New advances in deep learning

cosmos 3rd April 2019 at 1:47am
Artificial intelligence innovation Deep learning

https://paperswithcode.com/sota

review/position paper on giving neural nets the right biases – https://arxiv.org/abs/1806.01261

Deep Learning State of the Art (2019) - MIT

Advances in reinforcement learning

Advances in NLP, Transformer network, BERT (see here) – https://blog.openai.com/better-language-models/#task6

2017 updates: http://www.wildml.com/2017/12/ai-and-deep-learning-in-2017-a-year-in-review/

DeepMind publications: https://deepmind.com/research/publications/

Conferences: ICLR, NIPS

Transformer networks. https://research.googleblog.com/2017/08/transformer-novel-neural-network.html – New: Universal Transformers networks

Attention is all you need..

Data augmentation

Model compression –> Neural network compression


Generative adversarial networks, etc.

Generating video! Video-to-video synthesis, like Image-to-image synthesis (pix2pix, CycleGAN).

Variational autoencoder

WaveNet, PixelRNN, SampleRNN

Augmented RNNs (Memory-augmented)

Neural architecture search

Neural Architecture Search with Reinforcement Learning

Meta-learning

Avoiding catastrophic forgetting..

Capsules. Dynamic Routing Between Capsules

.. PoseNet

Object detection

neural theorem proving: https://arxiv.org/abs/1807.08204


Multisensory integration

Deep learning theory

Deep learning with Elastic Averaging SGD

We study the problem of stochastic optimization for deep learning in the parallel computing environment under communication constraints. A new algorithm is proposed in this setting where the communication and coordination of work among concurrent processes (local workers), is based on an elastic force which links the parameters they compute with a center variable stored by the parameter server (master).


Solve vanishing gradients problem, allow be computations to depend on all/any of the existing computations, thus creating a DAG structure instead of a layered one: http://www.jonolick.com/home/dagnn-a-deeper-fully-connected-network

Batch normalization

Deep Networks with Stochastic Depth Stochastic Depth Networks will Become the New Normal

Highway Networks

Dropout

https://en.m.wikipedia.org/wiki/Modular_neural_network STN?

DCGAN http://arxiv.org/abs/1511.06434

DRAW http://arxiv.org/abs/1502.04623

Soft/hard attention https://www.google.es/url?sa=t&source=web&rct=j&url=http://arxiv.org/pdf/1502.03044&ved=0ahUKEwi4yof-jPjLAhVC5xoKHcTjDM4QFgggMAA&usg=AFQjCNEs1Yw8fZF9oaqo73cwbHJqKwQHTw

CharCNN https://www.google.es/url?sa=t&source=web&rct=j&url=http://arxiv.org/pdf/1508.06615&ved=0ahUKEwjZqaXnk_jLAhWCsxQKHZsuApUQFgglMAM&usg=AFQjCNHk8JQpI98eUtyiluv7d2G9aWRtyA

NeuralStyle https://github.com/jcjohnson/neural-style

"Take a look at @karpathy's Tweet: https://twitter.com/karpathy/status/709465955223543808?s=09"

http://arxiv.org/abs/1604.00790 bidirectional LSTM

http://www.computervisionblog.com/2016/06/deep-learning-trends-iclr-2016.html

Adversarial networks

Neural Turing machines, neural programmers-interpreters

http://www.kdnuggets.com/2016/09/9-key-deep-learning-papers-explained.html


A couple of cool recent neural network developments! * Lip reading * Neural enhance! Super-resolution of images (like in movies but real ;) )

LipNet. Lip reading NN: http://prostheticknowledge.tumblr.com/post/152735696866/lipnet-deep-learning-research-from-the-university

image super-resolution using deep convolutional networks! try here: http://waifu2x.udp.jp/ NeuralEnhance lets you apply 4x super-resolution to your photos CSI-style in only 340 lines of code! https://github.com/alexjc/neural-enhance

https://twitter.com/karpathy

https://twitter.com/alexjc

https://twitter.com/NandoDF