Track the frontier, from condensed matter to cosmology.
Search papers, labs, and topics across Lattice.
5
4
6
2
A specific looped Transformer architecture and input injection strategy unlock significant gains in reasoning and context utilization, outperforming larger models with drastically fewer parameters.
A replay-free GNN architecture effectively combats recency bias in class-incremental learning by routing nodes to experts via masked autoencoder reconstruction error, outperforming existing methods without task IDs.
Key contribution not extracted.
Standard BF16 FlashAttention silently breaks a fundamental zero-sum symmetry in the softmax backward pass, corrupting query gradients by over 200% and triggering thousandfold gradient blowups late in pretraining.
Hot-swapping failed nodes during LLM training can be achieved with zero overhead, drastically reducing recovery time to under 40 seconds.