机器学习论文怎么读?大神给你三步妙招

本文介绍了如何高效阅读机器学习论文,包括在Reddit和arxiv-sanity.com等平台寻找论文,以及通过三步法(浏览理解、严谨阅读、关注数学细节)深入理解论文内容,强调了明确阅读目标和实践的重要性。
摘要由CSDN通过智能技术生成

读论文绝对是个技术活,无论是机器学习,还是密码学分布式共识算法,或者神经网络什么的,要想真正对计算机学科的某个领域有专业的理解,你必须紧跟这个方面最新研究。对相关专业有专业严谨的评估,是我们必备的能力。

只要有耐心并多加练习,以及很多很多的咖啡,你一定可以掌握高效读论文的方法。

我每周通常会每周读10-12篇研究论文,以紧跟计算机领域的潮流。随着时间推移,读论文的能力也越来越好。

找机器学习论文时最爱的资源是Reddit,在他的机器学习板块,每天都有好多人发表很有趣的论文,还有很酷的周报,显示你这周读了哪些论文,哪些最让你感兴趣。另外还有个很棒的网站arxiv-sanity.com,由大佬Andrei Karpathy开发,它会帮你寻找aRxiv上你最感兴趣的或者最热的论文,此外,谷歌和DeepMind也会在各自的网站上发表研究成果,并且在《nature》这样的期刊上也很容易找到顶级的论文。

那么,应该如何阅读论文?

最好带着目的去读,比如学习更好的激活函数,或者能帮我使用注意机制的概率模型等等。等明确目标后,就很容易确定读论文的策略,使我们直奔目标。只是从数学上看懂机器学习论文并不是我们的理想目标,能让自己有耐心有动力读下去更为重要。

下面是我读论文的三步矫正法

  • 第一步:仅浏览论文,理解论文大意。

首先看论文标题,要是觉得有意思,好了,可以开始了。然后看摘要,摘要是论文中研究工作的简短和独立的总结,可以用他作为文章的综述。然后仔细阅读段落和小节的题目而不管剩下的部分,主要是忽略数学部分。然后读末尾的总结,可能会扫一眼引用的论文,看看有没有已经看过的。在第一步,我会假设数学部分是正确的,这步的主要目的是弄懂作者的目的,论文的主要贡献、论文想解决什么问题以及我是否真的感兴趣。

在完成这一步后,我会回头看看别人怎么评价论文,把我的初步想法和大家对比一下。

这一步的主要目的就是确保论文是我想看且需要的,人生苦短嘛,毕竟还有好多东西要看。

  • 第二步:重新阅读

更严谨的读论文,并做笔记。我会读全部的内容,并尽力高度理解数学部分,并尽力评估实验结果,是否重复?结果有明显的证据支撑吗?然后去GitHub上找找是否有已实现的相关代码,自己读一遍然后运行一遍,复现结果。通常代码中的注释会帮助我理解,此外我还会搜寻其他有助于我理解代码的东西,比如文章、总结、教程等。通常一篇很火的论文会有很多人分享对他的见解。

  • 第三步:关注数学细节

拿出笔来自己解析论文里的公式,同时也会借助百度帮我理解公式。并用论文给出的参数尽可能的复现代码。

 

 

25篇机器学习经典论文合集,有需要欢迎积分自取 Efficient sparse coding algorithms论文附有代码 [1] Zheng S, Kwok J T. Follow the moving leader in deep learning[C]//Proceedings of the 34th International Conference on Machine Learning-Volume 70. JMLR. org, 2017: 4110-4119. [2] Kalai A, Vempala S. Efficient algorithms for online decision problems[J]. Journal of Computer and System Sciences, 2005, 71(3): 291-307. [3] Kingma, D. and Ba, J. Adam: A method for stochastic optimization. In Proceedings of the International Conference for Learning Representations, 2015. [4] Lee H, Battle A, Raina R, et al. Efficient sparse coding algorithms[C]//Advances in neural information processing systems. 2007: 801-808. [5] Fan J, Ding L, Chen Y, et al. Factor Group-Sparse Regularization for Efficient Low-Rank Matrix Recovery[J]. 2019. [6] Z. Lai, Y. Chen, J. Wu, W. W. Keung, and F. Shen, “Jointly sparse hashing for image retrieval,” IEEE Transactions on Image Processing, vol. 27, no. 12, pp. 6147–6158, 2018. [7] Z. Zhang, Y. Chen, and V. Saligrama, “Efficient training of very deep neural networks for supervised hashing,” in Proc. IEEE Int. Conf. Computer Vision and Pattern Recognition, 2016, pp. 1487–1495. [8] Wei-Shi Zheng, Shaogang Gong, Tao Xiang. Person re-identification by probabilistic relative distance comparison[C]// CVPR 2011. IEEE, 2011. [9] Liao S, Hu Y, Zhu X, et al. Person re-identification by local maximal occurrence representation and metric learning[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 2197-2206. [10] Liu X, Li H, Shao J, et al. Show, tell and discriminate: Image captioning by self-retrieval with partially labeled data[C]//Proceedings of the European Conference on Computer Vision (ECCV). 2018: 338-354. [11] Yao T, Pan Y, Li Y, et al. Exploring visual relationship for image captioning[C]//Proceedings of the European conference on computer vision (ECCV). 2018: 684-699. [12] Chao Dong, Chen Change Loy, Kaiming He, and Xiaoou Tang., ”Image Super-Resolution Using Deep Convolutional Networks, ” IEEE Transactions on Pattern Analysis and Machine Intelligence, Preprint, 2015. [13] M. D. Zeiler, D. Krishnan, Taylor, G. W., and R. Fergus, "Deconvolutional networks," in Proc. IEEE Comput. Soc. Conf. Comput. Vision Pattern Recog., 2010, pp. 2528-2535. [14] Girshick R, Donahue J, Darrell T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2014: 580-587. [15] Girshick R . Fast R-CNN[J]. Computer Science, 2015. [16] Joseph Redmon, Santosh Divvala, Ross Girshick, et al. You Only Look Once: Unified, Real-Time Object Detection[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2016. [17] LeCun Y, Bottou L, Bengio Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11): 2278-2324. [18] Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. science, 2006, 313(5786): 504-507. [19] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks[C]//Advances in neural information processing systems. 2012: 1097-1105. [20] Zeiler M D, Fergus R. Visualizing and understanding convolutional networks[C]//European conference on computer vision. Springer, Cham, 2014: 818-833. [21] Szegedy C, Liu W, Jia Y, et al. Going deeper with convolutions[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2015: 1-9. [22] Wu, Y., & He, K. (2018). Group normalization. In Proceedings of the European Conference on Computer Vision (ECCV) (pp. 3-19). [23] Goodfellow I,Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[C]//Advances in Neural Information Processing Systems. 2014: 2672-2680. [24] Tran, L., Yin, X., & Liu, X. (2017). Disentangled representation learning gan for pose-invariant face recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition (pp. 1415-1424). [25] Pu, Y., Gan, Z., Henao, R., Yuan, X., Li, C., Stevens, A., & Carin, L. (2016). Variational autoencoder for deep learning of images, labels and captions. In Advances in neural information processing systems (pp. 2352-2360).
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值