|May 4 · Issue #39 · View online |
Hiya deep learning ninjas and welcome to a new and exciting week.
This week NVIDIA added six AI Startups to its venture portfolio, deep learning is being employed to save our beloved honey bee, Google research open sources a fascination attention based OCR system used for automatic street name extraction for Google Street View and an exciting paper on a new learning paradigm for Neural Machine Translation has been released.
As always if you’d like to support this newsletter sharing it with friends and colleagues would really help us :)
| Top 5 Key Areas: Deep Learning in Retail & Advertising |
An insightful look at the future impact of deep learning in retail and advertisingGartner predicts that 85% of customer interactions will be managed autonomously by 2020, with cross-channel bots that are able to recognize the voices and faces of customers as soon as 2018. But the widespread impact of AI in retail and advertising won’t be limited to recommendations in the near future - in fact, it’s set to impact and improve every corner of the industry.
| NVIDIA Adds Six AI Startups to Its GPU Ventures Portfolio |
As the AI ecosystem continues its global expansion, NVIDIA is expanding its portfolio of startup investments, adding six companies in three countries over the past year.
| BeeScanning |
| What do deep neural networks understand of fractals? |
Grégory Châtel decided to investigate the influence of activation functions and the general architecture of neural networks on their performance. He tackled a quite interesting challenge of approximating the mandelbrot sets colors and complements his results with a Keras implementation.
| Have We Forgotten about Geometry in Computer Vision? |
This article by Alex Kendall reminds us about the role of geometry in computer vision by giving examples, where using geometry to define basic rules may help in producing better results. He covers unsupervised learning and the issues of purely semantics-based approaches, as well as successes in his recent research that were supported by geometry.
| Photo Editing with Generative Adversarial Networks (Part 2) |
Greg Heinrich shows us, what’s possible with GANs and creates a quite fun looking face generation tool on the way. There are some nice insights on design choices for such a system and it comes with extensive source code as well, although it’s tailored for Nvidia’s Digits web app.
| Google Research: Updating Google Maps with Deep Learning and Street View |
In order to enable the automatic extraction of information from Google’s geo-located street view images its Ground Truth team has pioneered an approach
for using a deep neural network to accurately read street names which has achieved 84,2% accuracy on the challenging French Street Name Signs (FSNS) dataset.
What is more the entire attention-ocr model has been open sourced!
| Expanded fastText library now fits on smaller-memory devices |
Facebook has updated its fastText library with a new technique to reduce its memory footprint. This allows usage on more devices and may enable some interesting use cases.
| Automatic Speech Recognition: End-to-end automatic speech recognition from scratch in Tensorflow |
A powerful library for automatic speech recognition implemented in TensorFlow and support training with CPU/GPU.
| STAIR Captions: Constructing a Large-Scale Japanese Image Caption Dataset |
The authors construct a large-scale Japanese image caption dataset based on images from MS-COCO, which is called STAIR Captions. STAIR Captions consists of 820,310 Japanese captions for 164,062 images. The neural network trained using STAIR Captions manages to generate more natural and better Japanese captions, compared to those generated using English-Japanese machine translation after generating English captions.
| Tweeting AI: Perceptions of AI-Tweeters (AIT) vs Expert AI-Tweeters |
This fun, little paper looks at how the public perceives the progress of AI by utilizing the data shared on Twitter. Specifically, this paper performs a comparative analysis on the understanding of users from two categories – general AI-Tweeters (AIT) and the expert AI-Tweeters (EAIT) who share posts about AI on Twitter.
| Adversarial Neural Machine Translation |
This paper presents a new learning paradigm for Neural Machine Translation (NMT). Instead of maximizing the likelihood of the human translation as in previous works, the authors minimize the distinction between human translation and the translation given by a NMT model.