Just concluded 1st-place in Kaggle with a great team consisting of Geremie Yeo, Gaurav Rawat, Young Min Paik, and Yevhenii Maslov in the PII Data Detection Competition, competing against over 2000 teams. Many thought NER is a simple task, not when you are faced with real world cases. See that Presidio just do not cut the task alone :) Our approach combined post-processing, ensemble techniques, and the training of transformer models like Deberta, Longformer and other models across various architectures, such as multi-dropout and BiLSTM, and knowledge distillation. Solution Write-up: https://lnkd.in/g5rzkiMx We continue to investigate various emerging LLM architectures and found that code pretrained LLMs are great at identifying NER too. Due to time constraints and work commitment, we have yet to integrate that to our solution. But the ensembles done by the team is already strong enough to land 1st. Cheers to the good team effort!
Congrats to you and the team, Patrick Chan! We've been testing with NER for some time, and I completely agree with you that it's not a simple task, especially when considering real-world use cases and trying to augment existing products! 👏
Congratulations Dr.
Congratulations !
Congrats Dr Patrick
Congrats Dr. Patrick ... Great going!
Great effort indeed ! Congratulations Dr. Patrick and team.
Congratulations Dr Patrick!
Congratulations Dr Patrick!
Quant | Kaggle Competition Master
4moCongratulations