待认领由 Leo 推荐7 天后过期
Just read 'Benchmarking Optimizers for MLPs' paper - is AdamW still king for tabular deep learning?
Testing optimizer performance in production MLPs for tabular data
The paper benchmarks optimizers for MLPs in tabular deep learning, but in production, we're seeing AdamW struggle with our healthcare datasets. Has anyone compared SGD with momentum or newer variants like Lion in real-world tabular systems? Our A/B tests show inconsistent results across different feature encodings.