Welcome to your first issue of AI Weekly!
This is a weekly collection of worthwhile news, resources, tools, books and startups in the fields of Artificial Intelligence, Machine Learning and Cognitive Sciences.
I hope you will find the issue worthwhile. If you don't or if you have suggestions, please reply to this email to let me know!
Thanks for subscribing!
David
Artificial Intelligence Weekly
In the News
Rise of the machines
The Economist published an excellent review on AI this week. This is a thorough article and a great introduction to the field.
For those who like the article, you may also want to read the leader published in the same issue: The Dawn of Artificial Intelligence and this other article on Money management being taken over by algorithms
Baidu Leads in Artificial Intelligence Benchmark
Earlier this year, Microsoft announced they had outperformed Google and humans at Imagenet image classification.
Baidu is now announcing performances that surpass those of Google and Facebook, in what is starting to look like a public arms race. This paper provides more technical details on the approach and the supercomputer Minwa built for this task.
The Machine Vision Algorithm Beating Art Historians at Their Own Game
Working on a database of paintings from 1000 artists and 27 different styles, teams at Rutgers University have developed machine vision algorithms that identify a painting's artist correctly over 60% of the time and the style over 40% of the time. According to this article, this performance is on par with what art historians would achieve.
An interesting aspect of the algorithms is that they can help discover links and similarities between different artists and different styles, in a way that was difficult and time consuming before.
This has fascinating implications for art history. Enjoy the read.
Spotify on AI in Music
Short interview with Nicola Montecchio, Spotify's Head of Deep Learning, on the way AI and Deep Learning are being leveraged by Spotify's recommendation engine.
Learning
Are there Deep Reasons Underlying the Pathologies of Today’s Deep Learning Algorithms?
Research paper by Ben Goertzel on the current limitations of Deep Learning Algorithms, namely the misclassification of random images and of minuscule perturbations to correctly classified images.
This topic has been previously discussed, namely through a great paper by Christian Szegedy et al. on the Intriguing properties of neural networks
Tutorial - Recognizing music genres from cover images
Nice tutorial by Alexandre Passant in which he shows how he used Google Prediction API and Clarifai's services to train a classifier recognizing music genres by looking at the album cover. Some nice insights too.
Software tools & code
Neon deep learning software
Nervana Systems announced the release of its Python-based Neon deep learning software under an Apache open-source license.
The startup suggests that the software outperforms Nvidia’s cuDNN and Facebook’s Torch7 libraries. See benchmark
Keras - A Python library for Deep Learning prototyping
Nice little library to help you prototype and iterate quickly with your models and datasets. Define new nets in a few lines and start training / testing models. The library uses Theano under the hood, so it should prove OK in terms of performance.
UCSB team created a simple artificial neural circuit
A team at UC Santa Barbara announced the creation of a circuit of about 100 artificial synapses. The circuit was used to perform simple classification tasks, such as classifying images of letters.
Tutorial - Google Prediction API
Google's prediction API is a a RESTful interface to build Machine Learning models. The API is not new but so is this thorough tutorial.
For those not familiar, Google's API Prediction is meant to work as a “black box”. It's easy to get started but you get no control over what happens under the hood: your model configuration is restricted to specifying “Classification” vs. “Regression,” or providing a preprocessing Predictive Model Markup Language file and a set of weighting parameters in the case of categorical models.
Brains & Neurons
Look, your eyes are wired backwards - here's why
Human vision, from the retina to the visual cortex, has been a great source of inspiration to develop algorithms that recognize objects and "understand" scenes, through convolutional neural networks.
Many aspects of the human vision remain mysterious though, and one of those mysteries seems to have been explained. Interesting read.
Some thoughts
Visualization of a genetic algorithm
Nice visualization tool developed by Karsten Ahnert. Shows the beauty of genetic operators in a fun and useful way.
Finding Topics in Harry Potter using K-Means Clustering
This article is a bit technical but on a rather distracting topic: classification of Harry Potter chapters.
I must confess I never read any Harry Potter book, but I found the article rather amusing! :)
About
This newsletter is a weekly collection of the best news and resources on Artificial Intelligence and Machine Learning. If you find it worthwhile, please forward to your friends and colleagues, or share on your favorite network.
Share on Twitter · Share on Linkedin · Share on Google+
Suggestions or comments are more than welcome, just reply to this email. Thanks!
Artificial Intelligence Weekly
- 1 of 475
- Next Issue May 21st 2015