For the past few days I've been working on how to implement recursive neural networks in TensorFlow. Recursive neural networks (which I'll call TreeNets from now on to avoid confusion with recurrent neural nets) can be used for learning tree-like structures (more generally, directed acyclic graph structures). They are highly useful for parsing natural scenes and language; see the work of Richard Socher (2011) for examples. More recently, in 2014, Ozan İrsoy used a deep variant of TreeNets to obtain some interesting NLP results. In RNNs, at each time step the network takes as input its previous state s(t-1) and its current input x(t) and produces an output y(t) and a new hidden state s(t).

But not a drop to drink," is one of the more famous lines of poetry from Samuel Coleridge about a dehydrated sailor adrift at sea; but it could just be describing the data challenge faced by banks. There is no shortage of bank data, in fact the likes of Oracle and Accenture have built whole divisions in helping build and manage data. The challenge is that accessing and using this data is a minefield for banks due to internal organizational and regulatory compliance constraints, which makes taking advantage of new tools like machine learning all that more challenging. Therefore, what are the approaches that other banks have used to navigate around this and what best practices can be applied to take advantage of data? One of the primary challenges banks face is that data has piled up in different systems, making the concept of a'universal' customer record something of a joke.

It seems like every year Microsoft CEO Satya Nadella tweaks the company's technological direction. Microsoft was the productivity company. After the Microsoft Build 2017 developer conference in Seattle last week, Microsoft is now the intelligent cloud, intelligent edge company. A casual observer might think that Nadella can't get his story straight. That's the wrong frame of mind to take.

The short answer: To make their lives easier. Marketers have depended on tools and technology to automate their work and reduce manual effort for a while. Yet, there has always been a gap in terms of effort and quantifiable results. Intuition on the right audience and time to send messages aren't enough to answer a digital marketer's basic questions. Who should I reach out to? What should I send?

If you want to get a lot of retailers nodding their heads, ask them this: "Are you pursuing a personalization strategy?" You'll get a pretty resounding "yes!". In fact, personalization already drives a lot of activity on eCommerce sites. It often influences the products or offers featured on the home page, the order of products you see on a category or search results page, and makes product recommendations both on the site and in the digital marketing (email, retargeting) that follow. If you've ever been browsing a site and suddenly had the order of products change around on a product listing page, that was probably a decision driven by personalization.

The rise of artificial intelligence and robotics in the workplace is inevitable, but what does that mean for the recruitment industry? With robots building everything from our computers to our cars, there is always the fear that the human workforce will become redundant in the future. Researchers from Oxford University found that over the following 20 years, 35% of current jobs in the UK are at high risk of computerisation. Artificial intelligence offers benefits such as efficient management of time, speed, precision and costs less than hiring an employee to perform the same job. Although this may seem impressive, robots lack human emotion, judgement and the ability to think independently. These traits are important in many industries, one of them being recruitment.

Every day people use artificial intelligence; most without realizing the amount of impact AI has on their lives. Chances are better than good that you're using AI far more than you realize. Netflix, Google, credit card companies, video games, customer support, social media, and many other industries use AI to streamline their business model and improve the customer experience. This means that businesses large and small need to capitalize on what's quickly becoming a thriving industry. For perspective, the artificial intelligence market is expected to grow from its current $600 million valuation, to over $36 billion in 2025.

Our Western diet is famously bad for the circulatory system, but for a long time, people thought the damage stopped there. Then around 10 years ago, Terry Davidson, a behavioral neuroscientist, wondered whether our modern eating habits might also affect our brains. To test it out, he fed lab rats a diet high in saturated fats and sugars. He found that the animals had problems learning various memory tasks for which they'd get rewards. Their difficulties were probably linked to changes in the way blood reaches a portion of the brain called the hippocampus.

Google Assistant might have made its debut at Google I/O 2016, but this year's event was its real coming-out party. Over the past several months, Google has been steadily adding features to its AI chatbot, and now we can finally see a fuller vision for Assistant, and it goes far beyond asking questions about our day. It's my number one most-wanted feature, and will easily double the amount of time I spend using Assistant. Now, if I'm in the office or in bed, I won't have to break the silence with an inopportune "OK Google." I tried it out at I/O, and typing is just as fast as speaking--and in some cases even faster.

Dig below the surface of some of today's biggest tech controversies and you are likely to find an algorithm misfiring:[1] These errors are not primarily caused by problems in the data that can make algorithms discriminatory, or their inability to improvise creatively. No, they stem from something more fundamental: the fact that algorithms, even when they are generating routine predictions based on non-biased data, will make errors. We should not stop using algorithms simply because they make errors.[2] Without them, many popular and useful services would be unviable.[3] However, we need to recognise that algorithms are fallible, and that their failures have costs. Economics is the science of trade-offs, so why not think about this topic like economists? This is what I have done ahead of this blog, creating three simple economics vignettes that look at key aspects of algorithmic decision-making.[4] The two sections that follow give the gist of the analysis and its implications.