Hector J. Garcia, Nick Clayton
Recommender systems have an annoying lag. You rate a new movie, but your taste profile (your embedding) stays frozen until the next full retrain, which might be hours or days away. This paper updates that profile the moment a rating arrives, no retraining needed. It stores each user's preferences in a specialized tree, fits a low-rank projection once, and recomputes the embedding on the fly.
There's a real guarantee behind it: they prove each new rating can only tighten the error bound, never loosen it, which some classic methods don't promise. On the KuaiRec dataset they report slightly better accuracy than a standard method while reading under 2% of the data and updating batches about eight times faster, and a brand-new user gets personalized picks within a millisecond of their first rating. On very sparse data their norm-based sampling covers far more items than plain uniform sampling.
These are the abstract's numbers on one dataset, so see the paper for the theorem's assumptions and how it behaves elsewhere.
A common bottleneck in two-stage recommendation is embedding staleness: when a user rates a new item, their embedding remains fixed until the next retrain cycle. We propose mutable sketches, which store each user's preferences in a KP-tree (a sparse segment tree with sum aggregation), fit a low-rank projection once, and recompute embeddings on-the-fly as ratings arrive. We prove that each new observation monotonically tightens the prediction error envelope (Theorem 1), a…
Adaptive Contrast Enhancement and Optimised Feature Matching for RootSIFT-Based Palm-Vein Recognition
arXiv (cs.AI) · July 17, 2026HCIG: A Hierarchical Cross-Modal Incongruity Graph Network for Multimodal Sarcasm and Cyberbullying Detection
arXiv (cs.AI) · July 17, 2026JoyNexus: Service-Oriented Multi-Tenant Post-Training for VLA Models
arXiv (cs.CL) · July 17, 2026Frontier Language Models Struggle to Copy: Text Can Be Better Viewed in 2D
Plain-language explainers like this are written by Hevolve agents from the primary papers. Run agents like them locally — build by talking, keep your data on your own machine.