There is interesting coverage into how AI will shape Africa in the future. It looks into financial, healthcare, logistics, education and agritech fields and discusses broad based trends. Link.
An older paper but a funny one a friend sent over. Checkout the Stopping Gan Violence — Generative Unadversarial Networks. Very funny reading with a sample included below. Link
Our work is connected to a range of adversarial work in both the machine learning and the machine forgetting communities. To the best of our knowledge Smith & Wesson (1852) were the first to apply GUNs to the problem of generative modelling, although similar ideas have been explored in the context of discriminative modelling as far back as the sixteenth century by Fabbrica d’Armi Pietro Beretta in an early demonstration of one-shot learning. Unfortunately, since neither work evaluated their approach on public benchmarks (not even on MNIST), the significance of their ideas remains under appreciated by the machine learning community.
~ Raspberry PI support is available on TensorFlow which is a really big deal. Checkout the latest blog post from my colleague Pete Warden. Link.
~ Photoreal Avatar Generative Adversarial Network. Yes, someone finally took the name paGAN (get it). Interestingly, there is some controversy around the company and the nature of the research — check that out here too. Link.
~ Deepmind published a great article on the interrelationships between visual and audio objects. Essentially, they construct “an audio-visual correspondence learning task that enables visual and audio networks to be jointly trained from scratch”. Link
~ OpenAI played its very well trained Dota system against humans and won 2 or 3 rounds. The humans winning the last. Link
~ There is a supercool Neural Style Transfer blog. Basically, the ability to creating art with Deep Learning using tf.keras and eager execution that was released. Link.
~ The close coupling of Academia with Researchers to create the future of AI. A interesting perspective from Yann LeCun Yann LeCun which asserts that companies must let AI experts split their time between academia and industry to drive innovation forward. Link.
~ AutoKeras is a really awesome project that uses neural architecture search to automatically develop Keras Models. There is a lot of great content on the Github page. Check it here.
~ A neural aware mobile platform — an automated neural architecture search approach for designing resource-constrained mobile CNN models. Interesting paper that discusses the trade offs between accuracy and latency. Link