Search papers, labs, and topics across Lattice.
Affiliation:
4
0
7
5
Pruning models can lead to smaller, more accurate prediction sets without sacrificing reliability, as shown by CPP's impressive performance on large-label tasks.
Uniform bagging's performance degrades significantly with Markov-dependent data, but a novel spectral routing algorithm can restore minimax optimality without knowledge of the mixing time.
Forget accuracy: a spectral "fingerprint" derived from Random Matrix Theory can reveal hidden overfitting in your model, even when standard metrics look good.
GPT-2 wastes 88% of its attention on predictable information, but this new method learns to forget, achieving a 37.8x compute reduction by dynamically consolidating episodic retrievals into parametric memory.
