Happy that a key paper from my PhD is finally out: PEDL+ (github.com/leonweber/pedl) is a tool for extracting protein-protein and protein-chemical associations from a large portion of the biomedical literature and processing the results (who doesn't love Excel tables?). Proud (1/2)
Latest posts 5
Come and check out our poster on joint 馃&馃pipelines for finding annotation errors. Today from 11:40 to 12:30 at TrustNLP @ #ACL2023NLP
馃摑 ActiveAED: A Human in the Loop Improves Annotation Error DetectionView quoted post on X
馃攳 This work shows that you can improve the performance of annotation error detection methods by using them in an active learning setup.
馃懃@weberple, @barbara_plank
馃敆arxiv.org/abs/2305.20045
馃搧Findings-Short
I will be at ACL in Toronto. If you want to (re-)connect, reach out!
If you want to detect labelling errors in your dataset, it might be a good idea to do it in an active learning setting for maximum performance. See our ACL 2023 Findings paper 猬囷笍 for how you could do that. Joint work with @barbara_plank
馃摑 ActiveAED: A Human in the Loop Improves Annotation Error DetectionView quoted post on X
馃攳 This work shows that you can improve the performance of annotation error detection methods by using them in an active learning setup.
馃懃@weberple, @barbara_plank
馃敆arxiv.org/abs/2305.20045
馃搧Findings-Short
Does anyone have an intuition about good hyperparameter ranges for finetuning Pythia models? @AiEleuther @BlancheMinerva