Recent studies have demonstrated the susceptibility of deep neural networks to backdoor attacks. Given a backdoored model, its prediction of a poisoned sample with trigger will be dominated by the trigger information, though trigger information and benign information coexist. Inspired by the mechanism of the optical polarizer that a polarizer could pass light waves with particular polarizations while filtering light waves with other polarizations, we propose a novel backdoor defense method by inserting a learnable neural polarizer into the backdoored model as an intermediate layer, in order to purify the poisoned sample via filtering trigger information while maintaining benign information. The neural polarizer is instantiated as one lightweight linear transformation layer, which is learned through solving a well designed bi-level optimization problem, based on a limited clean dataset. Compared to other fine-tuning-based defense methods which often adjust all parameters of the backdoored model, the proposed method only needs to learn one additional layer, such that it is more efficient and requires less clean data. Extensive experiments demonstrate the effectiveness and efficiency of our method in removing backdoors across various neural network architectures and datasets, especially in the case of very limited clean data.

Breakthroughs, discoveries, and DIY tips sent every weekday. The lifeblood of global communication flows through more than 807,800 miles worth of garden hose-wide cables woven across the sea floor. These cables, which reportedly transmit over 10 trillion worth of financial data every day, are vulnerable to extreme weather, decay, and, if recent reports are to be believed, acts of sabotage. The Associated Press estimates that at least 11 cables have been damaged since October 2023 in the Baltic Sea alone. Finnish and German authorities traced several of those incidents back to dragged anchors, which they allege may have been intentionally deployed to cause damage for political ends.

Recent studies have demonstrated the susceptibility of deep neural networks to backdoor attacks. Given a backdoored model, its prediction of a poisoned sample with trigger will be dominated by the trigger information, though trigger information and benign information coexist. Inspired by the mechanism of the optical polarizer that a polarizer could pass light waves with particular polarizations while filtering light waves with other polarizations, we propose a novel backdoor defense method by inserting a learnable neural polarizer into the backdoored model as an intermediate layer, in order to purify the poisoned sample via filtering trigger information while maintaining benign information. The neural polarizer is instantiated as one lightweight linear transformation layer, which is learned through solving a well designed bi-level optimization problem, based on a limited clean dataset. Compared to other fine-tuning-based defense methods which often adjust all parameters of the backdoored model, the proposed method only needs to learn one additional layer, such that it is more efficient and requires less clean data.

'Meeting a real-life cyborg was gobsmacking' For the past 20 years, self-declared cyborg artist Neil Harbisson has provoked debate with his eyeborg - a surgically attached antenna. Harbisson, who grew up in Barcelona, is colour blind, having been born with the rare condition achromatopsia, which affects one in 33,000 people. This means he sees in what he calls greyscale - only black, white and shades of grey. But he decided to have surgery in 2004 which changed his life - and his senses - attaching an antenna to the back of his head, which transforms light waves into sounds. When film director Carey Born came across Harbisson, classed by Guinness World Records as the first officially recognised'cyborg', she was gobsmacked and astonished.

Researchers have developed a new technique for performing computations straight on smart home gadgets. Their method moves the memory-intensive machine-learning model operations to a central server. The data is encoded onto light waves rather than sent hundreds of miles away from the device. Fiber optics is used to transfer the waves to a connected device, allowing massive amounts of data to be sent via a network at incredible speeds. The receiver then uses a specific optical device to quickly compute the model's components delivered by those light waves.

Ask a smart home device for the weather forecast, and it takes several seconds for the device to respond. One reason this latency occurs is because connected devices don't have enough memory or power to store and run the enormous machine-learning models needed for the device to understand what a user is asking of it. The model is stored in a data center that may be hundreds of miles away, where the answer is computed and sent to the device. MIT researchers have created a new method for computing directly on these devices, which drastically reduces this latency. Their technique shifts the memory-intensive steps of running a machine-learning model to a central server where components of the model are encoded onto light waves.

Ask a smart home device for the weather forecast, and it takes several seconds for the device to respond. One reason this latency occurs is because connected devices don't have enough memory or power to store and run the enormous machine-learning models needed for the device to understand what a user is asking of it. The model is stored in a data center that may be hundreds of miles away, where the answer is computed and sent to the device. MIT researchers have created a new method for computing directly on these devices, which drastically reduces this latency. Their technique shifts the memory-intensive steps of running a machine-learning model to a central server where components of the model are encoded onto light waves.

In order to scrutinize the internal features of a living organism using light, it is necessary to A) deliver sufficient light energy to the sample and B) accurately measure the signal reflected from the target tissue. However, in living tissues multiple scattering effects and severe aberration1 tend to occur when light hits the cells, which makes it difficult to obtain sharp images. In complex structures such as living tissue, light undergoes multiple scattering, which causes the photons to randomly change their direction several times as they travel through the tissue. Because of this process, much of the image information carried by the light becomes ruined. However, even if it is a very small amount of reflected light, it is possible to observe the features located relatively deep within the tissues by correcting the wavefront2 distortion of the light that was reflected from the target to be observed. However, the above-mentioned multiple scattering effects interfere with this correction process.

One of the imaging technologies that many robotics companies are integrating into their sensor packages is Light Detection and Ranging, or LiDAR for short. Currently commanding great attention and investment from self-driving car developers, the approach essentially works like radar, but instead of sending out broad radio waves and looking for reflections, it uses short pulses of light from lasers. Traditional time-of-flight LiDAR, however, has many drawbacks that make it difficult to use in many 3D vision applications. Because it requires detection of very weak reflected light signals, other LiDAR systems or even ambient sunlight can easily overwhelm the detector. It also has limited depth resolution and can take a dangerously long time to densely scan a large area such as a highway or factory floor.

A powerful holographic camera has been developed by scientists, and it is able to see though almost anything, including corners, fog, and even human flesh. The device, built by researchers from Northwestern University in Evanston, Illinois, uses a technique called'synthetic wavelength holography'. It works by indirectly scattering light onto hidden objects, which then scatters again and travels back to a camera, where AI is used to reconstruct the original object. The team say it is a decade away from being commercially available, but when it is the technology could be used in cars, CCTV and even as a medical scanner. One example could be to replace the use of an endoscope in colonoscopy, instead gathering the light waves to see around the folds inside the intestines.