Top 20 Recent Research Papers on Machine Learning and Deep Learning

Machine learning, especially its subfield of Deep Learning, had many amazing advances in the recent years, and important research papers may lead to breakthroughs in technology that get used by billions of people. The research in this field is developing very quickly and to help our readers monitor the progress we present the list of most important recent scientific papers published since 2014.
The criteria we used to select the 20 top papers are by using citation counts from three academic sources: scholar.google.com; academic.microsoft.com; and  semanticscholar.org. Since the number of citations varied among sources and are estimated, we listed the results from academic.microsoft.com which is slightly lower than others.

For each paper we also give the year it was published, a Highly Influential Citation count (HIC) and Citation Velocity (CV) measures provided by  semanticscholar.org.  HIC that presents how publications build upon and relate to each other is result of identifying meaningful citations. CV is the weighted average number of citations per year over the last 3 years. For some references, where CV is zero that means it was blank or not shown by semanticscholar.org.

Most (but not all) of these 20 papers, including the top 8, are on the topic of Deep Learning. However, we see strong diversity - only one author (Yoshua Bengio) has 2 papers, and the papers were published in many different venues: CoRR (3), ECCV (3), IEEE CVPR (3), NIPS (2), ACM Comp Surveys, ICML, IEEE PAMI, IEEE TKDE, Information Fusion, Int. J. on Computers & EE, JMLR, KDD, and Neural Networks. The top two papers have by far the highest citation counts than the rest. Note that the second paper is only published last year. Read (or re-read them) and learn about the latest advances.

1. Dropout: a simple way to prevent neural networks from overfitting, by Hinton, G.E., Krizhevsky, A., Srivastava, N., Sutskever, I., & Salakhutdinov, R. (2014). Journal of Machine Learning Research, 15, 1929-1958. (cited 2084 times, HIC: 142 , CV: 536).
   
   Summary: The key idea is to randomly drop units (along with their connections) from the neural network during training. This prevents units from co-adapting too much. This significantly reduces overfitting and gives major improvements over other regularization methods
2. Deep Residual Learning for Image Recognition, by He, K., Ren, S., Sun, J., & Zhang, X. (2016). CoRR, abs/1512.03385. (cited 1436 times, HIC: 137 , CV: 582).
   Summary: We present a residual learning framework to ease the training of deep neural networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth.
3. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift, by Sergey Ioffe, Christian Szegedy (2015) ICML. (cited 946 times, HIC: 56 , CV: 0).
   Summary: Training Deep Neural Networks is complicated by the fact that the distribution of each layer's inputs changes during training, as the parameters of the previous layers change.  We refer to this phenomenon as internal covariate shift, and address the problem by normalizing layer inputs.  Applied to a state-of-the-art image classification model, Batch Normalization achieves the same accuracy with 14 times fewer training steps, and beats the original model by a significant margin.
4. Large-Scale Video Classification with Convolutional Neural Networks , by Fei-Fei, L., Karpathy, A., Leung, T., Shetty, S., Sukthankar, R., & Toderici, G. (2014). IEEE Conference on Computer Vision and Pattern Recognition (cited 865 times, HIC: 24 , CV: 239)
   Summary: Convolutional Neural Networks (CNNs) have been established as a powerful class of models for image recognition problems. Encouraged by these results, we provide an extensive empirical evaluation of CNNs on large-scale video classification using a new dataset of 1 million YouTube videos belonging to 487 classes .
5. Microsoft COCO: Common Objects in Context , by Belongie, S.J., Dollár, P., Hays, J., Lin, T., Maire, M., Perona, P., Ramanan, D., & Zitnick, C.L. (2014). ECCV. (cited 830 times, HIC: 78 , CV: 279) Summary: We present a new dataset with the goal of advancing the state-of-the-art in object recognition by placing the question of object recognition in the context of the broader question of scene understanding. Our dataset contains photos of 91 objects types that would be easily recognizable by a 4 year old. Finally, we provide baseline performance analysis for bounding box and segmentation detection results using a Deformable Parts Model.
6. Learning deep features for scene recognition using places database , by Lapedriza, À., Oliva, A., Torralba, A., Xiao, J., & Zhou, B. (2014). NIPS. (cited 644 times, HIC: 65 , CV: 0)
   Summary: We introduce a new scene-centric database called Places with over 7 million labeled pictures of scenes. We propose new methods to compare the density and diversity of image datasets and show that Places is as dense as other scene datasets and has more diversity.
7. Generative adversarial nets, by Bengio, Y., Courville, A.C., Goodfellow, I.J., Mirza, M., Ozair, S., Pouget-Abadie, J., Warde-Farley, D., & Xu, B. (2014) NIPS. (cited 463 times, HIC: 55 , CV: 0)
   Summary: We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G.
8. High-Speed Tracking with Kernelized Correlation Filters, by Batista, J., Caseiro, R., Henriques, J.F., & Martins, P. (2015). CoRR, abs/1404.7584. (cited 439 times, HIC: 43 , CV: 0)
   Summary: In most modern trackers,  to cope with natural image changes, a classifier is typically trained with translated and scaled sample patches. We propose an analytic model for datasets of thousands of translated patches. By showing that the resulting data matrix is circulant, we can diagonalize it with the discrete Fourier transform, reducing both storage and computation by several orders of magnitude.
9. A Review on Multi-Label Learning Algorithms, by  Zhang, M., & Zhou, Z. (2014). IEEE TKDE,  (cited 436 times, HIC: 7 , CV: 91)
   Summary:  This paper aims to provide a timely review on multi-label learning studies the problem where each example is represented by a single instance while associated with a set of labels simultaneously.
10. How transferable are features in deep neural networks, by Bengio, Y., Clune, J., Lipson, H., & Yosinski, J. (2014) CoRR, abs/1411.1792. (cited 402 times, HIC: 14 , CV: 0)
    Summary: We experimentally quantify the generality versus specificity of neurons in each layer of a deep convolutional neural network and report a few surprising results. Transferability is negatively affected by two distinct issues: (1) the specialization of higher layer neurons to their original task at the expense of performance on the target task, which was expected, and (2) optimization difficulties related to splitting networks between co-adapted neurons, which was not expected.
11. Do we need hundreds of classifiers to solve real world classification problems, by Amorim, D.G., Barro, S., Cernadas, E., & Delgado, M.F. (2014).  Journal of Machine Learning Research (cited 387 times, HIC: 3 , CV: 0)
    Summary: We evaluate 179 classifiers arising from 17 families (discriminant analysis, Bayesian, neural networks, support vector machines, decision trees, rule-based classifiers, boosting, bagging, stacking, random forests and other ensembles, generalized linear models, nearest-neighbors, partial least squares and principal component regression, logistic and multinomial regression, multiple adaptive regression splines and other methods). We use 121 data sets from UCI data base to study the classifier behavior, not dependent on the data set collection. The winners are the random forest (RF) versions implemented in R and accessed via caret) and the SVM with Gaussian kernel implemented in C using LibSVM.
12. Knowledge vault: a web-scale approach to probabilistic knowledge fusion, by Dong, X., Gabrilovich, E., Heitz, G., Horn, W., Lao, N., Murphy, K., ... & Zhang, W. (2014, August). In Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining  ACM. (cited 334 times, HIC: 7 , CV: 107).
    Summary: We introduce Knowledge Vault, a Web-scale probabilistic knowledge base that combines extractions from Web content (obtained via analysis of text, tabular data, page structure, and human annotations) with prior knowledge derived from existing knowledge repositories for constructing knowledge bases. We employ supervised machine learning methods for fusing  distinct information sources. The Knowledge Vault is substantially bigger than any previously published structured knowledge repository, and features a probabilistic inference system that computes calibrated probabilities of fact correctness.
13. Scalable Nearest Neighbor Algorithms for High Dimensional Data, by Lowe, D.G., & Muja, M. (2014). IEEE Trans. Pattern Anal. Mach. Intell., (cited 324 times, HIC: 11 , CV: 69).
    Summary: We propose new algorithms for approximate nearest neighbor matching and evaluate and compare them with previous algorithms.  In order to scale to very large data sets that would otherwise not fit in the memory of a single machine, we propose a distributed nearest neighbor matching framework that can be used with any of the algorithms described in the paper.
14. Trends in extreme learning machines: a review, by Huang, G., Huang, G., Song, S., & You, K. (2015).  Neural Networks,  (cited 323 times, HIC: 0 , CV: 0)
    Summary: We aim to report the current state of the theoretical research and practical advances on Extreme learning machine (ELM). Apart from classification and regression, ELM has recently been extended for clustering, feature selection, representational learning and many other learning tasks.  Due to its remarkable efficiency, simplicity, and impressive generalization performance, ELM have been applied in a variety of domains, such as biomedical engineering, computer vision, system identification, and control and robotics.
15. A survey on concept drift adaptation, by Bifet, A., Bouchachia, A., Gama, J., Pechenizkiy, M., & Zliobaite, I.  ACM Comput. Surv., 2014 , (cited 314 times, HIC: 4 , CV: 23)
    Summary: This work aims at providing a comprehensive introduction to the concept drift adaptation that refers to an online supervised learning scenario when the relation between the input data and the target variable changes over time.
16. Multi-scale Orderless Pooling of Deep Convolutional Activation Features, by Gong, Y., Guo, R., Lazebnik, S., & Wang, L. (2014).  ECCV(cited 293 times, HIC: 23 , CV: 95)
    Summary:  To improve the invariance of CNN activations without degrading their discriminative power, this paper presents a simple but effective scheme called multi-scale orderless pooling (MOP-CNN).
17. Simultaneous Detection and Segmentation, by Arbeláez, P.A., Girshick, R.B., Hariharan, B., & Malik, J. (2014) ECCV , (cited 286 times, HIC: 23 , CV: 94)
    Summary: We aim to detect all instances of a category in an image and, for each instance, mark the pixels that belong to it. We call this task Simultaneous Detection and Segmentation (SDS).
18. A survey on feature selection methods, by Chandrashekar, G., & Sahin, F.  Int. J. on Computers & Electrical Engineering, (cited 279 times, HIC: 1 , CV: 58)
    Summary: Plenty of feature selection methods are available in literature due to the availability of data with hundreds of variables leading to data with very high dimension.
19. One Millisecond Face Alignment with an Ensemble of Regression Trees, by Kazemi, Vahid, and Josephine Sullivan, Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition 2014, (cited 277 times, HIC: 15 , CV: 0)
    Summary: This paper addresses the problem of Face Alignment for a single image. We show how an ensemble of regression trees can be used to estimate the face's landmark positions directly from a sparse subset of pixel intensities, achieving super-realtime performance with high quality predictions.
20. A survey of multiple classifier systems as hybrid systems , by Corchado, E., Graña, M., & Wozniak, M. (2014). Information Fusion, 16, 3-17. (cited 269 times, HIC: 1 , CV: 22)
    Summary: A current focus of intense research in pattern classification is the combination of several classifier systems, which can be built following either the same or different models and/or datasets building.