Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
2
0
3
Ditch quadratic attention and unlock structured time series representations: ChronoVAE-HOPE achieves strong classification performance by disentangling trend and seasonal components with a novel dual-memory VAE.
MoEITS slashes the computational cost of MoE-LLMs while boosting accuracy, outperforming existing pruning methods on models like Mixtral and DeepSeek.