Search papers, labs, and topics across Lattice.
4
0
8
Skip the needle: a new AI model estimates hemoglobin levels from PPG signals with surprising accuracy, paving the way for non-invasive anemia screening.
LLMs still struggle with multilingual reasoning, exhibiting marked accuracy drops in Hindi compared to English on real-world Indian educational benchmarks, even with chain-of-thought prompting.
LLM safety filters can get smarter: this adaptive abstention system slashes false positives by dynamically adjusting thresholds based on context, without sacrificing safety or speed.
L1-norm pruning is leaving performance on the table: SHAP-inspired pruning boosts object detection efficiency by up to 10% without sacrificing accuracy.
