«IEEE’s camera identification challenge — different approach to teaming up»

https://hackernoon.com/ieees-camera-identification-challenge-different-approach-to-teaming-up-28da44dfe635

#cv #dbrain

Neural nets are terrible at arithmetic & counting. If you train one in 1 to 10, it will do okay on 3 + 5 but fail miserably for 1000 + 3000. Resolving this, «Neural Arithmetic Logic Units» can track time, do arithmetic on images of numbers, & extrapolate, providing better results than other architectures.

https://arxiv.org/pdf/1808.00508.pdf

#nn #architecture #concept #deepmind #arithmetic

The Conversational Intelligence Challenge 2 (ConvAI second part) got announced today.

The aim of our competition is to establish a concrete scenario for testing chatbots that aim to engage humans, and become a standard evaluation tool in order to make such systems directly comparable.

The most promising team attending will receive a travel grant to attend #NIPS2018

#nlp #dl #dialoguesystem #competition

Stanford engineers have combined two types of computers to create a faster and less energy-intensive image processor for use in autonomous vehicles, security cameras and medical devices.

https://news.stanford.edu/2018/08/17/new-ai-camera-revolutionize-autonomous-vehicles/

Recent Advances for a Better Understanding of Deep Learning − Part I

https://towardsdatascience.com/recent-advances-for-a-better-understanding-of-deep-learning-part-i-5ce34d1cc914

#dl #theory

What is a Generative Adversarial Network?

Another article about how GANs work.

http://hunterheidenreich.com/blog/what-is-a-gan/

#gan #theory #dl

Curious About How To Be A Data Scientist? Hear From A Netflix Data Scientist

Article about how data science #production works. How problem should be defined and how project should be maintained and run.

https://towardsdatascience.com/a-peek-into-a-netflix-data-scientists-day-66bf3dacabb9

First 1e6 integers, represented as binary vectors indicating their prime factors, and laid out using the sparse matrix support in leland_mcinnes's UMAP dimensionality reduction algorithm. This is from a 1000000x78628 (!) binary matrix. Very pretty structure emerges.

"Learning Hierarchical Semantic Image Manipulation through Structured Representations":

https://arxiv.org/abs/1808.07535

#cv #dl

Online ad demand prediction #kaggle competition 1st place summary:
https://www.kaggle.com/c/avito-demand-prediction/discussion/59880

Winner explains how to combine categorical, numerical, image and text features into a single #NN that gets you into top 10 without stacking.

#PyTorch tensor basics: https://www.kdnuggets.com/2018/05/pytorch-tensor-basics.html

New model to naturally detect depression in conversations

Link: http://news.mit.edu/2018/neural-network-model-detect-depression-conversations-0830

#nlp #audio #dl

Tickets for one of the key conferences #NIPS2018 were sold out in 15 minutes.

NIPS Conference Registrations 2002 thru 2019.
[2018] War erupts for tickets
[2019] AI researchers discover time travel.

#NIPS2018

1st place solution in the recent Home Credit Default Risk #Kaggle competition

- extensive feature engineering, with ~700 of features total used
- XGBoost, LightGBM, CatBoost, FastRGF, DAE+NN, Lin Reg
- 3-level ensembling (stacking x2 + blending)

Link: https://www.kaggle.com/c/home-credit-default-risk/discussion/64821

More on that model in the Wired article: https://thenextweb.com/artificial-intelligence/2018/09/05/mits-depression-detecting-ai-might-be-its-scariest-creation-yet/

The article notes that the best advantage of suggesting approach is that it is context-free.

Those who want to understand #GAN s better can now play with them in the browser with GAN lab.

https://towardsdatascience.com/gan-lab-train-gans-in-the-browser-21a423585460

#Google introduced Conceptual Captions, a new dataset and challenge for image captioning consisting of ~3.3 million image/caption pairs for the machine learning community to train and evaluate their own image captioning models.

Link: https://ai.googleblog.com/2018/09/conceptual-captions-new-dataset-and.html

#dataset