Generative recommenders lose all top-10 paths at the first beam step; reallocation halves budget
_reachsumit · x · 2026-11-04
Paper 2611.01072 identifies a "first-step coverage bottleneck" in generative recommendation: across six TIGER test settings, every pruned exact top-10 path is lost at the initial decoding step.
- Moving beam width to the first step raises Search Recall@10 by 0.96–10.90 points at Beam-50's unchanged budget
- With RQ-VAE tokenization (77,373 test users), the reallocated schedule reproduces Beam-100's ordered top-10 at a nominal budget of 151 instead of 301, no retraining needed
- Matched-output timing saves 6.4% decoder latency
More from Research
- Noetra engineer's first project: building an image encoder from scratch for multimodal LLMs — HirokatuKataoka · 2026-11-04
- Japanese team trains image encoder from scratch to match SigLIP2, with better CJK performance — HirokatuKataoka · 2026-11-04
- SpecJudge author argues LLM-as-judge tools should never ask models for self-reported confidence — jokiruiz · 2026-11-04
- Tsinghua's BEMULA benchmark: no LLM agent memory system passes over 1.5% of forget requests — rohanpaul_ai · 2026-11-04
- KAIST AI's FactEmbed decouples query understanding to improve composed image retrieval — kaist-ai · 2026-11-04
- Intology's Locus auto-designs molecular models, beating GPT-6 and Claude agents — rohanpaul_ai · 2026-11-04