Corpus ID: 232352674

AutoLoss-Zero: Searching Loss Functions from Scratch for Generic Tasks

@article{Li2021AutoLossZeroSL,
  title={AutoLoss-Zero: Searching Loss Functions from Scratch for Generic Tasks},
  author={Hao Li and Tianwen Fu and Jifeng Dai and Hongsheng Li and Gao Huang and Xizhou Zhu},
  journal={ArXiv},
  year={2021},
  volume={abs/2103.14026}
}
  • Hao Li, Tianwen Fu, +3 authors Xizhou Zhu
  • Published 2021
  • Computer Science
  • ArXiv
Significant progress has been achieved in automating the design of various components in deep networks. However, the automatic design of loss functions for generic tasks with various evaluation metrics remains under-investigated. Previous works on handcrafting loss functions heavily rely on human expertise, which limits their extendibility. Meanwhile, existing efforts on searching loss functions mainly focus on specific tasks and particular metrics, with taskspecific heuristics. Whether such… Expand

Figures and Tables from this paper

References

SHOWING 1-10 OF 69 REFERENCES
AM-LFS: AutoML for Loss Function Search
  • 15
  • Highly Influential
  • PDF
Loss Function Discovery for Object Detection via Convergence-Simulation Driven Search
  • 1
  • Highly Influential
  • PDF
AutoML: A Survey of the State-of-the-Art
  • 130
  • PDF
Loss Function Search for Face Recognition
  • 5
  • Highly Influential
  • PDF
Efficient Optimization for Rank-Based Loss Functions
  • 17
  • PDF
Efficient Neural Architecture Search via Parameter Sharing
  • 1,180
  • Highly Influential
  • PDF
Evolving Normalization-Activation Layers
  • 16
  • Highly Influential
  • PDF
AutoML-Zero: Evolving Machine Learning Algorithms From Scratch
  • 39
  • PDF
AutoAugment: Learning Augmentation Strategies From Data
  • 488
  • Highly Influential
  • PDF
...
1
2
3
4
5
...