ReportWire

Tag: Quanta Magazine

  • The Brain Region That Controls Movement Also Guides Feelings

    The Brain Region That Controls Movement Also Guides Feelings

    The original version of this story appeared in Quanta Magazine.

    In recent decades, neuroscience has seen some stunning advances, and yet a critical part of the brain remains a mystery. I am referring to the cerebellum, so named for the Latin for “little brain,” which is situated like a bun at the back of the brain. This is no small oversight: The cerebellum contains three-quarters of all the brain’s neurons, which are organized in an almost crystalline arrangement, in contrast to the tangled thicket of neurons found elsewhere.

    Encyclopedia articles and textbooks underscore the fact that the cerebellum’s function is to control body movement. There is no question that the cerebellum has this function. But scientists now suspect that this long-standing view is myopic.

    Or so I learned in November in Washington, DC, while attending the Society for Neuroscience annual meeting, the largest meeting of neuroscientists in the world. There, a pair of neuroscientists organized a symposium on newly discovered functions of the cerebellum unrelated to motor control. New experimental techniques are showing that in addition to controlling movement, the cerebellum regulates complex behaviors, social interactions, aggression, working memory, learning, emotion, and more.

    A Crack in Dominant Wisdom

    The connection between the cerebellum and movement has been known since the 19th century. Patients suffering trauma to the brain region had obvious difficulties with balance and movement, leaving no doubt that it was critical for coordinating motion. Over the decades, neuroscientists developed a detailed understanding of how the cerebellum’s unique neural circuitry controls motor function. The explanation of how the cerebellum worked seemed watertight.

    Then, in 1998, in the journal Brain, neurologists reported on wide-ranging emotional and cognitive disabilities in patients with damage to the cerebellum. For example, in 1991, a 22-year-old female college student had fallen while ice skating; a CT scan revealed a tumor in her cerebellum. After it was removed surgically, she was a completely different person. The bright college student had lost her ability to write with proficiency, do mental arithmetic, name common objects, or copy a simple diagram. Her mood flattened. She hid under covers and behaved inappropriately, undressing in the corridors and speaking in baby talk. Her social interactions, including recognizing familiar faces, were also impaired.

    This and similar cases puzzled the authors. These high-level cognitive and emotional functions were understood to reside in the cerebral cortex and limbic system. “Precisely what that cerebellar role is, and how the cerebellum accomplishes it, is yet to be established,” they concluded.

    Despite these clues from clinical studies that conventional wisdom was on the wrong track, leading authorities still insisted that the function of the cerebellum was to control movement and nothing more. “It is kind of sad, because it has been 20 years” since these cases were reported, said Diasynou Fioravante, a neurophysiologist at the UC Davis, who co-organized the conference symposium.

    Other neurologists have noticed neuropsychiatric deficits in their patients all along, said the neuroscientist Stephanie Rudolph of Albert Einstein College of Medicine, who co-organized the symposium with Fioravante. However, there was no hard anatomical evidence for how the cerebellum’s unique neural circuitry could possibly regulate the reported psychological and emotional functions, so the clinical reports were overlooked.

    R Douglas Fields

    Source link

  • Large Language Models’ Emergent Abilities Are a Mirage

    Large Language Models’ Emergent Abilities Are a Mirage

    The original version of this story appeared in Quanta Magazine.

    Two years ago, in a project called the Beyond the Imitation Game benchmark, or BIG-bench, 450 researchers compiled a list of 204 tasks designed to test the capabilities of large language models, which power chatbots like ChatGPT. On most tasks, performance improved predictably and smoothly as the models scaled up—the larger the model, the better it got. But with other tasks, the jump in ability wasn’t smooth. The performance remained near zero for a while, then performance jumped. Other studies found similar leaps in ability.

    The authors described this as “breakthrough” behavior; other researchers have likened it to a phase transition in physics, like when liquid water freezes into ice. In a paper published in August 2022, researchers noted that these behaviors are not only surprising but unpredictable, and that they should inform the evolving conversations around AI safety, potential, and risk. They called the abilities “emergent,” a word that describes collective behaviors that only appear once a system reaches a high level of complexity.

    But things may not be so simple. A new paper by a trio of researchers at Stanford University posits that the sudden appearance of these abilities is just a consequence of the way researchers measure the LLM’s performance. The abilities, they argue, are neither unpredictable nor sudden. “The transition is much more predictable than people give it credit for,” said Sanmi Koyejo, a computer scientist at Stanford and the paper’s senior author. “Strong claims of emergence have as much to do with the way we choose to measure as they do with what the models are doing.”

    We’re only now seeing and studying this behavior because of how large these models have become. Large language models train by analyzing enormous data sets of text—words from online sources including books, web searches, and Wikipedia—and finding links between words that often appear together. The size is measured in terms of parameters, roughly analogous to all the ways that words can be connected. The more parameters, the more connections an LLM can find. GPT-2 had 1.5 billion parameters, while GPT-3.5, the LLM that powers ChatGPT, uses 350 billion. GPT-4, which debuted in March 2023 and now underlies Microsoft Copilot, reportedly uses 1.75 trillion.

    That rapid growth has brought an astonishing surge in performance and efficacy, and no one is disputing that large enough LLMs can complete tasks that smaller models can’t, including ones for which they weren’t trained. The trio at Stanford who cast emergence as a “mirage” recognize that LLMs become more effective as they scale up; in fact, the added complexity of larger models should make it possible to get better at more difficult and diverse problems. But they argue that whether this improvement looks smooth and predictable or jagged and sharp results from the choice of metric—or even a paucity of test examples—rather than the model’s inner workings.

    Stephen Ornes

    Source link

  • Never-Repeating Patterns of Tiles Can Safeguard Quantum Information

    Never-Repeating Patterns of Tiles Can Safeguard Quantum Information

    This extreme fragility might make quantum computing sound hopeless. But in 1995, the applied mathematician Peter Shor discovered a clever way to store quantum information. His encoding had two key properties. First, it could tolerate errors that only affected individual qubits. Second, it came with a procedure for correcting errors as they occurred, preventing them from piling up and derailing a computation. Shor’s discovery was the first example of a quantum error-correcting code, and its two key properties are the defining features of all such codes.

    The first property stems from a simple principle: Secret information is less vulnerable when it’s divided up. Spy networks employ a similar strategy. Each spy knows very little about the network as a whole, so the organization remains safe even if any individual is captured. But quantum error-correcting codes take this logic to the extreme. In a quantum spy network, no single spy would know anything at all, yet together they’d know a lot.

    Each quantum error-correcting code is a specific recipe for distributing quantum information across many qubits in a collective superposition state. This procedure effectively transforms a cluster of physical qubits into a single virtual qubit. Repeat the process many times with a large array of qubits, and you’ll get many virtual qubits that you can use to perform computations.

    The physical qubits that make up each virtual qubit are like those oblivious quantum spies. Measure any one of them and you’ll learn nothing about the state of the virtual qubit it’s a part of—a property called local indistinguishability. Since each physical qubit encodes no information, errors in single qubits won’t ruin a computation. The information that matters is somehow everywhere, yet nowhere in particular.

    “You can’t pin it down to any individual qubit,” Cubitt said.

    All quantum error-correcting codes can absorb at least one error without any effect on the encoded information, but they will all eventually succumb as errors accumulate. That’s where the second property of quantum error-correcting codes kicks in—the actual error correction. This is closely related to local indistinguishability: Because errors in individual qubits don’t destroy any information, it’s always possible to reverse any error using established procedures specific to each code.

    Taken for a Ride

    Zhi Li, a postdoc at the Perimeter Institute for Theoretical Physics in Waterloo, Canada, was well versed in the theory of quantum error correction. But the subject was far from his mind when he struck up a conversation with his colleague Latham Boyle. It was the fall of 2022, and the two physicists were on an evening shuttle from Waterloo to Toronto. Boyle, an expert in aperiodic tilings who lived in Toronto at the time and is now at the University of Edinburgh, was a familiar face on those shuttle rides, which often got stuck in heavy traffic.

    “Normally they could be very miserable,” Boyle said. “This was like the greatest one of all time.”

    Before that fateful evening, Li and Boyle knew of each other’s work, but their research areas didn’t directly overlap, and they’d never had a one-on-one conversation. But like countless researchers in unrelated fields, Li was curious about aperiodic tilings. “It’s very hard to be not interested,” he said.

    Ben Brubaker

    Source link

  • Google’s Chess Experiments Reveal How to Boost the Power of AI

    Google’s Chess Experiments Reveal How to Boost the Power of AI

    His group decided to find out. They built the new, diversified version of AlphaZero, which includes multiple AI systems that trained independently and on a variety of situations. The algorithm that governs the overall system acts as a kind of virtual matchmaker, Zahavy said: one designed to identify which agent has the best chance of succeeding when it’s time to make a move. He and his colleagues also coded in a “diversity bonus”—a reward for the system whenever it pulled strategies from a large selection of choices.

    When the new system was set loose to play its own games, the team observed a lot of variety. The diversified AI player experimented with new, effective openings and novel—but sound—decisions about specific strategies, such as when and where to castle. In most matches, it defeated the original AlphaZero. The team also found that the diversified version could solve twice as many challenge puzzles as the original and could solve more than half of the total catalog of Penrose puzzles.

    “The idea is that instead of finding one solution, or one single policy, that would beat any player, here [it uses] the idea of creative diversity,” Cully said.

    With access to more and different played games, Zahavy said, the diversified AlphaZero had more options for sticky situations when they arose. “If you can control the kind of games that it sees, you basically control how it will generalize,” he said. Those weird intrinsic rewards (and their associated moves) could become strengths for diverse behaviors. Then the system could learn to assess and value the disparate approaches and see when they were most successful. “We found that this group of agents can actually come to an agreement on these positions.”

    And, crucially, the implications extend beyond chess.

    Real-Life Creativity

    Cully said a diversified approach can help any AI system, not just those based on reinforcement learning. He has long used diversity to train physical systems, including a six-legged robot that was allowed to explore various kinds of movement, before he intentionally “injured” it, allowing it to continue moving using some of the techniques it had developed before. “We were just trying to find solutions that were different from all previous solutions we have found so far.” Recently, he has also been collaborating with researchers to use diversity to identify promising new drug candidates and develop effective stock-trading strategies.

    “The goal is to generate a large collection of potentially thousands of different solutions, where every solution is very different from the next,” Cully said. So—just as the diversified chess player learned to do—for every type of problem, the overall system could choose the best possible solution. Zahavy’s AI system, he said, clearly shows how “searching for diverse strategies helps to think outside the box and find solutions.”

    Zahavy suspects that in order for AI systems to think creatively, researchers simply have to get them to consider more options. That hypothesis suggests a curious connection between humans and machines: Maybe intelligence is just a matter of computational power. For an AI system, maybe creativity boils down to the ability to consider and select from a large enough buffet of options. As the system gains rewards for selecting a variety of optimal strategies, this kind of creative problem-solving gets reinforced and strengthened. Ultimately, in theory, it could emulate any kind of problem-solving strategy recognized as a creative one in humans. Creativity would become a computational problem.

    Liemhetcharat noted that a diversified AI system is unlikely to completely resolve the broader generalization problem in machine learning. But it’s a step in the right direction. “It’s mitigating one of the shortcomings,” she said.

    More practically, Zahavy’s results resonate with recent efforts that show how cooperation can lead to better performance on hard tasks among humans. Most of the hits on the Billboard 100 list were written by teams of songwriters, for example, not individuals. And there’s still room for improvement. The diverse approach is currently computationally expensive, since it must consider so many more possibilities than a typical system. Zahavy is also not convinced that even the diversified AlphaZero captures the entire spectrum of possibilities.

    “I still [think] there is room to find different solutions,” he said. “It’s not clear to me that given all the data in the world, there is [only] one answer to every question.”

    Original story reprinted with permission from Quanta Magazine, an editorially independent publication of the Simons Foundation whose mission is to enhance public understanding of science by covering research developments and trends in mathematics and the physical and life sciences.

    Stephen Ornes

    Source link

  • A Celebrated Cryptography-Breaking Algorithm Just Got an Upgrade

    A Celebrated Cryptography-Breaking Algorithm Just Got an Upgrade


    This is a job for LLL: Give it (or its brethren) a basis of a multidimensional lattice, and it’ll spit out a better one. This process is known as lattice basis reduction.

    What does this all have to do with cryptography? It turns out that the task of breaking a cryptographic system can, in some cases, be recast as another problem: finding a relatively short vector in a lattice. And sometimes, that vector can be plucked from the reduced basis generated by an LLL-style algorithm. This strategy has helped researchers topple systems that, on the surface, appear to have little to do with lattices.

    In a theoretical sense, the original LLL algorithm runs quickly: The time it takes to run doesn’t scale exponentially with the size of the input—that is, the dimension of the lattice and the size (in bits) of the numbers in the basis vectors. But it does increase as a polynomial function, and “if you actually want to do it, polynomial time is not always so feasible,” said Léo Ducas, a cryptographer at the national research institute CWI in the Netherlands.

    In practice, this means that the original LLL algorithm can’t handle inputs that are too large. “Mathematicians and cryptographers wanted the ability to do more,” said Keegan Ryan, a doctoral student at the University of California, San Diego. Researchers worked to optimize LLL-style algorithms to accommodate bigger inputs, often achieving good performance. Still, some tasks have remained stubbornly out of reach.

    The new paper, authored by Ryan and his adviser, Nadia Heninger, combines multiple strategies to improve the efficiency of its LLL-style algorithm. For one thing, the technique uses a recursive structure that breaks the task down into smaller chunks. For another, the algorithm carefully manages the precision of the numbers involved, finding a balance between speed and a correct result. The new work makes it feasible for researchers to reduce the bases of lattices with thousands of dimensions.

    Past work has followed a similar approach: A 2021 paper also combines recursion and precision management to make quick work of large lattices, but it worked only for specific kinds of lattices, and not all the ones that are important in cryptography. The new algorithm behaves well on a much broader range. “I’m really happy someone did it,” said Thomas Espitau, a cryptography researcher at the company PQShield and an author of the 2021 version. His team’s work offered a “proof of concept,” he said; the new result shows that “you can do very fast lattice reduction in a sound way.”

    The new technique has already started to prove useful. Aurel Page, a mathematician with the French national research institute Inria, said that he and his team have put an adaptation of the algorithm to work on some computational number theory tasks.

    LLL-style algorithms can also play a role in research related to lattice-based cryptography systems designed to remain secure even in a future with powerful quantum computers. They don’t pose a threat to such systems, since taking them down requires finding shorter vectors than these algorithms can achieve. But the best attacks researchers know of use an LLL-style algorithm as a “basic building block,” said Wessel van Woerden, a cryptographer at the University of Bordeaux. In practical experiments to study these attacks, that building block can slow everything down. Using the new tool, researchers may be able to expand the range of experiments they can run on the attack algorithms, offering a clearer picture of how they perform.

    Original story reprinted with permission from Quanta Magazine, an editorially independent publication of the Simons Foundation whose mission is to enhance public understanding of science by covering research developments and trends in mathematics and the physical and life sciences.



    Madison Goldberg

    Source link

  • Quanta Magazine Partners With PRX and Cosmologist Janna Levin for Season 3 of ‘The Joy of Why’ Podcast

    Quanta Magazine Partners With PRX and Cosmologist Janna Levin for Season 3 of ‘The Joy of Why’ Podcast


    NEW YORK, January 25, 2024 (Newswire.com)
    –
    For two seasons of Quanta Magazine’s “The Joy of Why” podcast, mathematician and author Steven Strogatz has invited listeners to learn about the questions that drive the work of leading researchers.

    Beginning February 1, listeners can look forward to season three of the podcast, with 24 new episodes and a new co-host, the cosmologist Janna Levin. A new episode will go out every other Thursday on the Quanta website, Apple Podcasts, Spotify and other podcast platforms.

    This season’s guests include Fields Medalist and mathematician Terence Tao, Nobel Prize–winning physicist Frank Wilczek and psychedelics neuroscientist Gül Dölen. Tao’s episode, about the makings of good mathematics, will kick off the new season.

    To help bring “The Joy of Why” to listeners everywhere on-demand, Quanta has partnered with public media organization PRX — one of the world’s top podcast publishers and public radio distributors.

    “I have listened to previous seasons with delight,” Levin writes in a new column for Quanta. “An irrepressible curiosity ignites the science lover, and in this capacity, as hosts of a math and science podcast, we are proxies for you, and our curiosity is a proxy for yours.”

    In season three, Strogatz and Levin take turns interviewing researchers about questions surrounding the neurobiology of depression, the nature of time and the flocking behaviors of animals. Levin, an author and the Claire Tow Professor of Physics and Astronomy at Barnard College of Columbia University, is an experienced interviewer. She is the director of sciences and co-founder of Pioneer Works in Brooklyn, where she hosts rich public discussions with scientists and mathematicians that weave together curiosity and culture.

    “We hope you’ll marvel with us this season on ‘The Joy of Why,’” Levin writes in her new column.

    Quanta Magazine is a Pulitzer Prize–winning, editorially independent online publication of the Simons Foundation. In 2023, Strogatz received a National Academies Eric and Wendy Schmidt Award for Excellence in Science Communications partly for his work on “The Joy of Why.”

    Celebrating more than 20 years as a nonprofit public media company, PRX works in partnership with leading independent creators, organizations and stations to bring meaningful audio storytelling into millions of listeners’ lives. PRX’s portfolio of broadcast productions and podcast partners have received recognition from the Peabody Awards, the Tribeca Festival, the International Documentary Association and the Pulitzer Prizes.

    Source: Quanta Magazine

    Source link