NeurIPS’22 Cross-Domain MetaDL Challenge: Results and lessons learned

Dustin Carrión-Ojeda, Mahbubul Alam, Sergio Escalera, Ahmed Farahat, Dipanjan Ghosh, Teresa Gonzalez Diaz, Chetan Gupta, Isabelle Guyon, Joël Roman Ky, Xian Yeow Lee, Xin Liu, Felix Mohr, Manh Hung Nguyen, Emmanuel Pintelas, Stefan Roth, Simone Schaub-Meyer, Haozhe Sun, Ihsan Ullah, Joaquin Vanschoren, Lasitha Vidyaratne, Jiamin Wu, Xiaotian Yin
Proceedings of the NeurIPS 2022 Competitions Track, PMLR 220:50-72, 2022.

Abstract

Deep neural networks have demonstrated the ability to outperform humans in multiple tasks, but they often require substantial amounts of data and computational resources. These resources may be limited in certain fields. Meta-learning seeks to overcome these challenges by utilizing past task experiences to efficiently solve new tasks, achieving better performance with limited training data and modest computational resources. To further advance the ChaLearn MetaDL competition series, we organized the Cross-Domain MetaDL Challenge for NeurIPS’22. This challenge aimed to solve “any-way" and “any-shot" tasks from 10 domains through cross-domain meta-learning. In this paper, authored collaboratively by the competition organizers, top-ranked participants, and external collaborators, we describe the technical aspects of the competition, baseline methods, and top-ranked approaches that have been open-sourced. Additionally, we provide a detailed analysis of the competition results. Lessons learned from this competition include the critical role of pre-trained backbones, the necessity of preventing overfitting, and the significance of using data augmentation or domain adaptation techniques in conjunction with extra optimizations to improve performance.

Cite this Paper

BibTeX
@InProceedings{pmlr-v220-carrion-ojeda23a, title = {NeurIPS’22 Cross-Domain MetaDL Challenge: Results and lessons learned}, author = {Carri\'{o}n-Ojeda, Dustin and Alam, Mahbubul and Escalera, Sergio and Farahat, Ahmed and Ghosh, Dipanjan and Gonzalez Diaz, Teresa and Gupta, Chetan and Guyon, Isabelle and Ky, Jo\"el Roman and Lee, Xian Yeow and Liu, Xin and Mohr, Felix and Nguyen, Manh Hung and Pintelas, Emmanuel and Roth, Stefan and Schaub-Meyer, Simone and Sun, Haozhe and Ullah, Ihsan and Vanschoren, Joaquin and Vidyaratne, Lasitha and Wu, Jiamin and Yin, Xiaotian}, booktitle = {Proceedings of the NeurIPS 2022 Competitions Track}, pages = {50--72}, year = {2022}, editor = {Ciccone, Marco and Stolovitzky, Gustavo and Albrecht, Jacob}, volume = {220}, series = {Proceedings of Machine Learning Research}, month = {28 Nov--09 Dec}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v220/carrion-ojeda23a/carrion-ojeda23a.pdf}, url = {https://proceedings.mlr.press/v220/carrion-ojeda23a.html}, abstract = {Deep neural networks have demonstrated the ability to outperform humans in multiple tasks, but they often require substantial amounts of data and computational resources. These resources may be limited in certain fields. Meta-learning seeks to overcome these challenges by utilizing past task experiences to efficiently solve new tasks, achieving better performance with limited training data and modest computational resources. To further advance the ChaLearn MetaDL competition series, we organized the Cross-Domain MetaDL Challenge for NeurIPS’22. This challenge aimed to solve “any-way" and “any-shot" tasks from 10 domains through cross-domain meta-learning. In this paper, authored collaboratively by the competition organizers, top-ranked participants, and external collaborators, we describe the technical aspects of the competition, baseline methods, and top-ranked approaches that have been open-sourced. Additionally, we provide a detailed analysis of the competition results. Lessons learned from this competition include the critical role of pre-trained backbones, the necessity of preventing overfitting, and the significance of using data augmentation or domain adaptation techniques in conjunction with extra optimizations to improve performance.} }
Endnote
%0 Conference Paper %T NeurIPS’22 Cross-Domain MetaDL Challenge: Results and lessons learned %A Dustin Carrión-Ojeda %A Mahbubul Alam %A Sergio Escalera %A Ahmed Farahat %A Dipanjan Ghosh %A Teresa Gonzalez Diaz %A Chetan Gupta %A Isabelle Guyon %A Joël Roman Ky %A Xian Yeow Lee %A Xin Liu %A Felix Mohr %A Manh Hung Nguyen %A Emmanuel Pintelas %A Stefan Roth %A Simone Schaub-Meyer %A Haozhe Sun %A Ihsan Ullah %A Joaquin Vanschoren %A Lasitha Vidyaratne %A Jiamin Wu %A Xiaotian Yin %B Proceedings of the NeurIPS 2022 Competitions Track %C Proceedings of Machine Learning Research %D 2022 %E Marco Ciccone %E Gustavo Stolovitzky %E Jacob Albrecht %F pmlr-v220-carrion-ojeda23a %I PMLR %P 50--72 %U https://proceedings.mlr.press/v220/carrion-ojeda23a.html %V 220 %X Deep neural networks have demonstrated the ability to outperform humans in multiple tasks, but they often require substantial amounts of data and computational resources. These resources may be limited in certain fields. Meta-learning seeks to overcome these challenges by utilizing past task experiences to efficiently solve new tasks, achieving better performance with limited training data and modest computational resources. To further advance the ChaLearn MetaDL competition series, we organized the Cross-Domain MetaDL Challenge for NeurIPS’22. This challenge aimed to solve “any-way" and “any-shot" tasks from 10 domains through cross-domain meta-learning. In this paper, authored collaboratively by the competition organizers, top-ranked participants, and external collaborators, we describe the technical aspects of the competition, baseline methods, and top-ranked approaches that have been open-sourced. Additionally, we provide a detailed analysis of the competition results. Lessons learned from this competition include the critical role of pre-trained backbones, the necessity of preventing overfitting, and the significance of using data augmentation or domain adaptation techniques in conjunction with extra optimizations to improve performance.
APA
Carrión-Ojeda, D., Alam, M., Escalera, S., Farahat, A., Ghosh, D., Gonzalez Diaz, T., Gupta, C., Guyon, I., Ky, J.R., Lee, X.Y., Liu, X., Mohr, F., Nguyen, M.H., Pintelas, E., Roth, S., Schaub-Meyer, S., Sun, H., Ullah, I., Vanschoren, J., Vidyaratne, L., Wu, J. & Yin, X.. (2022). NeurIPS’22 Cross-Domain MetaDL Challenge: Results and lessons learned. Proceedings of the NeurIPS 2022 Competitions Track, in Proceedings of Machine Learning Research 220:50-72 Available from https://proceedings.mlr.press/v220/carrion-ojeda23a.html.

Related Material