Search papers, labs, and topics across Lattice.
Affiliation:
2
0
3
Foundation models are practically useless for lossless time-series compression, yet shifting to error-bounded lossy regimes lets them outperform classical predictors across the board by collapsing in-band residual costs to zero.
A 135M-parameter transformer-based compressor, Nacrith, achieves state-of-the-art lossless compression, beating gzip by 3.1x and ts_zip by 20% on alice29.txt, and surpassing FineZip by 8% on enwik8, all with a model 60x smaller and without fine-tuning.
