One of the main difficulties in analyzing neural networks is the non-convexity of the loss function which may have many bad local minima. In this paper, we study the landscape of neural networks for binary classification tasks. Under mild assumptions, we prove that after adding one special neuron with a skip connection to the output, or one special neuron per layer, every local minimum is a global minimum.

One of the main difficulties in analyzing neural networks is the non-convexity of the loss function which may have many bad local minima. In this paper, we study the landscape of neural networks for binary classification tasks. Under mild assumptions, we prove that after adding one special neuron with a skip connection to the output, or one special neuron per layer, every local minimum is a global minimum.

One of the main difficulties in analyzing neural networks is the non-convexity of the loss function which may have many bad local minima. In this paper, we study the landscape of neural networks for binary classification tasks. Under mild assumptions, we prove that after adding one special neuron with a skip connection to the output, or one special neuron per layer, every local minimum is a global minimum.

Drones, robots and self-driving cars could dodge accidents in future thanks to a new sensor inspired by the collision avoidance neurons of swarming locusts. Flying locusts travel in swarms of millions at 2–3 miles an hour -- but are capable of making evasive manoeuvres within just hundreds of milliseconds to avoid collisions. Researchers from the US have developed an electronic version of a special neuron, unique to locusts, that allows them to react rapidly without using much energy. Unlike current collision sensors for self-driving cars -- which tend to be bulky and heavy -- the insect-inspired detector is tiny and can respond in just two seconds. 'We are always looking for animals with unusual abilities, ones that do something better than humans,' said paper author and engineer Saptarshi Das of the Pennsylvania State University.

Specific neurons in fruit flies fire according to wind direction, helping them form a neural map of their surroundings. Algorithms inspired by this may be able to help robots to better navigate their environment. Tatsuo Okubo at Harvard Medical School and his colleagues wanted to determine how wind direction was characterised by a fruit fly's brain. While it is well known that wind direction affects the behaviour of insects, no one had yet developed a map of the neurons involved in this phenomenon for any animal.

One of the main difficulties in analyzing neural networks is the non-convexity of the loss function which may have many bad local minima. In this paper, we study the landscape of neural networks for binary classification tasks. Under mild assumptions, we prove that after adding one special neuron with a skip connection to the output, or one special neuron per layer, every local minimum is a global minimum. Papers published at the Neural Information Processing Systems Conference.

Drones and self-driving cars may soon come with'spidey' senses. That's according to engineers in America, who believe the unmanned machines would benefit from sensory detectors similar to those often seen in arachinds. Specifically, they're referring the hairs on a spider's legs, which are linked to special neurons called mechanoreceptors, which flag-up danger through vibrations. If machines had similar characteristics, they'd be able to navigate more effectively in dangerous environments. Until now, sensor technology hasn't always been able to process data fast enough, or as smoothly, as nature.