A quantum computer version of a single artificial neural network neuron was tested on an actual quantum computer with 2 qubits.

A new technique, called Monte Carlo Neural Fictitious Self Play (MC-NFSP), improves the performance of AIs in "imperfect information" games like Texas Hold'em and multiplayer video games.

TripAdvisor just rolled out a deep learning recommendation system.

A neural network did better than a physics-based model at predicting how a vehicle will maneuver on a variety of road surfaces ranging from dry to snow without the need for explicit road road-tire friction coefficients.

"This robot, like many others, is designed to tolerate the dynamics of the unstructured world. But instead of just tolerating dynamics, can robots learn to use them advantageously, developing an 'intuition' of physics that would allow them to complete tasks more efficiently?"

"To explore this concept, we worked with researchers at Princeton, Columbia, and MIT to develop TossingBot: a picking robot for our real, random world that learns to grasp and throw objects into selected boxes outside its natural range. We find that by learning to throw, TossingBot is capable of achieving picking speeds that are twice as fast as previous systems, with twice the effective placing range. TossingBot jointly learns grasping and throwing policies using an end-to-end neural network that maps from visual observations (RGB-D images) to control parameters for motion primitives. Using overhead cameras to track where objects land, TossingBot improves itself over time through self-supervision."

Yann LeCun, Yoshua Bengio, and Geoffrey Hinton won the Turning Award, the top award in computer science, for their work on neural networks.

"The computer industry has been busy in recent years trying to figure out how to speed up the calculations needed for artificial neural networks."

"Operating these networks typically hinges on how fast the hardware can perform matrix multiplications, which are used to determine the output of each artificial neuron -- its 'activation' -- for a given set of input values."

"The Rice University researchers, like others before them, realized that many of the neurons in a given layer of a neural network have very small activations, meaning that their outputs don't contribute to what gets computed by subsequent layers. So if you knew which neurons those were, you could just ignore them."

"We treat this as a search problem."

The European Ministry of Justice has asked its Chief Information Officer to design an automated judge that can adjudicate small claims disputes up to $8,000, clearing a backlog of cases for judges and court clerks. "In this system, the two parties will upload relevant information and documents, and the artificial intelligence will make a ruling. The decision can still be appealed to be seen by a human judge, though."

Estonia's chief information officer says the country has deployed AI or machine learning in places where an algorithm has replaced government workers. "Inspectors no longer check on farmers who receive government subsidies to cut their hay fields each summer. Satellite images taken by the European Space Agency each week from May to October are fed into a deep learning algorithm originally developed by the Tartu Observatory. The images are overlaid onto a map of fields where farmers receive the hay cutting subsidies to prevent them from turning forests over time." They say the system has already saved the country $755,000 in its first year but an inspector still has to drive out and check because the system still isn't perfect.

Estonia may have an easier time at implementing these kinds of automated solutions because already their 1.3 million population use a national ID card that allows them digital services for voting and tax filing. Outside experts found a vulnerability in Estonia's ID system in 2017, but it was fixed and the ID cards replaced, and only led to some embarrassment. Government officials say the country hasn't had a major data breach or theft since it began its digital drive in the early 2000s.

The video also describes a new report from Spain's Instituto de Empresa (IE) University shows that one in four Europeans trusts artificial intelligence to make political choices more than living breathing politicians. "70% believe that if not appropriately controlled, new technologies will cause more harm than good to society in the coming decade."

"The researchers point out a strange paradox. While the public are disillusioned with the government and favor AI in the driver's seat, people are deeply fearful of advancements in tech, particularly increased automation. 70% of respondents called for politicians to take strong policy measures to curb automation and tackle job displacement, even if that means slowing technological progress."

"40% of Europeans think that the company they work for will disappear in the next 10 years, unless they implement profound and fast changes, and that most Europeans believe governments should intervene to limit automation and tackle negative effects on society."

"60% of university graduates think their schools did not train them well to deal with technological transformation. The age group that felt the strongest about this was people over 50, at 68 to 69%."

Efficient Neural Architecture Search (ENAS) "employs reinforcement learning to build convolutional neural networks (CNNs) and recurrent neural networks (RNNs)." The authors proposed a predefined neural network to generate new neural networks guided by a reinforcement learning framework using macro and micro search. That's right -- a neural network building another neural network."

"In ENAS, there are 2 types of neural networks involved: Controller -- a predefined RNN, which is a long short-term memory (LSTM) cell, and child model -- the desired CNN for image classification."

"The controller controls or directs the building of the child model's architecture by 'generating a set of instructions' (or, more rigorously, making decisions or sampling decisions) using a certain search strategy. These decisions are things like what types of operations (convolutions, pooling etc.) to perform at a particular layer of the child model. Using these decisions, a child model is built."

"This particular child model is then trained to convergence (~95% training accuracy) using stochastic gradient descent to minimise the expected loss function between the predicted class and ground truth class (for an image classification task)."

"Then, we update the controller's parameters using REINFORCE, a policy-based reinforcement learning algorithm, to maximise the expected reward function which is the validation accuracy."

"When talking about how neurons work, we usually end up with the sum-up-inputs-and-spit-out-spike idea. In this idea, the dendrites are just a device to collect inputs. Activating each input alone makes a small change to the neuron's voltage. Sum up enough of these small changes, from all across the dendrites, and the neuron will spit out a spike from its body, down its axon, to go be an input to other neurons. It's a handy mental model for thinking about neurons. It forms the basis for all artificial neural networks. It's wrong. Those dendrites are not just bits of wire: they also have their own apparatus for making spikes."

"In visual cortex, we know that many neurons respond only to things in the world moving at a certain angle (like most, but by no means all of us, they have a preferred orientation). Some neurons fire their spikes to things at 60 degrees; some at 90 degrees; some at 120 degrees. But when we record what their dendrites respond to, we see responses to every angle. The dendrites know a hell of a lot more about how objects in the world are arranged than the neuron's body does."