機器學習已經發展多年，至今仍然持續進行多方研究，例如深度學習。 然而，隨著硬體的革新，傳統的機器學習算法可能無法在新型態的硬體上獲得最佳效率。 根據不同的硬體特性，本文提出了一種現代的系統架構來提高隨機森林算法的性能。基於相變化記憶體，我們發現讀寫操作之間的延遲存在著顯著的差異。因此，我們除了設置檢查點以防止不必要的寫入並且使用讀取來替換寫入的動作。 接下來，考慮到記憶體的讀取局部性，通過不同的實作方式實現出更適合新型態記憶體的記憶體配置。實驗顯示，我們所提出的系統架構最多可以提高隨機森林演算法70％的效能。

Machine learning has been accessible for a long time and is still a trend to keep developing, such as deep learning. Nevertheless, with the upgrade of hardware, the traditional algorithm of machine learning might no longer get the best efficiency. According to the different features of new generation hardware, this paper proposes a modern architecture to enhance the performance of the random forest algorithm. Based on phase change memory, we found the discrepancy on latency between read and write operations. Therefore, we set checkpoints to prevent unnecessary write operations and replace write with read operations. Next, with considering to read locality, modify the implementation by a different perspective. The results show that the proposed design can improve performance by 70%.

1. Introduction 5
2. Background 7
2.1 Phase Change Memory 7
2.2 Decision Tree and Random Forest 8
2.3 Motivation 10
3. NVM-friendly Random Forest 11
3.1 Overview 11
3.2 Check Point 12
3.3 Tracing Strategy 14
3.3.1 Naïve Solution 14
3.3.2 Bit Flag 15
3.4 Feature Base to Sample Base 17
4. Cache Analysis 19
4.1 Optimization of Required Cache Size 19
4.2 Lack of Cache 20
5. Evaluation 21
5.1 Experiment Setup 21
5.2 Performance 22
5.3 Coverage Rate of Cache 23
6. Conclusion 24
7. Reference 25

[1.] Xiangyu Dong and Yuan Xie. Adams: Adaptive mlc/slc phase-change memory design for file storage. In Proceedings of the 16th Asia and South Pacific Design Automation Conference, pages 31–36. IEEE Press, 2011.
[2.] Yuntan Fang, Huawei Li, and Xiaowei Li. Lifetime enhancement techniques for pcm-based image buffer in multimedia applications. IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 22(6):1450–1455, 2013.
[3.] Meikang Qiu, Zhi Chen, and Meiqin Liu. Low-power low-latency data allocation for hybrid scratch-pad memory. IEEE Embedded Systems Letters, 6(4):69–72, 2014.
[4.] T Nirschl, JB Philipp, TD Happ, Geoffrey W Burr, B Rajendran, M-H Lee, A Schrott, M Yang, M Breitwisch, C-F Chen, et al. Write strategies for 2 and 4-bit multi-level phase-change memory. In 2007 IEEE International Electron Devices Meeting, pages 461–464. IEEE, 2007.
[5.] Rich Caruana, Nikos Karampatziakis, and Ainur Yessenalina. An empirical evaluation of supervised learning in high dimensions. In Proceedings of the 25th international conference on Machine learning, pages 96–103. ACM, 2008.
[6.] Manuel Fern´andez-Delgado, Eva Cernadas, Sen´en Barro, and Dinani Amorim. Do we need hundreds of classifiers to solve real world classification problems? The Journal of Machine Learning Research, 15(1):3133–3181, 2014.
[7.] Sebastian Buschjager, Kuan-Hsun Chen, Jian-Jia Chen, and Katharina Morik. Realization of random forest for real-time evaluation through tree framing. In 2018 IEEE International Conference on Data Mining (ICDM), pages 19–28. IEEE, 2018.
[8.] Claudio Lucchese, Franco Maria Nardini, Salvatore Orlando, Raffaele Perego, Nicola Tonellotto, and Rossano Venturini. Quickscorer: Efficient traversal of large ensembles of decision trees. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, pages 383–387. Springer, 2017.
[9.] M Hosomi, H Yamagishi, T Yamamoto, K Bessho, Y Higo, K Yamane, H Yamada, M Shoji, H Hachino, C Fukumoto, et al. A novel nonvolatile memory with spin torque transfer magnetization switching: Spin-ram. In IEEE InternationalElectron Devices Meeting, 2005. IEDM Technical Digest., pages 459–462. IEEE, 2005.
[10.] E Chen, D Apalkov, Z Diao, A Driskill-Smith, D Druist, D Lottis, V Nikitin, X Tang, S Watts, S Wang, et al. Advances and future prospects of spin-transfer torque random access memory. IEEE Transactions on Magnetics, 46(6):1873–1878, 2010.
[11.] Prashant J Nair, Chiachen Chou, Bipin Rajendran, and Moinuddin K Qureshi. Reducing read latency of phase change memory via early read and turbo read. In 2015 IEEE 21st International Symposium on High Performance Computer Architecture (HPCA), pages 309–319. IEEE, 2015.
[12.] Hung-Sheng Chang, Yuan-Hao Chang, Yuan-Hung Kuan, Xiang-Zhi Huang, Tei-Wei Kuo, and Hsiang-Pang Li. Pattern-aware write-back strategy to minimize energy consumption of pcm-based storage systems. In 2016 5th Non-Volatile Memory Systems and Applications Symposium (NVMSA), pages 1–6. IEEE, 2016.
[13.] Ming-Chang Yang, Yuan-Hao Chang, and Che-Wei Tsao. Byte-addressable update scheme to minimize the energy consumption of pcm-based storage systems. ACM Transactions on Embedded Computing Systems (TECS), 15(3):55, 2016.

[14.] Jue Wang, Xiangyu Dong, Guangyu Sun, Dimin Niu, and Yuan Xie. Energy-efficient multi-level cell phase-change memory system with data encoding. In 2011 IEEE 29th International Conference on Computer Design (ICCD), pages 175–182. IEEE, 2011.
[15.] Jingtong Hu, Chun Jason Xue, Wei-Che Tseng, Yi He, Meikang Qiu, and Edwin H-M Sha. Reducing write activities on non-volatile memories in embedded cmps via data migration and recomputation. In Proceedings of the 47th Design Automation Conference, pages 350–355. ACM, 2010.
[16.] Moinuddin K Qureshi, Michele M Franceschini, Ashish Jagmohan, and Luis A Lastras. Preset: improving performance of phase change memories by exploiting asymmetry in write times. ACM SIGARCH Computer Architecture News, 40(3):380–391, 2012.
[17.] Meikang Qiu, Zhong Ming, Jiayin Li, Keke Gai, and Ziliang Zong. Phase-change memory optimization for green cloud with genetic algorithm. IEEE Transactions on Computers, 64(12):3528–3540, 2015.
[18.] Shimin Chen, Phillip B Gibbons, Suman Nath, et al. Rethinking database algorithms for phase change memory. In Cidr, pages 21–31, 2011.
[19.] Geoff Hulten and Pedro Domingos. Mining complex models from arbitrarily large databases in constant time. In Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining, pages 525–531. ACM, 2002.
[20.] John C Shafer, Rakesh Agrawal, and Manish Mehta. A scalable parallel classi er for data mining. In Proc. 22nd International Conference on VLDB, Mumbai, India, 1996.
[21.] Manish Mehta, Rakesh Agrawal, and Jorma Rissanen. Sliq: A fast scalable classifier for data mining. In International conference on extending database technology, pages 18–32. Springer, 1996.
[22.] Leo Breiman. Random forests. Machine learning, 45(1):5–32, 2001.
[23.] Marvin N Wright and Andreas Ziegler. ranger: A fast implementation of random forests for high dimensional data in c++ and r. arXiv preprint arXiv:1508.04409, 2015.
[24.] Dheeru Dua and Casey Graff. UCI machine learning repository, 2017.