Samsung is rumored to be making its own smart speaker that will be powered by the Bixby smart assistant. However, it looks like the South Korean company has doubts on the marketability of such a device. "Samsung currently does not view Al speakers as marketable, as the global market is already dominated by unbeatable Amazon and the Korean market is too small to make profits," an anonymous source told The Korea Herald. "More importantly, Samsung cannot afford to focus on the uncertain market, as most of its AI specialists -- whose number is much less than that of the US tech giants -- are currently going all out to develop the Bixby version in English." Samsung might simply be taking a "wait-and-see attitude" towards its ambitions for a smart speaker, mobile industry specialist and researcher Kim Jong-ki suggested.

Samsung Electronics said its Bixby voice assistant for smartphones will start speaking English but only in two countries: South Korea and the U.S. The South Korean tech giant said starting Wednesday, users of the Galaxy S8 smartphones can speak in American English to their phones to turn on the flashlight, take a selfie or make the phone search for say, pictures from a summer vacation, and create an album. Samsung did not say why it could not make Bixby in English available to users outside South Korea or the U.S. or when it will become available in other languages in other countries. The service, comparable to Apple's Siri or Google Assistant, was previously available only in Korean. Samsung says its virtual assistant can be activated by saying "Hi Bixby" to the phone or by clicking a button on the side of the S8 phones. The rollout of the Bixby English version was delayed by a couple of months.

We should have been more specific. Organizers say this event is the first global robotics competition specifically for high school students. There have been other robotics competitions with teams from multiple countries.]

STEVE INSKEEP: Republicans promised for years to repeal and replace the Affordable Care Act. In fact, they said they'd replace it with something better. President Trump says he would now rather just repeal. Trump said that last night after a Senate bill to replace Obamacare collapsed. Two more Republican senators objected to it. And since they were trying to pass it with GOP votes alone, it was assured of failure.

The voice component of Samsung's Bixby assistant has been a long time in coming. The company was quick to boast about its AI helper at the Galaxy S8 launch, but revealed that the signature voice feature wouldn't arrive until later in the spring... and even then, it only had a full launch in South Korea. Americans had to make do with a preview. At last, though, it's becoming widely available: Samsung is officially rolling out Bixby's voice assistance to S8 and S8 Plus owners in the US. Every American with one of the flagship phones will get to talk to Bixby once an update arrives "tonight."

A team of Afghan girls whose plight resounded with the world won a silver medal for "courageous achievement" at an international robotics contest in the United States, with judges praising the group's "can-do attitude". The First Global Challenge event in Washington ended on Tuesday, having attracted teams of teenagers from more than 150 nations. But all eyes were on the squad of girls from Afghanistan, who had twice travelled 800 kilometres to the American embassy in Kabul, only to have their visa applications turned down. They were finally granted entry with just one week to go until the event began after their story had gone viral. In an interview with Al Jazeera, before US officials decided to allow them in the country, team member Rodaba Noori said: "We wanted to show our talents to the world so they would know that we do have skills."

The Association for Advancing Automation (A3) cites that between 2010 and 2016, 136,748 robots were shipped to the US --the most in any seven-year period in the US robotics industry. At the same time, US manufacturing employment increased by 894,000 and the unemployment rate fell from 9.8% to 4.7%. Yet manufacturers, robotics associations, ethicists and media pundits are still fighting the robotics and jobs issue. Brett Brune, Editor in Chief of Smart Manufacturing magazine, argues that "the hand-wringing around robotics and jobs in the US really needs to stop." Manufacturers around the world, including in China, are busy figuring out how quickly to acquire robots.

I've been at every one, from the tiny one with perhaps 100-200 people to this one with 1,400 that fills a large ballroom. Tuesday morning did not offer too many surprises. The first was an announcement by Toyota Research Institute of a $100M venture fund. Toyota committed $1B to this group a couple of years ago, but surprisingly Gil Pratt (who ran the DARPA Robotics Challenge for humanoid-like robots) has been somewhat a man of mixed views, with less optimistic forecasts. Different about this VC fund will be the use of DARPA like "calls."

Trump intervened and enabled all the teams headed to Washington, DC for the F.I.R.S.T Global Robotics Championship, whose visas had been held up or denied, to get their visas. Some visas were received as late as two days before the event. Although the Afghan team got all the press, the team from Gambia was also denied when they first applied. The three-day event which started Sunday evening in Washington, DC with opening ceremonies, has teams from 157 countries (including the Afghanistan and Gambia teams) -- and some multinational teams representing continents. FIRST has organized competitions for many years but this is the first year it is hosting an international competition.

Algorithm-dependent generalization error bounds are central to statistical learning theory. A learning algorithm may use a large hypothesis space, but the limited number of iterations controls its model capacity and generalization error. The impacts of stochastic gradient methods on generalization error for non-convex learning problems not only have important theoretical consequences, but are also critical to generalization errors of deep learning. In this paper, we study the generalization errors of Stochastic Gradient Langevin Dynamics (SGLD) with non-convex objectives. Two theories are proposed with non-asymptotic discrete-time analysis, using Stability and PAC-Bayesian results respectively. The stability-based theory obtains a bound of $O\left(\frac{1}{n}L\sqrt{\beta T_k}\right)$, where $L$ is uniform Lipschitz parameter, $\beta$ is inverse temperature, and $T_k$ is aggregated step sizes. For PAC-Bayesian theory, though the bound has a slower $O(1/\sqrt{n})$ rate, the contribution of each step is shown with an exponentially decaying factor by imposing $\ell^2$ regularization, and the uniform Lipschitz constant is also replaced by actual norms of gradients along trajectory. Our bounds have no implicit dependence on dimensions, norms or other capacity measures of parameter, which elegantly characterizes the phenomenon of "Fast Training Guarantees Generalization" in non-convex settings. This is the first algorithm-dependent result with reasonable dependence on aggregated step sizes for non-convex learning, and has important implications to statistical learning aspects of stochastic gradient methods in complicated models such as deep learning.