A mathematical proof shows that some quantum states can resist nature’s tendency to disorder – but only under very specific conditions.
Overcoming ‘catastrophic forgetting’: Algorithm inspired by brain allows neural networks to retain knowledge
Posted in biological, information science, robotics/AI, transportation | Leave a Comment on Overcoming ‘catastrophic forgetting’: Algorithm inspired by brain allows neural networks to retain knowledge
Neural networks have a remarkable ability to learn specific tasks, such as identifying handwritten digits. However, these models often experience “catastrophic forgetting” when taught additional tasks: They can successfully learn the new assignments, but “forget” how to complete the original. For many artificial neural networks, like those that guide self-driving cars, learning additional tasks thus requires being fully reprogrammed.
Biological brains, on the other hand, are remarkably flexible. Humans and animals can easily learn how to play a new game, for instance, without having to re-learn how to walk and talk.
Inspired by the flexibility of human and animal brains, Caltech researchers have now developed a new type of algorithm that enables neural networks to be continuously updated with new data that they are able to learn from without having to start from scratch. The algorithm, called a functionally invariant path (FIP) algorithm, has wide-ranging applications from improving recommendations on online stores to fine-tuning self-driving cars.
Be it water, light or sound: waves usually propagate in the same way forwards as in the backward direction. As a consequence, when we are speaking to someone standing some distance away from us, that person can hear us as well as we can hear them. This is useful when having a conversation, but in some technical applications one would prefer the waves to be able to travel only in one direction – for instance, in order to avoid unwanted reflections of light or microwaves.
For sound waves, ten years ago researchers succeeded in suppressing their propagation in the backward direction; however, this also attenuated the waves travelling forwards. A team of researchers at ETH Zurich led by Nicolas Noiray, professor for Combustion, Acoustics and Flow Physics, in collaboration with Romain Fleury at EPFL, has now developed a method for preventing sound waves from travelling backwards without deteriorating their propagation in the forward direction. In the future, this method, which has recently been published in the scientific journal external page Nature Communications, could also be applied to electromagnetic waves.
The basis of this one-way street for sound waves are self-oscillations, in which a dynamical system periodically repeats its behaviour. “I’ve actually spent a good part of my career preventing such phenomena”, says Noiray. Amongst other things, he studies how self-sustaining thermo-acoustic oscillations can arise from the interplay between sound waves and flames in the combustion chamber of an aircraft engine, which can lead to dangerous vibrations. In the worst case, these vibrations can destroy the engine.
It’s clear that 2024 is the year of AI breaking into the world’s most prestigious scientific awards.
A nanotube researcher in Japan has earned 13 retractions, with more to come, after an extensive investigation by the country’s National Institute of Advanced Industrial Science and Technology (AIST) revealed widespread misconduct in his work.
AIST’s investigation found Naohiro Kameta, senior principal researcher at the Nanomaterials Research Institute located in AIST’s Ibaraki campus, fabricated and falsified dozens of studies. He was apparently dismissed from his role following the findings.
The institute first learned of the problems in Kameta’s work in November 2022, according to a translated version of the investigation report. Initially, they looked into five papers, but eventually expanded their scrutiny to 61 articles on which Kameta was the lead or responsible author.
Implementation of new pricing and business structures and improved licensing and manufacturing processes could reduce the per-patient cost of gene therapy tenfold.
SpaceX announced a new capability for the Dragon spacecraft on Sept. 27 in the unlikely event of a parachute failure. Dragon now has built-in redundancy to propulsively land using its SuperDraco thrusters, saving the vehicle and potential crew from a rough landing or imminent danger.
SpaceX introduced the concept of a propulsive landing Dragon over ten years ago. When SpaceX revealed Dragon 2, it was marketed as capable of propulsively landing anywhere on Earth and was initially designed to land exclusively with the SuperDracos. However, SpaceX ultimately pursued the use of parachutes as the main form of recovery for Dragon 2 missions.
Much has had to change with Dragon 2 since May 30, 2014, to make it the reliable crewed spacecraft we know it as today. Now, SpaceX has decided to bring back one of the main capabilities that was believed to have been left behind in development.
