Close Menu
    Facebook LinkedIn YouTube WhatsApp X (Twitter) Pinterest
    Trending
    • Supermassive black holes may create millions of new planets
    • Cheque in: 3 startups ended May by raising $15.5 million
    • Universal Audio Volt 876 USB Audio Interface Review: Pro-Level Polish
    • New York City-based Mecka AI, which trains robots with human data sourced from body sensors and iPhones, raised $60M, including a $25M Series A (Ben Weiss/Fortune)
    • Is Instagram Down? What to Know
    • It’s the Lessons We Learned Along the Way. Or, Is It?
    • The forever chemicals impacting your health
    • WiseTech CEO threatened amid job cuts; founder Richard White calls in police
    Facebook LinkedIn WhatsApp
    Times FeaturedTimes Featured
    Monday, June 1
    • Home
    • Founders
    • Startups
    • Technology
    • Profiles
    • Entrepreneurs
    • Leaders
    • Students
    • VC Funds
    • More
      • AI
      • Robotics
      • Industries
      • Global
    Times FeaturedTimes Featured
    Home»News»Researchers isolate memorization from problem-solving in AI neural networks
    News

    Researchers isolate memorization from problem-solving in AI neural networks

    Editor Times FeaturedBy Editor Times FeaturedNovember 17, 2025No Comments3 Mins Read
    Facebook Twitter Pinterest Telegram LinkedIn Tumblr WhatsApp Email
    Share
    Facebook Twitter LinkedIn Pinterest Telegram Email WhatsApp Copy Link


    Mathematical operations and closed-book truth retrieval shared pathways with memorization, dropping to 66 to 86 p.c efficiency after enhancing. The researchers discovered arithmetic significantly brittle. Even when fashions generated similar reasoning chains, they failed on the calculation step after low-curvature parts had been eliminated.

    Determine 3 from the paper “From Memorization to Reasoning within the Spectrum of Loss Curvature.”


    Credit score:

    Merullo et al.


    “Arithmetic issues themselves are memorized on the 7B scale, or as a result of they require narrowly used instructions to do exact calculations,” the workforce explains. Open-book query answering, which depends on offered context quite than inside data, proved most strong to the enhancing process, sustaining practically full efficiency.

    Curiously, the mechanism separation diverse by info kind. Frequent details like nation capitals barely modified after enhancing, whereas uncommon details like firm CEOs dropped 78 p.c. This means fashions allocate distinct neural assets based mostly on how steadily info seems in coaching.

    The Ok-FAC method outperformed present memorization removing strategies with no need coaching examples of memorized content material. On unseen historic quotes, Ok-FAC achieved 16.1 p.c memorization versus 60 p.c for the earlier finest methodology, BalancedSubnet.

    Imaginative and prescient transformers confirmed comparable patterns. When skilled with deliberately mislabeled photographs, the fashions developed distinct pathways for memorizing fallacious labels versus studying appropriate patterns. Eradicating memorization pathways restored 66.5 p.c accuracy on beforehand mislabeled photographs.

    Limits of reminiscence removing

    Nevertheless, the researchers acknowledged that their method isn’t good. As soon as-removed reminiscences would possibly return if the mannequin receives extra coaching, as other research has proven that present unlearning strategies solely suppress info quite than utterly erasing it from the neural community’s weights. Meaning the “forgotten” content material will be reactivated with just some coaching steps focusing on these suppressed areas.

    The researchers can also’t absolutely clarify why some skills, like math, break so simply when memorization is eliminated. It’s unclear whether or not the mannequin truly memorized all its arithmetic or whether or not math simply occurs to make use of comparable neural circuits as memorization. Moreover, some subtle capabilities would possibly appear like memorization to their detection methodology, even once they’re truly advanced reasoning patterns. Lastly, the mathematical instruments they use to measure the mannequin’s “panorama” can grow to be unreliable on the extremes, although this doesn’t have an effect on the precise enhancing course of.

    This text was up to date on November 11, 2025 at 9:16 am to make clear a proof about sorting weights by curvature.



    Source link

    Share. Facebook Twitter Pinterest LinkedIn Tumblr Email
    Editor Times Featured
    • Website

    Related Posts

    New York City-based Mecka AI, which trains robots with human data sourced from body sensors and iPhones, raised $60M, including a $25M Series A (Ben Weiss/Fortune)

    June 1, 2026

    SpaceX will reserve up to 5% of its Class A shares for select employees and executives’ friends and family; 60%+ of shares have an extended lock-up (Charles Capel/Bloomberg)

    June 1, 2026

    Netherlands-based Invisix, which is developing advanced chipmaking measurement tools, raised a €20M seed, with the participation of a “tier-one” chipmaker (Tamara Djurickovic/Tech.eu)

    June 1, 2026

    Nvidia unveils DGX Station, a desktop Windows PC powered by its GB300 Grace Blackwell chip with up to 748 GB of memory, capable of running 1T-parameter models (Mike Wheatley/SiliconANGLE)

    June 1, 2026

    Intel teases its Xeon 7 Diamond Rapids CPUs, built on 18A-P node, with PCIe 6.0, and 50% more cores and twice the memory bandwidth vs. Xeon 6, launching in 2027 (Jake Roach/Tom’s Hardware)

    June 1, 2026

    Dell introduces the $699+ Dell XPS 13, starting with 8GB of RAM, a six-core Intel Core 5 320 chip, and a 13.4-inch touchscreen, rivaling the MacBook Neo (Antonio G. Di Benedetto/The Verge)

    June 1, 2026

    Comments are closed.

    Editors Picks

    Supermassive black holes may create millions of new planets

    June 1, 2026

    Cheque in: 3 startups ended May by raising $15.5 million

    June 1, 2026

    Universal Audio Volt 876 USB Audio Interface Review: Pro-Level Polish

    June 1, 2026

    New York City-based Mecka AI, which trains robots with human data sourced from body sensors and iPhones, raised $60M, including a $25M Series A (Ben Weiss/Fortune)

    June 1, 2026
    Categories
    • Founders
    • Startups
    • Technology
    • Profiles
    • Entrepreneurs
    • Leaders
    • Students
    • VC Funds
    About Us
    About Us

    Welcome to Times Featured, an AI-driven entrepreneurship growth engine that is transforming the future of work, bridging the digital divide and encouraging younger community inclusion in the 4th Industrial Revolution, and nurturing new market leaders.

    Empowering the growth of profiles, leaders, entrepreneurs businesses, and startups on international landscape.

    Asia-Middle East-Europe-North America-Australia-Africa

    Facebook LinkedIn WhatsApp
    Featured Picks

    AI in the Workplace Statistics 2025–2035

    February 15, 2026

    Not on my watch: Christine Holgate on leadership, support, resilience, women aiming for the top – and how to survive

    October 6, 2025

    How pigeons use liver cells for magnetic sensing

    May 28, 2026
    Categories
    • Founders
    • Startups
    • Technology
    • Profiles
    • Entrepreneurs
    • Leaders
    • Students
    • VC Funds
    Copyright © 2024 Timesfeatured.com IP Limited. All Rights.
    • Privacy Policy
    • Disclaimer
    • Terms and Conditions
    • About us
    • Contact us

    Type above and press Enter to search. Press Esc to cancel.