LoopRec applies item-level adaptive recursion to sequential recommendation
October 20–21 reports restate the LoopRec method, benchmark gains, and GitHub release, adding no facts after the October 24 abstract note.
LoopRec, a shared-block recommender with per-item exit gates, beats matched fixed-depth models on public benchmarks.
LoopRec replaces a fixed computation budget in sequential recommendation with item-level adaptive recursion. One weight-shared block is applied repeatedly, and a lightweight gate independently decides whether each history item exits or continues, without auxiliary depth labels. The authors report that most items converge after shallow processing while a minority needs full depth, and that an oracle non-uniform allocation outranks a fixed budget. On public benchmarks across four backbone families, LoopRec beats matched fixed-depth models and assigns more depth to lower-frequency and more recent items. Code is released at github.com/Elpmis117/LoopRec.
- A binary exit-or-continue gate runs at each recursion boundary.
- Training uses no auxiliary depth labels.
- Tests span four backbone families on public benchmarks.
- Lower-frequency and more recent items receive deeper computation.
- Code is at github.com/Elpmis117/LoopRec.
What each outlet adds
beyond the shared facts- Hugging Face daily papersStates the code release and independent per-item exit decisions.
- arXiv cs.AI / cs.LG / cs.CLMost items converge early; oracle non-uniform depth beats fixed allocation.
Coverage
oldest first · one row per article · UTC- · 5d ago
Hugging Face daily papersFirst report
LoopRec assigns per-item recursion depth in sequential recommenders
A shared block and per-item exit gate beat matched fixed-depth models; rarer and more recent items get deeper compute.
- · 4d ago
arXiv cs.AI / cs.LG / cs.CL
LoopRec varies recursion depth per item in sequential recommenders
A shared block and per-item exit gate beat matched fixed-depth models on public benchmarks across four backbone families.