Revolutionizing Medical Diagnosis with Deep Learning: TED Talk

https://www.youtube.com/watch?v=w2_N_p_Y-W4&feature=youtu.be

Understanding and Controlling Memory in Recurrent Neural Networks (ICML'19 oral)

This paper shows that RNNs are able to form long-term memories despite being trained only for short-term with a limited amount of timesteps, but that not all memories are created equal. The authors find that each memory is correlated with a dynamical object in the hidden-state phase space and that the objects properties can quantitatively predict long term effectiveness. By regularizing the dynamical object, the long-term functionality of the RNN is significantly improved, while not adding to the computational complexity of training.

Link to PDF: https://proceedings.mlr.press/v97/haviv19a/haviv19a.pdf

Study shows that artificial neural networks can be used to drive brain activity.



MIT neuroscientists have performed the most rigorous testing yet of computational models that mimic the brain’s visual cortex.

Using their current best model of the brain’s visual neural network, the researchers designed a new way to precisely control individual neurons and populations of neurons in the middle of that network. In an animal study, the team then showed that the information gained from the computational model enabled them to create images that strongly activated specific brain neurons of their choosing.

The findings suggest that the current versions of these models are similar enough to the brain that they could be used to control brain states in animals. The study also helps to establish the usefulness of these vision models, which have generated vigorous debate over whether they accurately mimic how the visual cortex works, says James DiCarlo, the head of MIT’s Department of Brain and Cognitive Sciences, an investigator in the McGovern Institute for Brain Research and the Center for Brains, Minds, and Machines, and the senior author of the study.



Full article: https://news.mit.edu/2019/computer-model-brain-visual-cortex-0502

Science paper: https://science.sciencemag.org/content/364/6439/eaav9436

Biorxiv (open access): https://www.biorxiv.org/content/10.1101/461525v1

Andrew Ng and Masoumeh Haghpanahi the team's new paper -- Cardiologist-level arrhythmia detection from ECG using deep learning.

https://www.nature.com/articles/s41591-018-0268-3


https://stanfordmlgroup.github.io/projects/ecg2/

https://t.iss.one/ArtificialIntelligenceArticles

#DeepLearning #MachineLearning #artificalintelligence

@ArtificialIntelligenceArticles

Introducing FastBert — A simple Deep Learning library for BERT Models
Blog by Kaushal Trivedi: https://medium.com/huggingface/introducing-fastbert-a-simple-deep-learning-library-for-bert-models-89ff763ad384
#MachineLearning #ArtificialIntelligence #NLP #Bert #NaturalLanguageProcessing

A curated list of gradient boosting research papers from the last 25 years with implementations. It covers NeurIPS, ICML, ICLR, KDD, ICDM, CIKM, AAAI etc.
https://github.com/benedekrozemberczki/awesome-gradient-boosting-papers

Very interesting work applying machine learning to higher order logics and theorem proofs. This could eventually change how we understand and program many different things.

https://arxiv.org/abs/1904.03241

DeepMind Made a Math Test For Neural Networks
https://arxiv.org/abs/1904.01557

Should AI Research Try to Model the Human Brain?


https://www.cell.com/trends/cognitive-sciences/fulltext/S1364-6613(19)30061-0

17 equations that changed the world

Attentive Generative Adversarial Network for Raindrop Removal from A Single Image

Abstract : "Raindrops adhered to a glass window or camera lens can severely hamper the visibility of a background scene and degrade an image considerably. In this paper, we address the problem by visually removing raindrops, and thus transforming a raindrop degraded image into a clean one. The problem is intractable, since first the regions occluded by raindrops are not given. Second, the information about the background scene of the occluded regions is completely lost for most part. To resolve the problem, we apply an attentive generative network using adversarial training (...)."

Qian et al.: https://arxiv.org/pdf/1711.10098.pdf

#artificialintelligence #deeplearning #generativeadversarialnetwork

DLTK – Deep Learning Toolkit for Medical Image Analysis, built on top of TensorFlow
Fast prototyping with a low entry threshold & reproducibility in image analysis applications, with a particular focus on medical imaging.

https://github.com/DLTK/DLTK

News & Views: In Nature Methods, two research teams report substantial improvements in the accurate prediction of fragment ion spectra by deep neural networks.

https://www.nature.com/articles/s41592-019-0428-5.epdf

This Review, published in Nature Reviews Genetics, describes different deep learning techniques and how they can be applied to extract biologically relevant information from large, complex genomic data sets.
https://www.nature.com/articles/s41576-019-0122-6.epdf

Discovering Neural Wirings (https://arxiv.org/abs/1906.00586)

In the past years developing deep neural architectures either required manual design (e.g. AlexNet, ResNet, MobileNet, ...) or require expensive search among possible predefined block structures of layers (NAS, MNas, DART,...). What if we see a neural network as a completely unstructured graph? where each node is running a simple sensing operation over a single data-point or a channel (e.g. 2d filter) and all the nodes are wired up massively in the network. In this paper we explain how to discover a good wiring of a neural network that minimizes the loss function with a limited amount of computation. We relax the typical notion of layers and instead enable channels to form connections independent of each other. This allows for a much larger space of possible networks. The wiring of our network is not fixed during training – as we learn the network parameters we also learn the structure itself.

MelNet: A Generative Model for Audio in the Frequency Domain
Sean Vasquez and Mike Lewis: https://arxiv.org/abs/1906.01083
Blog: https://sjvasquez.github.io/blog/melnet/
#ArtificialIntelligence #DeepLearning #MachineLearning

A summary of the debate on human-level AI organized by the World Science Festival last Friday.
I shared the stage with Gary Kasparov, Shannon Vallor, Hod Lipson, and moderator Daniel Sieberg.


https://www.zdnet.com/article/artificial-general-intelligence-is-a-rorschach-test-do-we-need-orangutans/

Functional Adversarial Attacks
Cassidy Laidlaw and Soheil Feizi: https://arxiv.org/abs/1906.00001
#ArtificialIntelligence #DeepLearning #MachineLearning

Deep Learning and the Game of Go

GitHub : https://github.com/maxpumperla/deep_learning_and_the_game_of_go

#artificialintelligence #machinelearning #reinforcementlearning

IoT Network Security from the Perspective of Adversarial Deep Learning. arxiv.org/abs/1906.00076