Search papers, labs, and topics across Lattice.
1
0
Concatenating LLM features can lead to a surprising 17% drop in accuracy for GNNs on homophilous benchmarks, challenging the assumption that more features always enhance performance.
