Chinese facial recognition technology developer SenseTime Group Ltd said it has raised $620 million in a second round of funding in just two months, adding that it continued to be the world's most valuable unicorn in artificial intelligence. The financing, which values the company at more than $4.5 billion, was led by Fidelity International, Hopu Capital, Silver Lake and Tiger Global and follows a $600 million funding round in April led by Alibaba Group. SenseTime, which became profitable in 2017, said in a statement it would invest further in research and development, and in securing talented researchers after the latest round of funding. SenseTime, based in both Hong Kong and Beijing, develops technology that employs AI to quickly identify and analyze identities using cameras, and has been used by Chinese authorities to track and capture suspects in public spaces such as airports and festivals.

"I don't think Homo sapiens-type people will exist in 10 or 20 years' time," Getnet Assefa, 31, speculates as he gazes into the reconstructed eye sockets of Lucy, one of the oldest and most famous hominid skeletons known, at the National Museum of Ethiopia. "Slowly the biological species will disappear and then we will become a fully synthetic species," Assefa says. "I believe [we] can inspire the Ethiopian youth to actually get really engaged in AI and feel like it's their thing." "Perception, memory, emotion, intelligence, dreams -- everything that we value now -- will not be there," he adds. Assefa is a computer scientist, a futurist, and a utopian -- but a pragmatic one at that. He is founder and chief executive of iCog, the first artificial intelligence (AI) lab in Ethiopia, and a stone's throw from the home of Lucy.

Inventor and artist Aaron Taylor Kuffner's artistic robotic orchestra, Gamelatron fuses technology, sound, sculpture, and engineering to create a visceral experience based on acoustic resonance and robotic technology. Based on ancient Indonesian bronze gongs called the gamelan, Kuffner brings the past to the present to create a kinetic sculpture. He works with master craftsmen from Indonesia to fabricate the gongs and then alters their shape and then adds a finish to create unique pieces of sculpture. He crafts each robotic part and electronic components by hand to create a'digital brain,' (made of pulse width modulation chips, circuit boards, etc.) installs mechanical mallets and builds custom mounts for each sound sculpture. The circuit boards Kuffner uses today are made in partnership with Lumigeek.

The tech is pretty cool, but don't let new developments in partially self-driving cars distract you from your responsibilities behind the wheel. A Tesla that the driver said was in Autopilot mode struck a parked police vehicle in Laguna Beach, Calif. SAN FRANCISCO -- Tesla's innovative and controversial Autopilot software -- which powers the partially self-driving features of its electric cars -- is most often used for highway driving, according to the initial findings of an MIT study using volunteer owners. The research, shared at a conference in Cambridge, Mass. Wednesday, came a day after the latest crash of a Tesla using Autopilot, and as two consumer groups renewed criticism of the software's name and marketing, which they say dangerously misleads drivers.

Black swans are the ultimate outliers. They have the ability to surprise and disrupt the status quo. I was recently in Shenzhen and was permitted access to Huawei's campus. I know I didn't see it all, but I saw enough to get me thinking. I had heard lots of stories, but reality was even more interesting. Seeing and talking to the people gave me new insights. I'd heard that in China, tech employees worked 10 hours straight a day. The offices, campus and the university (yes, a University where all employees study) are perhaps even more modern and inviting than many I've seen in the United States.

I'm reading "Artificial Intelligence -- The Revolution Hasn't Happened Yet" by Berkeley computer science professor Michael I Jordan. It is a quick, light, read but thought provoking. His point is that once the dust settles after the hype around ubiquitous Artificial Intelligence (AI), we'll realize that what is being called AI is less general purpose intelligence than a collection of powerful tools for augmenting human intelligence (e.g., Siri, what he calls intelligence augmentation or IA) and the intelligent infrastructure (II) that makes these tools possible. There are also some good historical anecdotes in the article. Whether or not you agree with Jordan's point in this piece, I'd still encourage you to check out the reading list he's long suggested to his post docs and grad students, it is a gem.

Susceptibility of deep neural networks to adversarial attacks poses a major theoretical and practical challenge. All efforts to harden classifiers against such attacks have seen limited success. Two distinct categories of samples to which deep networks are vulnerable, "adversarial samples" and "fooling samples", have been tackled separately so far due to the difficulty posed when considered together. In this work, we show how one can address them both under one unified framework. We tie a discriminative model with a generative model, rendering the adversarial objective to entail a conflict. Our model has the form of a variational autoencoder, with a Gaussian mixture prior on the latent vector. Each mixture component of the prior distribution corresponds to one of the classes in the data. This enables us to perform selective classification, leading to the rejection of adversarial samples instead of misclassification. Our method inherently provides a way of learning a selective classifier in a semi-supervised scenario as well, which can resist adversarial attacks. We also show how one can reclassify the rejected adversarial samples.

We present a new algorithm for stochastic variational inference that targets at models with non-differentiable densities. One of the key challenges in stochastic variational inference is to come up with a low-variance estimator of the gradient of a variational objective. We tackle the challenge by generalizing the reparameterization trick, one of the most effective techniques for addressing the variance issue for differentiable models, so that the trick works for non-differentiable models as well. Our algorithm splits the space of latent variables into regions where the density of the variables is differentiable, and their boundaries where the density may fail to be differentiable. For each differentiable region, the algorithm applies the standard reparameterization trick and estimates the gradient restricted to the region. For each potentially non-differentiable boundary, it uses a form of manifold sampling and computes the direction for variational parameters that, if followed, would increase the boundary's contribution to the variational objective. The sum of all the estimates becomes the gradient estimate of our algorithm. Our estimator enjoys the reduced variance of the reparameterization gradient while remaining unbiased even for non-differentiable models. The experiments with our preliminary implementation confirm the benefit of reduced variance and unbiasedness.

In statistics and machine learning, approximation of an intractable integration is often achieved by using the unbiased Monte Carlo estimator, but the variances of the estimation are generally high in many applications. Control variates approaches are well-known to reduce the variance of the estimation. These control variates are typically constructed by employing predefined parametric functions or polynomials, determined by using those samples drawn from the relevant distributions. Instead, we propose to construct those control variates by learning neural networks to handle the cases when test functions are complex. In many applications, obtaining a large number of samples for Monte Carlo estimation is expensive, which may result in overfitting when training a neural network. We thus further propose to employ auxiliary random variables induced by the original ones to extend data samples for training the neural networks. We apply the proposed control variates with augmented variables to thermodynamic integration and reinforcement learning. Experimental results demonstrate that our method can achieve significant variance reduction compared with other alternatives.

We study a distributed estimation problem in which two remotely located agents, Alice and Bob, observe an unlimited number of i.i.d. samples corresponding to different parts of a random vector. Alice can send $k$ bits on average to Bob, who in turn wants to estimate the cross-correlation matrix between the two parts of the vector. In the case where the agents observe jointly Gaussian scalar random variables with an unknown correlation $\rho$, we obtain two constructive and simple unbiased estimators attaining a variance of $\frac{1-\rho^2}{2k\ln 2}$, which coincides with a known but non-constructive random coding result of Zhang and Berger. We extend our approach to the vector Gaussian case, which has not been treated before, and construct an estimator that is uniformly better than the scalar estimator applied separately to each of the correlations. We then show that the Gaussian performance can essentially be attained even when the distribution is completely unknown. This in particular implies that in the general problem of distributed correlation estimation, the variance can decay at least as $O(1/k)$ with the number of transmitted bits. This behavior is however not tight: we give an example of a rich family of distributions where a slightly modified estimator attains a variance of $2^{-\Omega(k)}$.