An Atari Model Zoo for Analyzing, Visualizing, and Comparing Deep Reinforcement Learning Agents
Such et al.: https://arxiv.org/abs/1812.07069
Code: https://github.com/uber-research/atari-model-zoo
Blog: https://eng.uber.com/atari-zoo-deep-reinforcement-learning/
#ArtificialIntelligence #DeepLearning #ReinforcementLearning

MegatronLM: Training Billion+ Parameter Language Models Using GPU Model Parallelism
"... training an 8.3 billion parameter transformer language model with 8-way model parallelism and 64-way data parallelism on 512 GPUs, making it the largest transformer based language model ever trained at 24x the size of BERT and 5.6x the size of GPT-2."
Blog by NVIDIA Applied Deep Learning Research : https://nv-adlr.github.io/MegatronLM
Code: https://github.com/nvidia/megatron-lm
#ArtificialIntelligence #DeepLearning #NLP #PyTorch #Transformer

"There has been surprisingly little mainstream discussion about how the techniques we classify as AI actually work.”

Andrew Ng and Derrick Harris discuss enterprise AI in just 15 minutes: https://content.pivotal.io/intersect/ai-in-15-minutes

The functional organisation of the hippocampus along its long axis is gradual and predicts recollection
https://reader.elsevier.com/reader/sd/pii/S0010945219301832?token=7ED07A1848AFEC6DC2EA8E0B689CC9EB30478F019977CF3DB31E898DE8DFF75456605EE621D0E2B7D2212D3FD9E00C72

Understanding XLNet
https://www.borealisai.com/en/blog/understanding-xlnet/

ICYMI: Best demo paper from ACL 2019 (super recent)

https://www.profillic.com/paper/arxiv:1902.08646

The paper introduces the Pytorch-based open source framework OpenKiwi for translation quality estimation

Generating Diverse High-Fidelity Images with VQ-VAE-2

Ali Razavi, Aaron van den Oord, Oriol Vinyals : https://arxiv.org/abs/1906.00446

#DeepLearning #VariationalAutoEncoder #VAE

Project Euphonia’s Personalized Speech Recognition for Non-Standard Speech
Blog by Joel Shor and Dotan Emanuel : https://ai.googleblog.com/2019/08/project-euphonias-personalized-speech.html
#ArtificialIntelligence #DeepLearning #NeuralNetworks

Is Deep Reinforcement Learning Really Superhuman on Atari?
Marin Toromanoff, Emilie Wirbel, Fabien Moutarde : https://arxiv.org/abs/1908.04683
#deeplearning #machinelearning #reinforcementlearning

Object as Distribution #NeurIPS2019


Propose bivariate normal distribution for object detection representation.
Benefits detection of highly-overlapping objects and downstream tracking

https://arxiv.org/abs/1907.12929

The following is our paper on driver functional vigilance during use of Tesla Autopilot driver assistance system. We analyzed 18,928 Autopilot disengagements. 3+ years of hard work with an incredible research team at MIT. Example videos out next week.

link: https://hcai.mit.edu/human-side-of-tesla-autopilot/

If a neural network generates an image, who owns the copyright? The owner of the dataset that the net was trained on? The designer of the network architecture? The person running the code? Or... the AI system itself? @lexfridman

Top 10 of 2019: AI and Deep Learning Content Recommended by Experts

https://medium.com/@teamrework/top-10-of-2019-ai-and-deep-learning-content-recommended-by-experts-58a19166e5bd

Multiscale Representations for Manifold-Valued Data

Rahman et al.: https://statweb.stanford.edu/~symmlab/SymmPaper.pdf

#SymmetricSpace #Wavelets #Denoising

Visionary GPU Architecture Paper Wins SC19 Test of Time Award

Blog by Jeffrey K. Hollingsworth : https://sc19.supercomputing.org/2019/08/13/visionary-gpu-architecture-paper-wins-sc19-test-of-time-award/

#ArtificialIntelligence #DeepLearning #GPU

Reward Tampering Problems and Solutions in Reinforcement Learning: A Causal Influence Diagram Perspective

Tom Everitt and Marcus Hutter : https://arxiv.org/abs/1908.04734

#ArtificialIntelligence #MachineLearning #ReinforcementLearning

Creating a Pop Music Generator with the Transformer

Blog by Andrew Shaw : https://towardsdatascience.com/creating-a-pop-music-generator-with-the-transformer-5867511b382a

#Music #DeepLearning #ArtificialIntelligence

SLIDES
Transfer Learning in Natural Language Processing
June 2, 2019
NAACL-HLT 2019
https://docs.google.com/presentation/d/1fIhGikFPnb7G5kr58OvYC3GN4io7MznnM0aAgadvJfc/edit#slide=id.g5888218f39_177_4

The Math behind Neural Networks: Part 1 - The Rosenblatt Perceptron
https://www.codeproject.com/Articles/4047091/The-Math-behind-Neural-Networks-Part-1-The-Rosenbl

Why BLEU score sucks for evaluating translation systems.
(Or rather, why BLEU score works fine when you translation system sucks, but sucks when it's good).

https://arxiv.org/abs/1908.05204

Welcome SUPERGLUE from Facebook AI, DeepMind, University of Washington and New York University.
It comprises new ways to test creative approaches on a range of difficult NLP tasks and serves a series of benchmark tasks to measure the performance of modern, high performance language-understanding AI.

Made on the premise that deep learning models for conversational AI have “hit a ceiling” and need greater challenges .

Read https://arxiv.org/pdf/1905.00537.pdf