By experimenting, computers are figuring out how to do things that no programmer could teach them.

Marketing and sales seemed to be much easier in the past. Customers simply visited a retail shop, where they could ask a knowledgeable salesperson about a product they discovered in a local newspaper. In recent years, the ubiquity of the internet and a state-of-the-art technology changed everything. Customers became prosumers, well informed about the product before the purchase. And what is even more important, customers frequently use a variety of channels: online and traditional stores, mobile apps, online auctions, price comparison websites, social media and more. Today, in spite of all the available technologies, life is more challenging for both marketers and salespeople.

Yuval Noah Harari's first book, Sapiens, was an international sensation. The Israeli historian's mind-bending tour through the triumph of Homo sapiens is a favorite of, among others, Bill Gates, Mark Zuckerberg, and Barack Obama. His new book, Homo Deus: a Brief History of Tomorrow, is about what comes next for humanity -- and the threat our own intelligence and creative capacity poses to our future. I spoke with Harari recently for my podcast, The Ezra Klein Show. To hear our whole conversation, subscribe on iTunes (or wherever you get your podcasts) or stream it off SoundCloud. In this excerpt, which has been edited for length and clarity, Harari and I discuss the rise of artificial intelligence, whether digital consciousness is a necessary byproduct of digital intelligence, and what it will all mean for human beings. As you'll see, I'm a bit less convinced than Harari is that the computers are coming for our jobs, and that human beings are on the edge of economic uselessness.

WASHINGTON – The world's largest manufacturer of civilian drones is proposing that the craft continually transmit identification information to help government security agencies and law enforcement figure out which might belong to rogue operators. DJI, a Chinese company, said in a paper released Monday that radio transmissions of an identification code, possibly the operator's Federal Aviation Administration's registration number, could help allay security concerns while also protecting the operator's privacy. The paper suggests steps that can be taken to use existing technologies to develop an identification system, and that operators could include more identification information in addition to a number if they wish. Anyone with the proper radio receiver could obtain those transmissions from the drone, but only law enforcement officials or aviation regulators would be able to use that registration number to identify the registered owner. Law enforcement agencies and the U.S. military raised security concerns last year after FAA officials proposed permitting more civilian drone flights over crowds and densely populated areas.

Microsoft (MSFT) ended 2016 in the green, up more than 10%. Moreover, the stock is off to a good start this year, with a 5% rise year to date. As a matter of fact, the company's upward movement started in early 2013. Since then, it has continued rewarding investors every year. Microsoft reported second-quarter results in January. The company's revenue came in at $26 billion, again exceeding analysts' estimate by $790 million.

The other day, I went to meet someone in downtown Sydney, Australia. On my way, back on the local train, I looked at my mobile to check my emails and found a message asking me whether I would like to meet the person I had just connected with on my LinkedIn network. So, was this some form of artificial intelligence (AI) at play? We now live in a brave new world where AI is the next frontier. We keep hearing about bots, chatbots, teacherbots, digital assistants, machine learning, deep learning and many more such words and often wonder what do they mean.

Part of the money will help secure research grants designed to stop a'brain drain' and ensure that top computing talent and academics remain in the country, avoiding the lure of more lucrative funding for projects in other nations. The money will also help to nurture post-graduate trainees and researchers who wish to study artificial intelligence. There will also be a co-ordinated attempt to bring together Canada's main centers of computer expertise. These are located in Montreal, Toronto-Waterloo and Edmonton. The funding strategy will be channeled through the Canadian Institute for Advanced Research.

In many scientific and engineering applications, we are tasked with the optimisation of an expensive to evaluate black box function $f$. Traditional settings for this problem assume just the availability of this single function. However, in many cases, cheap approximations to $f$ may be obtainable. For example, the expensive real world behaviour of a robot can be approximated by a cheap computer simulation. We can use these approximations to eliminate low function value regions cheaply and use the expensive evaluations of $f$ in a small but promising region and speedily identify the optimum. We formalise this task as a \emph{multi-fidelity} bandit problem where the target function and its approximations are sampled from a Gaussian process. We develop MF-GP-UCB, a novel method based on upper confidence bound techniques. In our theoretical analysis we demonstrate that it exhibits precisely the above behaviour, and achieves better regret than strategies which ignore multi-fidelity information. Empirically, MF-GP-UCB outperforms such naive strategies and other multi-fidelity methods on several synthetic and real experiments.

In high-dimensional and/or non-parametric regression problems, regularization (or penalization) is used to control model complexity and induce desired structure. Each penalty has a weight parameter that indicates how strongly the structure corresponding to that penalty should be enforced. Typically the parameters are chosen to minimize the error on a separate validation set using a simple grid search or a gradient-free optimization method. It is more efficient to tune parameters if the gradient can be determined, but this is often difficult for problems with non-smooth penalty functions. Here we show that for many penalized regression problems, the validation loss is actually smooth almost-everywhere with respect to the penalty parameters. We can therefore apply a modified gradient descent algorithm to tune parameters. Through simulation studies on example regression problems, we find that increasing the number of penalty parameters and tuning them using our method can decrease the generalization error.

Automatic Music Transcription (AMT) is one of the oldest and most well-studied problems in the field of music information retrieval. Within this challenging research field, onset detection and instrument recognition take important places in transcription systems, as they respectively help to determine exact onset times of notes and to recognize the corresponding instrument sources. The aim of this study is to explore the usefulness of multiscale scattering operators for these two tasks on plucked string instrument and piano music. After resuming the theoretical background and illustrating the key features of this sound representation method, we evaluate its performances comparatively to other classical sound representations. Using both MIDI-driven datasets with real instrument samples and real musical pieces, scattering is proved to outperform other sound representations for these AMT subtasks, putting forward its richer sound representation and invariance properties.