According to common belief, neural networks' main advantage over traditional ML algorithms is that NNs learn features by themselves while in the traditional ML, you handcraft features. This is not exactly true. Well, it's true for vanilla feed-forward NNs consisting only of fully connected layers. But those are very hard to train for high dimensional inputs like images.

When you use a convolutional neural network, you already use two types of handcrafted features: convolution filters and pooling filters.

The designer of the convolutional NN for image classification has looked into the input data (this is what traditional ML engineers do to invent features) and decided that patches of pixels close to each other contain information that could help in classification, and at the same time reduce the number of NN parameters.

The same reasoning is used when we classify texts using bag-of-words features. We look at the data and decide that individual words and n-grams of words would be good features to classify a document. This reduces the number of input features while allowing us to accurately classify documents.

BTW, the way convolutional filters apply (sum of element-wise multiplications, spanning over channels resulting in one number) is a hell of a feature!
Burkov

❇️ @AI_Python_EN

Google’s SummAE generates abstract summaries of paragraphs
#DataScience #MachineLearning #ArtificialIntelligence

http://bit.ly/2pVMjZJ

❇️ @AI_Python_EN

.

just published my (free) 81-page guide on learning #ComputerVision, #DeepLearning, and #OpenCV!
Includes step-by-step instructions on:
- Getting Started
- Face Applications
- Object Detection
- OCR
- Embedded/IOT
- and more!
Check it out here:
http://pyimg.co/getstarted
And if you liked it, please do give it a share to spread the word. Thank you!
#Python #Keras #MachineLearning #ArtificialIntelligence #AI

❇️ @AI_Python_EN

The war between ML frameworks has raged on since the rebirth of deep learning. Who is winning? Horace He data analysis shows clear trends: PyTorch is winning dramatically among researchers, while Tensorflow still dominates industry.
#PyTorch #Tensorflow

https://thegradient.pub/state-of-ml-frameworks-2019-pytorch-dominates-research-tensorflow-dominates-industry/

❇️ @AI_Python_EN

.


Free Book: Deep Learning and Computer Vision with CNNs

https://www.datasciencecentral.com/profiles/blogs/free-book-deep-learning-and-computer-vision-with-cnns

❇️ @AI_Python_EN

If you're interested in using pytorch on free Colab TPUs, here are some notebooks to get you started

https://github.com/pytorch/xla/tree/master/contrib/colab

❇️ @AI_Python_EN

Automatically generate floor plans designs.
https://t.co/WEMrTKCI86
https://t.co/GKxF2a7H72

Image Quality Assessment for Rigid Motion Compensation.
https://t.co/3TC1f2fst3
https://t.co/ZpwS8PIYGj

"Span-core Decomposition for Temporal.." is the #1 paper on Arxiv today in data structures and algorithms. Github code (span_cores) supports their results.See it at
http://assert.pub/arxiv/cs/cs.ds
http://assert.pub/papers/1910.03645

❇️ @AI_Python_EN

Spoken Language Identification using ConvNets.
https://arxiv.org/abs/1910.04269

❇️ @AI_Python_EN

Torch vs Theano vs TensorFlow vs Keras

☞ https://morioh.com/p/a80813c4a01c

#ai #deeplearning

❇️ @AI_Python_EN

"...can we say now, finally, that computers are as powerful as the human brain? No. Focusing on raw computing power misses the point entirely. Speed alone won’t give us AI. Running a poorly designed algorithm on a faster computer doesn’t make the algorithm better; it just means you get the wrong answer more quickly. (And with more data there are more opportunities for wrong answers!)

The principal effect of faster machines has been to make the time for experimentation shorter, so that research can progress more quickly. It’s not hardware that is holding AI back; it’s software. We don’t yet know how to make a machine really intelligent—even if it were the size of the universe...

Turing himself proved that some problems are undecidable by any computer: the problem is well defined, there is an answer, but there cannot exist an algorithm that always finds that answer...

The machine may be far more capable than us, but it will still be far from perfectly rational."

Stuart Russell in "Human Compatible"

❇️ @AI_Python_EN

Predicting survival from colorectal cancer histology slides using #deeplearning

1. They conducted the study because:
• Colorectal cancer (CRC) is a common disease with a variable clinical course, and there is a high clinical need to more accurately predict the outcome of individual patients.
• For almost every CRC patient, histological slides of tumor tissue are routinely available.
• Deep learning can be used to extract information from very complex images, and we hypothesized that deep learning can predict clinical outcome directly from histological images of CRC.

2. What did the researchers do and find?
• We trained a deep neural network to identify different tissue types & demonstrated that it can decompose complex tissue into its constituent parts and thereby showed that this score improves survival prediction compared to the SOTA avaiable.

3. Conclusion
• Deep learning is an inexpensive tool to predict the clinical course of CRC patients based on ubiquitously available histological images.
• Prospective validation studies are needed to firmly establish this biomarker for routine clinical use.

☞ Link to research

#healthcare #AI #machinelearning

❇️ @AI_Python_EN

The State of Transfer Learning in NLP

http://ruder.io/state-of-transfer-learning-in-nlp/

#TransferLearning #NaturalLanguageProcessing
#NLP

❇️ @AI_Python_EN

Autoencoders for Image Reconstruction in Python and Keras

https://stackabuse.com/autoencoders-for-image-reconstruction-in-python-and-keras/


❇️ @AI_Python_EN

DeepPrivacy model for making people on photoes unrecognizable (by humans)


ArXiV: https://arxiv.org/pdf/1909.04538.pdf

#MaskRCNN #DeepPrivacy #CV #DL

❇️ @AI_Python_EN

Self-supervised QA from Facebook AI

The researchers from Facebook AI published a paper with the results of exploring the idea of unsupervised extractive question answering and the following training of the supervised question answering model. This approach achieves 56.41F1 on SQuAD2 dataset.

Original paper:
https://research.fb.com/wp-content/uploads/2019/07/Unsupervised-Question-Answering-by-Cloze-Translation.pdf?
Code for experiments:
https://github.com/facebookresearch/UnsupervisedQA


#NLP #BERT #FacebookAI #SelfSupervised

❇️ @AI_Python_EN

1-point RANSAC for Circular Motion Estimation in Computed Tomography (CT)
https://deepai.org/publication/1-point-ransac-for-circular-motion-estimation-in-computed-tomography-ct

by Mikhail O. Chekanov et al.
#Statistics #Estimator

❇️ @AI_Python_EN

5 types of #machinelearning algorithms you should know
Model development is not one-size-fits-all -- there are different types of #machinelearning algorithm

https://www.7wdata.be/big-data/5-types-of-machine-learning-algorithms-you-should-know/

❇️ @AI_Python_EN

Deep Learning:
http://course.fast.ai
NLP:
http://bit.ly/fastai-nlp
Comp Linear Algebra:  
http://github.com/fastai/numerical-linear-algebra
Bias, Ethics, & AI:
http://fast.ai/topics/#ai-in-society
Debunk Pipeline Myth:
http://bit.ly/not-pipeline
AI Needs You:
http://bit.ly/rachel-TEDx
Ethics Center: 
http://bit.ly/USF-CADE

❇️ @AI_Python_EN