neural network quantization classification free electron lasers data acquisition |
| title | Data reduction through optimized scalar quantization for more compact neural networks |
| title_full | Data reduction through optimized scalar quantization for more compact neural networks |
| title_fullStr | Data reduction through optimized scalar quantization for more compact neural networks |
| title_full_unstemmed | Data reduction through optimized scalar quantization for more compact neural networks |
| title_short | Data reduction through optimized scalar quantization for more compact neural networks |
| title_sort | data reduction through optimized scalar quantization for more compact neural networks |
| topic | machine learning - ML neural network quantization classification free electron lasers data acquisition |
| url | https://www.frontiersin.org/articles/10.3389/fphy.2022.957128/full |
| work_keys_str_mv | AT berthiegouinferland datareductionthroughoptimizedscalarquantizationformorecompactneuralnetworks AT ryancoffee datareductionthroughoptimizedscalarquantizationformorecompactneuralnetworks AT audreyctherrien datareductionthroughoptimizedscalarquantizationformorecompactneuralnetworks |