Say we want to build system to detect dresses in images using a deep convolutional network. What we have is a database of 64x128 pixels images that either fully contain a dress or another object (a tree, the sky, a building, a car…). With that data we train a deep convolutional network and we end up successfully with a high accuracy rate in the test set. The problem comes when trying to detect dresses on arbitrarily large images. As images from cameras are usually far larger than 64x128 pixels, the output of the last convolutional layer will also be larger. Thus, the fully connected layer won't be able to use it as the dimensions will be incompatible.

This week, a lot of attention was given during the kick-off of the annual Google I/O conference to three of this year's key technologies: artificial intelligence, conversational interfaces and virtual reality. The attractive force of Google's technologies was in full display with the opening viewed by more than 1 million users in China alone. Google's CEO, Sundar Pichai, kicked it off by demonstrating that Google had not missed out on the shift to mobile and that 50% of all search engine queries were already today made from phones, including 20% through voice recognition in the United States. Already, the results of a mobile study that were unveiled extend beyond a simple list of links to include "cards" used to preview related content (photos, results, artists) without having to exit the search engine. The head of Google went on to highlight his company's progress with respect to artificial intelligence and concrete applications for search engines.

Google plans on opening a research and development center in Novi, Michigan, just outside of Detroit. Or to put it another way: more self-driving cars are heading to Motor City. "Many of our current partners are based here," read a note on the Google page for Google's Self-Driving Car Project, "so having a local facility will help us collaborate more easily and access Michigan's top talent in vehicle development and engineering." Based on the photo above, though, it seems pretty empty at the moment. Google claims Rousch, Bosch, Continental, FRIMO, LG Electronics, Prefix, and RCO among its partners on the self-driving car project.

It's no longer up for debate that AI is set to have a major impact on most businesses, if it isn't already--and any company that wants to stay ahead must figure out how to integrate the new technology into its structure. But how is a successful AI platform built? Perhaps best known as the guy who introduced Steve Jobs and Steve Wozniak, Bill Fernandez speaks out on Apple's founding magic, how love built the first Mac, and the interface of the future. In Mehanna's session, he explained how Facebook developed its own machine learning platform, and how Facebook employees are using it. In 2012, Mehanna said, Facebook's AI platform was "a snowball of complexity"--a system that slowed progress down significantly.

At any given moment, pharmaceutical firms have a massive library of compounds and no clue what to do with them. Tucked away in extensive collections of synthetic and theoretical drug banks lie hidden gems -- drugs to treat perhaps even the most devastating diseases -- but identifying them is a pain: testing can take years, even decades, and often researchers aren't even sure what they're looking for. What they need is a way to sort through the duds -- and now, it's looking like artificial intelligence can help. Scientists from Insilico Medicine, a bioinformatics firm, have figured out how to teach A.I. to predict the therapeutic use of new drugs before they're even tested. Publishing their work today in the journal Molecular Pharmaceutics, they discuss their A.I.'s training regimen, which involves taking in huge amounts of data from experiments on human cells using known drugs.

The R platform for statistical computing is perhaps the most popular and powerful platform for applied machine learning. The caret package in R has been called "R's competitive advantage". It makes the process of training, tuning and evaluating machine learning models in R consistent, easy and even fun. In this post you will discover the caret package in R, it's key features and where to go to learn more about it. Caret was built on a key philosophy in machine learning, that of the no free lunch theorem.

The smartphone boom upended Google's advertising profit engine and it took years for the Internet giant to adjust to the new mobile world. The next wave of computing will be an even bigger challenge. At Google's I/O developer conference this week near its Silicon Valley headquarters, the company unveiled new technology that will rely less and less on physical devices with screens to deliver information and services to consumers. Google hopes these advances will capture the human attention its business depends upon, and then it can figure out how to make money later, one executive said. Google Home will sit in living rooms sucking in voice-based queries and delivering verbal answers from an artificially intelligent "Google assistant."

The News and information ecosystem is in the midst of change -- again. Mobile-first consumption is on the rise, smart homes are becoming mainstream and connected cars will soon take over the roads of major cities around the world. Smart devices will require "smart content." It's only a matter of time before artificial intelligence (AI) becomes the backbone of the media industry of the future. Today, most people find information via search or social. And while these two channels are radically different in functionality, they have one thing in common -- any given article surfaced through these platforms is exactly the same for everyone in the world.

According to MarketandMarkets, the artificial intelligence (AI) market is estimated to grow from 419.7 million in 2014 to 5.05 billion by 2020, growing at a CAGR of 53.65% from 2015 to 2020. The Media and Advertising sector is expected to drive the growth of AI during this period. IBM, Microsoft, and Google are key players in the market, and now Salesforce is trying to make inroads into it. For the first quarter of fiscal 2017, Salesforce's revenue grew 27% over the year to 1.92 billion, above analyst estimate of 1.89 billion. Net income was 38.8 billion or 0.06 per share. Non GAAP EPS was 0.24, beating analyst forecast of 0.25.

Many times after my talks, people ask me whether algorithms could theoretically be better at making a diagnosis than doctors. With my doctor's cap on, I must defend the art of medicine. But as a medical futurist I need to tell my honest views. Making a diagnosis is an art. We humans are not engineering products, and therefore measuring a few parameters and tweaking a few knobs will not diagnose and cure our diseases.