A low-power microwave neural network fabricated using complementary metal–oxide–semiconductor technology can perform broadband computations using a slow control mechanism.
Category: robotics/AI
Rethinking brain-like artificial intelligence: New study reveals hidden mismatches
A new study by York University researchers has found a potential striking flaw in artificial intelligence (AI) models. Artificial neural networks (ANNs), a type of AI model built to solve vision tasks for computers, have surprisingly emerged as the current best understanding of how our own brain’s visual system works, in the last decade. But does current AI really work like a primate brain?
“Artificial intelligence systems are often described as ‘brain-like’ because they can predict activity in parts of the brain that help us recognize objects,” says York University Assistant Professor Kohitij Kar, senior author of a new study. “Until now, scientists mostly tested this in one direction. They asked whether AI models can predict brain activity.”
In this study, the researchers flipped the question—if AI truly mirrors the brain, shouldn’t brain activity also be able to predict what’s happening inside the AI model?—and developed a reverse predictivity test to find the answer. The findings are published in the journal Nature Machine Intelligence.
The influencers with millions of followers who don’t actually exist
Lil Miquela has 2.5 million Instagram followers, a high-fashion wardrobe, and a clear political voice. She has advocated for Black Lives Matter and the LGBTQI+ community, fronted major brand campaigns, and built a devoted global fanbase. She also has no pulse.
Lil Miquela is a virtual influencer, a computer-generated character designed to look, sound, and behave like a real person. And she is not alone.
In China, Liu Yexi blends traditional aesthetics with cyberpunk visuals to amass a huge following. Ling, created by Chinese AI startup Xmov, has promoted Tesla, Vogue, and luxury tea brand Nayuki.
Chip-scale light technology could power faster AI and data center communications
Researchers at Trinity have developed a new light-based technology on a tiny chip that could help make the data centers behind cloud computing, artificial intelligence, and global internet services faster and more efficient. In the new research, recently published in Nature Communications, the Trinity team reported one such promising advance with collaborators at the University of Bath and the Swiss Federal Institute of Technology Lausanne (EPFL).
The team developed a new way to generate extremely stable signals of light using microscopic ring-shaped devices called “microresonators.” These signals form what scientists call optical frequency combs, sometimes described as “optical rulers” because they produce a series of evenly spaced colors of light that can be used to measure light with remarkable precision.
The researchers also demonstrated a new type of light pulse called a “hyperparametric soliton.” This stable pulse is the key behind the major advancement in this work, as it allows the comb signals to be produced at different colors of light from the laser that powers the device.
How “mindreading” AI detects hidden suicidal thoughts in the brains of young adults
A recent study published in Human Brain Mapping provides evidence that young adults experiencing suicidal thoughts process concepts related to death differently in their brains compared to healthy individuals. The findings indicate that these individuals reflexively associate death-related ideas with their own sense of self. This research suggests that brain imaging combined with artificial intelligence could eventually help identify people at risk for suicide based on how their brains represent specific words.
If you or someone you know is experiencing suicidal thoughts or a mental health crisis, help is available. Call or text 988 to reach the free and confidential Suicide & Crisis Lifeline, or chat live at 988lifeline.org.
While mental health professionals typically rely on patients to report their feelings, people at risk for suicide do not always disclose their struggles. Finding an objective physical measurement in the brain could help identify those in need of support.
Long-term inflammatory memory driver identified!
The researchers first gave a bout of psoriasis to mice when they were young. They discovered that about 10–15% of the memories that persisted a month later stuck around even to the end of the mouse’s life (~2 years). To see why these long-term memories lingered while their short-term counterparts faded within six months, they analyzed the DNA sequence characteristics within each of the memories by using a deep learning model customized by the third co-first author.
“When we compared the DNA sequences of short and long-term memory domains, they looked very similar in terms of the numbers and kinds of transcription factor binding sites,” says the author. “We realized we needed to develop a new metric that specifically captures memory persistence across time, not just total accessibility at any one point.”
Soto-Ugaldi’s adaptation, called PersistNet, quickly identified a telling trait: The longest lasting memory domains had an unusually high frequency of CpG dinucleotides—short DNA sequences of cytosine followed by guanine, which are known to play a key role in gene regulation. In fact, the model predicted that CpG density hardwires a timer into every memory domain: The more CpG’s, the longer the memory.
When they tested the prediction, that’s exactly what they found. “Looking across all 1,000 memory domains, we discovered that these nucleotide densities alone, and no other DNA sequence pattern, could distinguish how long each memory would linger,” says the author.
Back in the lab, the team discovered that these genetically wired densities enabled a host of epigenetic changes in memory domains, including DNA demethylation (the removal of a methyl group specifically found on CpG dinucleotides); the binding of transcription factors that prefer demethylated states; and the recruitment of a histone variant called H2A.Z, which preferentially seeks out demethylated sites and boosts chromatin accessibility while staving off future re-methylation. Together, these changes stabilized the open chromatin formation and its gene-priming activity. As the authors discovered, this structure could crucially be passed down across cellular generations, essentially keeping the doors open for life. Science Mission sciencenewshighlights.
One of the most puzzling aspects of common chronic inflammatory skin diseases such as psoriasis is how they become chronic. What allows an ongoing condition to stay dormant for months or even years, then seemingly spring back out of nowhere?