[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-a-memory-trick-lets-weaker-devices-join-federated-learning":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},9110,"a-memory-trick-lets-weaker-devices-join-federated-learning","A Memory Trick Lets Weaker Devices Join Federated Learning","Researchers split model training between cheap and costly methods so memory-limited edge devices can still join federated learning.","A federated learning framework now splits AI training into two optimization styles so budget edge devices can take part without running out of memory.\n\nThe new method, called HO-FL, tackles a known problem in federated learning: devices with little memory can't use standard backpropagation-based training, so they rely on a slower technique called zeroth-order optimization instead. HO-FL trains the bottom layers of a model with that slower, lighter method and the top layers with full backpropagation, letting each device pick its own split point based on how much memory it has. All these differently-configured devices still contribute to training the same shared model. The researchers, from HKU's WILL Lab, also found that devices doing more backpropagation-based training produce more accurate updates, but weighting the global model toward those devices risks ignoring the data held by everyone else - a trade-off they address with a sampling scheme that balances accuracy against representation.\n\nFederated learning's pitch has always been training one model across many devices without centralizing anyone's data, but that promise breaks down when some devices simply can't afford the memory. By letting each device set its own point on the memory-to-accuracy dial, HO-FL sidesteps the usual choice between leaving slow devices out entirely or dragging the whole system down to the slowest common denominator. That matters most for anyone trying to run on-device training on phones, IoT hardware, or other constrained gear rather than in a data center.\n\nThe tests so far are limited to language tasks and a single lab's benchmarks, so whether the memory savings hold up on vision or multimodal models - or at real-world scale - is still an open question.","[\"federated-learning\",\"edge-computing\",\"machine-learning\",\"optimization\"]","2026-10-01T04:00:00.000Z","2026-10-01T20:27:20.935Z","2026-10-01T20:27:24.322Z","published",null,[],"ai",[26,27,28,29],"federated-learning","edge-computing","machine-learning","optimization",[31],{"name":32,"url":33},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2609.39074",0,{"sections":36},[37,40,44,48,53,58,62,67,72,76,81,86,91,96],{"name":38,"slug":24,"count":39,"latest_published_at":18},"AI",5572,{"name":41,"slug":42,"count":43,"latest_published_at":18},"Security","security",815,{"name":45,"slug":46,"count":47,"latest_published_at":18},"Policy","policy",430,{"name":49,"slug":50,"count":51,"latest_published_at":52},"Deals","deals",298,"2026-09-30T21:00:26.000Z",{"name":54,"slug":55,"count":56,"latest_published_at":57},"Hardware","hardware",196,"2026-09-30T13:00:00.000Z",{"name":59,"slug":60,"count":61,"latest_published_at":18},"Science","science",163,{"name":63,"slug":64,"count":65,"latest_published_at":66},"Consumer Tech","consumer-tech",149,"2026-09-30T22:57:11.000Z",{"name":68,"slug":69,"count":70,"latest_published_at":71},"Dev Tools","dev-tools",93,"2026-10-01T02:30:48.000Z",{"name":73,"slug":74,"count":70,"latest_published_at":75},"Software","software","2026-09-30T21:41:11.000Z",{"name":77,"slug":78,"count":79,"latest_published_at":80},"Startups","startups",84,"2026-09-30T20:39:09.000Z",{"name":82,"slug":83,"count":84,"latest_published_at":85},"Gaming","gaming",51,"2026-09-30T16:24:30.000Z",{"name":87,"slug":88,"count":89,"latest_published_at":90},"General","general",50,"2026-09-30T21:37:54.000Z",{"name":92,"slug":93,"count":94,"latest_published_at":95},"Reviews","reviews",31,"2026-09-28T14:31:34.000Z",{"name":97,"slug":98,"count":99,"latest_published_at":100},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]