Last Week in AI

Every week, my team at Invector Labs publishes a newsletter to track the most recent developments in AI research and technology. You can find this week’s issue below. You can sign up for it below. Please do so, our guys worked really hard on this:

From the Editor: Paradoxes that Puzzle AI Systems

Paradoxes are one of the elements of human cognition that are constantly present in AI systems. Conceptually, a paradox is a statement that leads to an apparent self-contradictory conclusion based on the original premises of the problem. The thing about paradoxes is that they have to do more with philosophical viewpoints and just common sense than with mathematics and statistics. The simple fact of judging an outcome as a paradox assumes a predefined criteria about what that outcome suppose to be. While some paradoxes are really easy to explain using match and data, others are firmly rooted in human pillars such as morals or ethics. As AI systems processes more complex datasets, identifying and understanding paradoxes becomes more relevant.

The field of statistics and mathematics is full of interesting paradoxes that are present in datasets used to train machine learning models. To some extent, you can argue that AI systems don’t necessarily see those “contradictions” are paradoxes since they don’t operate with a predefined criteria. However, we can’t make the same argument for paradoxes that face moral or ethical arguments. In those areas human intervention is still very much required. This week, our team at Invector Labs published a good summary of some of the most common paradoxes in the current generation of machine learning solutions.

Now let’s take a look at the core developments in AI research and technology this week:


Microsoft Researchers published a paper proposing a deep learning technique to improve editing processes.

>Read more in this blog post from Microsoft Research

Uber unveiled their AI research publication website with papers about image analysis, natural language processing and self-driving vehicles.

>Read more at the new Uber Research website

Facebook published a summary of the AI techniques they are leveraging to improve the privacy of the platform.

>Read more in this blog post from the Facebook engineering team

Cool Tech Releases

PyTorch added a few cool new tools to its ecosystem including BoTorch and Ax which streamling experimentation in the platform.

>Read more in this blog from the PyTorch team

Google released Landmarks-v2, the world’s largest dataset for landmark recognition.

>Read more in this blog post from the Google AI Research team

The Linkedin Engineering team published an insightful post about how they leverage machine learning to help users smarter decision.

>Read more in this blog post from the LinkedIn Engineering team

AI in the Real World

Japanese firm DataGrid showed an AI system that can generate super realistic fashion models which are be used in digital advertising.

>Read more in this coverage from The Next Web

MIT AI Researchers published a study that show how vision models can be used to stimulate specific areas of the brain.

>Read more in this article from MIT News

Wired published a great interview of best seller author Yuval Noah Harari and AI legend Fei-Fei Li about the benefits and challenges of AI.

>Read the entire interview at Wired