Search papers, labs, and topics across Lattice.
2
0
4
0
Forget test-time training: this work bakes optimal control directly into LLMs, yielding up to 27.8% gains in mathematical reasoning.
Forget scaling bottlenecks: DG-PG uses differentiable analytical models to slash gradient variance in cooperative MARL, achieving convergence in a 200-agent cloud scheduling task where standard methods fail.
