Neural networks are very powerful tools to classify data but they are very hard to debug. Indeed, they do a lot of computation with low level operations so they are like black boxes: we provide inputs and get outputs without any understanding on how the neural network is finding the results. Few years ago some scientists found ways to delve into the networks used for image categorization. Instead of doing backpropagation on weights like during the learning phase of a neural network, they did backpropagation on the images themselves: in the example below (edited from CS231n), considering x are inputs and w are weights, each learning step, the gradient (red) is applied to the x instead of the w. In this article, we will use the method and code from Google, Simonyan, Yosinski and Chollet to try to visualize the classes and convolutional layers learnt by popular neural networks. The code provided in this article uses the Keras library.

New drugs typically take 12-14 years to make it to market, with a 2014 report finding that the average cost of getting a new drug to market had ballooned to a whopping 2.6 billion. It's a topic I've covered before, with a study published earlier this year highlighting how automation could be used to reduce the cost of drug discovery by approximately 70%. It's an approach that a number of companies are taking to market. For instance, London based start-up Benevolent.AI utilizes complex AI to look for patterns in the scientific literature. They have already managed to identify two potential drug targets for Alzheimer's that has already attracted the attention of pharmaceutical companies.

Microsoft says it is going to "solve" cancer in the next 10 years. The company is working at treating the disease like a computer virus, that invades and corrupts the body's cells. Once it is able to do so, it will be able to monitor for them and even potentially reprogram them to be healthy again, experts working for Microsoft have said. The company has built a "biological computation" unit that says its ultimate aim is to make cells into living computers. As such, they could be programmed and reprogrammed to treat any diseases, such as cancer.

WASHINGTON – When Hillary Clinton attended her first major White House meeting on Russia in February 2009, the new secretary of state insisted that she wanted to play a leading role in President Barack Obama's effort to "reset" U.S. relations with Moscow. But while Clinton became implementer-in-chief for one of Obama's signature first-term initiatives, she was consistently more skeptical than most of his top aides about how far Russian leader Vladimir Putin was prepared to go in turning the page, according to current and former U.S. officials. That stance is indicative of how she will go about dealing with Moscow if she is elected U.S. president on Nov. 8, aides to both Clinton and Obama said. With U.S. relations with Moscow already plumbing post-Cold War lows, the aides and veteran Russia watchers said she will likely take a harder line than Obama or Republican nominee Donald Trump, who has praised Putin as a strong leader. Dealing with Putin, who is flexing his geopolitical muscle from Ukraine to Syria to cyberspace, will be among Clinton's biggest foreign policy challenges -- one made more daunting by the personal bad blood between them.

There are hundreds of new cancer drugs in development and new research published minute to minute, helping doctors treat patients with personalized combinations that target the specific building blocks of their disease. The problem is there's too much to read and too many drug combinations for doctors to choose the best option every time. Enter a Microsoft Research machine-learning project, dubbed Hanover, that aims to ingest all the papers and help predict which drugs and which combinations are most effective, according to the company. Researchers at Oregon Health & Science University's Knight Cancer Institute are working with Hanover's architect, Hoifung Poon, to use the system to find drug combinations effective in fighting acute myeloid leukemia, an often-fatal cancer where treatment hasn't improved much in decades. They include Jeff Tyner, and the institute's director, Brian Druker, best known for pioneering Gleevec, a blockbuster drug for a different type of leukemia now owned by Novartis, that's helped double those patients' five-year survival rate since the 1990s.

CRM company Salesforce, powered by the artificial intelligence of Einstein, is making additional in-roads into the contact center. Salesforce today unveiled Salesforce Einstein, which is being described as not only providing artificial intelligence for every company, but as creating the world's smartest customer relationship management (CRM) software. Whenever I cover a vendor announcement on No Jitter, one of my goals is to deliver value beyond the press release. And after listening to an hour long pre-briefing for press and analysts on the announcement last week, I was eager to get to work on coverage... Then I saw the press release. In 1,600 words, the Salesforce team has done an impressive job of laying out the Einstein vision and story.

Note from the Editor, Tricia Wang: Next up in our Co-designing with machines edition is Steven Gustafson (@stevengustafson), founder of the Knowledge Discovery Lab at the General Electric Global Research Center in Niskayuna, New York. In this post, he asked what is the role of humans in the future of intelligent machines. He makes the case that in the foreseeable future, artificially intelligent machines are the result of creative and passionate humans, and as such, we embed our biases, empathy, and desires into the machines making them more "human" that we often think. I first came across Steven's work while he was giving a talk hosted by Madeleine Clare Elish (edition contributor) at Data & Society, where he spoke passionately about the need for humans to move up the design process and to bring in ethical thinking in AI innovation. Steven is a former member of the Machine Learning Lab and Computational Intelligence Lab, where he developed and applied advanced AI and machine learning algorithms for complex problem solving.

According to market research firm Forrester Research, more than 6% of jobs currently performed by human beings will be taken over by robots in the next five years. We all knew the day was coming when robots would become more intelligent and start replacing human beings at the workplace, especially in jobs that need to follow a set pattern or jobs that are repetitive in nature. But that day seems closer than ever with retailers like Amazon (AMZN) and Wal-Mart (WMT) seemingly on the cusp of using automation on a large scale. For a company like Amazon, automation is nothing new. They already have robots manning their warehouses, and the company has been working for a while testing drone deliveries in the U.K. The U.S. government refused permission for drone testing, so Amazon is doing it across the pond.

HANOI – Younger students in Vietnam are learning Japanese after language classes were introduced at five elementary schools in Hanoi and Ho Chi Minh City this month. The schools are now offering Japanese lessons from the third grade, and classes are expected to be rolled out at elementary schools elsewhere, too. The Vietnamese education system has five grades in elementary school, four grades in junior high school and three years in high school. Japanese language classes are already offered in junior high and high school. "I became interested in Japanese language after reading the'Inazuma Eleven' manga series," said one student at Chu Van An Elementary School in Hanoi, which kicked off its Japanese language class on Thursday.

It's not out of the realm of possibility for fully self-driving cars to hit public roads in the next few years, what with Google, Tesla, Uber and other big-name companies working on the technology. Now, the US government has officially thrown its weight behind the technology and released a Federal Automated Vehicles policy. In it, the Department of Transportation outlines a set of 15 safety assessment objectives manufacturers have to meet to ensure their vehicles can meet regulators' requirements. It also clarifies the division of state and federal responsibilities, as well as the regulatory tools the DOT plans to use. According to the department, the policy is "rooted in [its] view that automated vehicles hold enormous potential benefits for safety, mobility and sustainability."