[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-dual-noise-trick-fixes-a-blind-spot-in-ai-robustness-proofs":10,"sections":35},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":30,"feedback":34,"feedback_at":22,"cost_usd":34,"total_tokens":34},6455,"dual-noise-trick-fixes-a-blind-spot-in-ai-robustness-proofs","Dual Noise Trick Fixes a Blind Spot in AI Robustness Proofs","A new randomized smoothing method uses input-specific noise instead of a single global variance, improving certified robustness on CIFAR-10 and ImageNet.","A new twist on a popular AI defense technique swaps one fixed noise setting for noise tuned to each individual input.\n\nResearchers behind the method, called dual randomized smoothing, tackle a known weakness in randomized smoothing (RS), a technique used to mathematically certify that a neural network's predictions won't flip under small adversarial perturbations. Standard RS relies on a single global noise variance: turn it up and the model handles bigger attacks but loses accuracy on easy cases; turn it down and the reverse happens. The new approach trains a separate estimator to pick a locally-constant noise level for each input, then feeds that into a standard RS classifier, with both parts trained together. On CIFAR-10 it beat prior input-dependent noise methods by 15.6, 20.0, and 15.7 percentage points at radii of 0.5, 0.75, and 1.0; on ImageNet the gains were 8.6, 17.1, and 9.1 points at radii 0.5, 1.0, and 1.5.\n\nThat is a meaningful fix, not a marginal tweak. Certified robustness has always forced a tradeoff between defending against small versus large perturbations, and papers claiming state of the art on this benchmark are common but rarely move both ends of the curve at once. Dual RS does, and it does so for a roughly 60% inference overhead, which is steep but not exotic for certified-defense research.\n\nStill, this is a CIFAR-10 and ImageNet result from an academic paper, run through the usual certified-robustness benchmarks, not a production system fending off real attackers.","[\"adversarial robustness\",\"randomized smoothing\",\"ai security\",\"machine learning\"]","2026-09-16T04:00:00.000Z","2026-09-17T17:29:09.454Z","2026-09-17T17:29:21.369Z","published",null,[],"ai",[26,27,28,29],"adversarial robustness","randomized smoothing","ai security","machine learning",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2512.01782",0,{"sections":36},[37,41,46,51,56,60,64,69,74,78,83,88,93,98],{"name":38,"slug":24,"count":39,"latest_published_at":40},"AI",3853,"2026-09-17T08:27:09.000Z",{"name":42,"slug":43,"count":44,"latest_published_at":45},"Security","security",648,"2026-09-17T04:00:00.000Z",{"name":47,"slug":48,"count":49,"latest_published_at":50},"Policy","policy",338,"2026-09-11T04:00:00.000Z",{"name":52,"slug":53,"count":54,"latest_published_at":55},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":57,"slug":58,"count":59,"latest_published_at":45},"Hardware","hardware",154,{"name":61,"slug":62,"count":63,"latest_published_at":45},"Science","science",114,{"name":65,"slug":66,"count":67,"latest_published_at":68},"Consumer Tech","consumer-tech",99,"2026-09-09T17:27:33.000Z",{"name":70,"slug":71,"count":72,"latest_published_at":73},"Software","software",75,"2026-09-10T20:41:21.000Z",{"name":75,"slug":76,"count":77,"latest_published_at":45},"Dev Tools","dev-tools",73,{"name":79,"slug":80,"count":81,"latest_published_at":82},"Startups","startups",55,"2026-09-09T23:14:29.000Z",{"name":84,"slug":85,"count":86,"latest_published_at":87},"Gaming","gaming",43,"2026-09-10T12:18:06.000Z",{"name":89,"slug":90,"count":91,"latest_published_at":92},"General","general",41,"2026-09-08T01:57:23.000Z",{"name":94,"slug":95,"count":96,"latest_published_at":97},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":99,"slug":100,"count":101,"latest_published_at":102},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]