Search papers, labs, and topics across Lattice.
Rensselaer Polytechnic Institute
3
0
6
LLMs can slash memory use by 4x during reasoning without sacrificing accuracy, simply by "zooming in" on relevant cached information instead of attending to everything.
Contrary to claims that RLVR can handle noisy data, this work reveals that current RLVR methods still suffer significantly from data quality issues, with performance dropping 8-12% when trained on truly noisy data.
By grounding reasoning within the topology of a global interaction graph, ManCAR achieves up to 46.88% relative improvement in NDCG@10 compared to state-of-the-art sequential recommendation baselines.
