#Alice_Zheng_Feature_Engineering_for_Machine_Learning_2018
#book
----------
@machinelearning_tuts

Advanced Analytics with Spark — S. Ryza и др. (en) 2017
#book #middle #spark
----------
@machinelearning_tuts

Advanced Analytics with Spark — S. Ryza и др. (en) 2017
#book #middle #spark
----------
@machinelearning_tuts

Allocated time to media per person

#statistics #visualization

Source:Nielsen
----------
@machinelearning_tuts

#google_assistant
----------
@machinelearning_tuts

https://www.analyticsindiamag.com/iit-hyd-btech-ai-admission-jee-advanced/
----------
@machinelearning_tuts

💡 We have enlisted THE BEST RESOURCES for learning - Statistics and Probability, for you all! Download the document for free, from the link specified below! Happy Learning! ♥

Link: bit.ly/Statistics-Probability-Resources
----------
@machinelearning_tuts

#مقاله
Cross-entropy loss can be equated to the Jensen-Shannon distance metric, and it was shown in early 2017 by Arjovsky et al. that this metric would fail in some cases and not point in the right direction in other cases. This group showed that the Wasserstein distance metric (also known as the earth mover or EM distance) worked and worked better in many more cases.

https://arxiv.org/abs/1701.07875

#GAN @WGAN
----------
@machinelearning_tuts

#datascience #machinelearning #learning #AI #beginner
Coursera MOOC for "Neural Networks for Machine Learning" by Geoffrey Hinton (Known as GodFather of AI) was prepared in 2012. But the lectures are still a good introduction to many of the basic ideas and are available at
https://www.cs.toronto.edu/~hinton/coursera_lectures.html

----------
@machinelearning_tuts

#datascience #machinelearning #R #learning #beginner
Data Science with R - Beginners. Free for limited time. Hurry up.

----------
@machinelearning_tuts

Generalization and Expressivity for Deep Nets

--Abstract
Along with the rapid development of deep learning in practice, thetheoretical explanations for its success become urgent. Generalization andexpressivity are two widely used measurements to quantify theoretical behaviorsof deep learning. The expressivity focuses on finding functions expressible bydeep nets but cannot be approximated by shallow nets with the similar number ofneurons. It usually implies the large capacity. The generalization aims atderiving fast learning rate for deep nets. It usually requires small capacityto reduce the variance. Different from previous studies on deep learning,pursuing either expressivity or generalization, we take both factors intoaccount to explore the theoretical advantages of deep nets. For this purpose,we construct a deep net with two hidden layers possessing excellentexpressivity in terms of localized and sparse approximation. Then, utilizingthe well known covering number to measure the capacity, we find that deep netspossess excellent expressive power (measured by localized and sparseapproximation) without enlarging the capacity of shallow nets. As aconsequence, we derive near optimal learning rates for implementing empiricalrisk minimization (ERM) on the constructed deep nets. These resultstheoretically exhibit the advantage of deep nets from learning theoryviewpoints.

2018-03-10T07:41:25Z

@machinelearning_tuts
----------
Link : http://arxiv.org/abs/1803.03772v2

Ophthalmic Diagnosis and Deep Learning -- A Survey

--Abstract
This survey paper presents a detailed overview of the applications for deeplearning in ophthalmic diagnosis using retinal imaging techniques. The need ofautomated computer-aided deep learning models for medical diagnosis isdiscussed. Then a detailed review of the available retinal image datasets isprovided. Applications of deep learning for segmentation of optic disk, bloodvessels and retinal layer as well as detection of red lesions arereviewed.Recent deep learning models for classification of retinal diseaseincluding age-related macular degeneration, glaucoma, diabetic macular edemaand diabetic retinopathy are also reported.

2018-12-09T05:57:17Z

@machinelearning_tuts
----------
Link : http://arxiv.org/abs/1812.07101v1

Deep Embedding Kernel

--Abstract
In this paper, we propose a novel supervised learning method that is calledDeep Embedding Kernel (DEK). DEK combines the advantages of deep learning andkernel methods in a unified framework. More specifically, DEK is a learnablekernel represented by a newly designed deep architecture. Compared withpre-defined kernels, this kernel can be explicitly trained to map data to anoptimized high-level feature space where data may have favorable featurestoward the application. Compared with typical deep learning using SoftMax orlogistic regression as the top layer, DEK is expected to be more generalizableto new data. Experimental results show that DEK has superior performance thantypical machine learning methods in identity detection, classification,regression, dimension reduction, and transfer learning.

2018-04-16T17:25:24Z

@machinelearning_tuts
----------
Link : http://arxiv.org/abs/1804.05806v1

Split learning for health: Distributed deep learning without sharing raw patient data

--Abstract
Can health entities collaboratively train deep learning models withoutsharing sensitive raw data? This paper proposes several configurations of adistributed deep learning method called SplitNN to facilitate suchcollaborations. SplitNN does not share raw data or model details withcollaborating institutions. The proposed configurations of splitNN cater topractical settings of i) entities holding different modalities of patient data,ii) centralized and local health entities collaborating on multiple tasks andiii) learning without sharing labels. We compare performance and resourceefficiency trade-offs of splitNN and other distributed deep learning methodslike federated learning, large batch synchronous stochastic gradient descentand show highly encouraging results for splitNN.

2018-12-03T05:43:20Z

@machinelearning_tuts
----------
Link : http://arxiv.org/abs/1812.00564v1

Deep-learning technique reveals “invisible” objects in the dark

--Abstract
Method could illuminate features of biological tissues in low-exposure images.

December 12, 2018

@machinelearning_tuts
----------
Link : http://news.mit.edu//2018/deep-learning-dark-objects-1212

Opportunities for materials innovation abound

--Abstract
Faculty researchers share insights into new capabilities at the annual Industrial Liaison Program Research and Development Conference.

December 14, 2018

@machinelearning_tuts
----------
Link : http://news.mit.edu//2018/mit-industrial-liaison-program-conference-1214

Recent Advances in Deep Learning: An Overview

--Abstract
Deep Learning is one of the newest trends in Machine Learning and ArtificialIntelligence research. It is also one of the most popular scientific researchtrends now-a-days. Deep learning methods have brought revolutionary advances incomputer vision and machine learning. Every now and then, new and new deeplearning techniques are being born, outperforming state-of-the-art machinelearning and even existing deep learning techniques. In recent years, the worldhas seen many major breakthroughs in this field. Since deep learning isevolving at a huge speed, its kind of hard to keep track of the regularadvances especially for new researchers. In this paper, we are going to brieflydiscuss about recent advances in Deep Learning for past few years.

2018-07-21T15:40:10Z

@machinelearning_tuts
----------
Link : http://arxiv.org/abs/1807.08169v1

Geometric Understanding of Deep Learning

--Abstract
Deep learning is the mainstream technique for many machine learning tasks,including image recognition, machine translation, speech recognition, and soon. It has outperformed conventional methods in various fields and achievedgreat successes. Unfortunately, the understanding on how it works remainsunclear. It has the central importance to lay down the theoretic foundation fordeep learning. In this work, we give a geometric view to understand deep learning: we showthat the fundamental principle attributing to the success is the manifoldstructure in data, namely natural high dimensional data concentrates close to alow-dimensional manifold, deep learning learns the manifold and the probabilitydistribution on it. We further introduce the concepts of rectified linear complexity for deepneural network measuring its learning capability, rectified linear complexityof an embedding manifold describing the difficulty to be learned. Then we showfor any deep neural network with fixed architecture, there exists a manifoldthat cannot be learned by the network. Finally, we propose to apply optimalmass transportation theory to control the probability distribution in thelatent space.

2018-05-26T09:15:53Z

@machinelearning_tuts
----------
Link : http://arxiv.org/abs/1805.10451v2

Machine Learning Refined — J. Watt, R. Borhani, A. K. Katsaggelos (en) 2016
#book #middle #theory

----------
@machinelearning_tuts
@drivelesscar
@autonomousvehicle

Machine Learning Refined — J. Watt, R. Borhani, A. K. Katsaggelos (en) 2016
#book #middle #theory

----------
@machinelearning_tuts
@drivelesscar
@autonomousvehicle