Meta-Learning with Task-Adaptive Loss Function for Few-Shot Learning
Abstract
In few-shot learning scenarios, the challenge is to generalize and perform well on new unseen examples when only very few labeled examples are available for each task. Model-agnostic meta-learning (MAML) has gained the popularity as one of the representative few-shot learning methods for its flexibility and applicability to diverse problems. However, MAML and its variants often resort to a simple loss function without any auxiliary loss function or regularization terms that can help achieve better generalization. The problem lies in that each application and task may require different auxiliary loss function, especially when tasks are diverse and distinct. Instead of attempting to hand-design an auxiliary loss function for each application and task, we introduce a new meta-learning framework with a loss function that adapts to each task. Our proposed framework, named Meta-Learning with Task-Adaptive Loss Function (MeTAL), demonstrates the effectiveness and the flexibility across various domains, such as few-shot classification and few-shot regression.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
MoE models represent a significant innovation in neural network design, offering enhanced scalability and efficiency in training and inference.
“MoE models represent a significant innovation in neural network design, offering enhanced scalability and efficiency in training and inference [287], [288], [110].”
From From Google Gemini to OpenAI Q* (Q-Star): A Survey on Reshaping the Generative Artificial Intelligence (AI) Research Landscape (2025) · cites this paper as [287]
MoE models demonstrate a substantial advantage in pretraining speed, outperforming dense models.
“MoE models demonstrate a substantial advantage in terms of pretraining speed, outperforming dense models [94], [287].”
From From Google Gemini to OpenAI Q* (Q-Star): A Survey on Reshaping the Generative Artificial Intelligence (AI) Research Landscape (2025) · cites this paper as [287]
Models like Mixtral and Switch Transformer, with over 1.6 trillion parameters, achieved computational efficiency equivalent to a 10 billion-parameter dense model, because they benefited from the sublinear scaling of MoE compute versus model size, leading to substantial accuracy gains within fixed compute budgets.
“Notably, models like Mixtral and Switch Transformer, with over 1.6 trillion parameters, achieved computational efficiency equivalent to a 10 billion-parameter dense model, because they benefited from the sublinear scaling of MoE compute versus model size, leading to substantial accuracy gains within fixed compute budgets [21], [289], [287], [110].”
From From Google Gemini to OpenAI Q* (Q-Star): A Survey on Reshaping the Generative Artificial Intelligence (AI) Research Landscape (2025) · cites this paper as [287]
DeepSpeed-MoE included model compression techniques, reducing model size by up to 3.7x while maintaining accuracy, and an end-to-end MoE training and inference solution, part of the DeepSpeed library, which was instrumental in serving large-scale MoE models with enhanced speed and cost efficiency.
“Moreover, DeepSpeed-MoE included model compression techniques, reducing model size by up to 3.7x while maintaining accuracy, and an end-to-end MoE training and inference solution, part of the DeepSpeed library, which was instrumental in serving large-scale MoE models with enhanced speed and costefficiency [287].”
From From Google Gemini to OpenAI Q* (Q-Star): A Survey on Reshaping the Generative Artificial Intelligence (AI) Research Landscape (2025) · cites this paper as [287]
DeepSpeedMoE offered new architectural designs and model compression, decreasing MoE model size by approximately 3.7x and optimizing inference to achieve up to 7.3x better latency and cost efficiency.
“Innovations like DeepSpeedMoE [287] offered new architectural designs and model compression, decreasing the MoE model size by approximately 3.7x and optimizing inference to achieve up to 7.3x better latency and cost efficiency.”
From From Google Gemini to OpenAI Q* (Q-Star): A Survey on Reshaping the Generative Artificial Intelligence (AI) Research Landscape (2025) · cites this paper as [287]
DeepSpeed-MoE introduces advanced parallelism modes like data parallelism, tensor-slicing for non-expert parameters, and expert parallelism for expert parameters, enhancing model efficiency, optimizing latency and throughput in MoE model inference, and offering scalable solutions in production environments using multiple GPU devices.
“DeepSpeed-MoE, for instance, introduces advanced parallelism modes like data parallelism, tensor-slicing for non-expert parameters, and expert parallelism for expert parameters, enhancing model efficiency, as their approach optimizes both latency and throughput in MoE model inference, offering scalable solutions in production environments using multiple Graphics Processing Unit (GPU) devices [287]”
From From Google Gemini to OpenAI Q* (Q-Star): A Survey on Reshaping the Generative Artificial Intelligence (AI) Research Landscape (2025) · cites this paper as [287]
Cites
ImageNet Large Scale Visual Recognition Challenge
Russakovsky, O., Deng, J., Su, H., Krause, J., Satheesh, S., Ma, S., Huang, Z., Karpathy, A., Khosla, A., Bernstein, M., et al. (2015) · International Journal of Computer Vision · doi:10.1007/s11263-015-0816-y · Cited by 39,380 · see citing works
Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
Finn, C., Abbeel, P., & Levine, S. (2017) · arXiv · doi:10.48550/arxiv.1703.03400 · Cited by 8,512 · see citing works
Prototypical Networks for Few-shot LearningPreprint
Snell, J., Swersky, K., & Zemel, R. S. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1703.05175 · Cited by 5,863 · see citing works
Caltech-UCSD Birds-200-2011 Dataset
Wah, C. (2024) · CaltechAUTHORS (California Institute of Technology) · doi:10.57702/i9oeewf6 · Cited by 3,875 · see citing works
Siamese Neural Networks for One-shot Image Recognition
Koch, G. E., Zemel, R. S., & Salakhutdinov, R. (2015) · Cited by 3,712 · see citing works
Optimization as a Model for Few-Shot Learning
Ravi, S., & Larochelle, H. (2017) · International Conference on Learning Representations · Cited by 2,766 · see citing works
Film: Visual reasoning with a general conditioning layer
Perez, E., Strub, F., De Vries, H., Dumoulin, V., & Courville, A. (2018) · Proceedings of the AAAI Conference on Artificial Intelligence · doi:10.1609/aaai.v32i1.11671 · Cited by 2,490 · see citing works
Meta-learning with memory-augmented neural networks
Santoro, A., Bartunov, S., Botvinick, M., Wierstra, D., & Lillicrap, T. (2016) · International Conference on Machine Learning · Cited by 1,504 · see citing works
On First-Order Meta-Learning Algorithms
Nichol, A., Achiam, J., & Schulman, J. (2018) · arXiv · doi:10.48550/arxiv.1803.02999 · Cited by 1,438 · see citing works
Meta-Learning With Differentiable Convex Optimization
Lee, K., Maji, S., Ravichandran, A., & Soatto, S. (2019) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2019.01091 · Cited by 1,351 · see citing works
Self-citationMeta-SGD: Learning to Learn Quickly for Few-Shot LearningPreprint
Li, Z., Zhou, F., Chen, F., & Li, H. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1707.09835 · Cited by 1,026 · see citing works
A Simple Neural Attentive Meta-LearnerPreprint
Mishra, N., Rohaninejad, M., Chen, X., & Abbeel, P. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1707.03141 · Cited by 961 · see citing works
Meta-Learning for Semi-Supervised Few-Shot ClassificationPreprint
Ren, M., Triantafillou, E., Ravi, S., Snell, J., Swersky, K., Tenenbaum, J. B., Larochelle, H., & Zemel, R. S. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1803.00676 · Cited by 892 · see citing works
Learning to Learn Using Gradient Descent
Hochreiter, S., Younger, A. S., & Conwell, P. R. (2001) · Lecture notes in computer science · doi:10.1007/3-540-44668-0_13 · Cited by 575 · see citing works
Task Agnostic Meta-Learning for Few-Shot Learning
Jamal, M. A., & Qi, G.-J. (2019) · 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2019.01199 · Cited by 519 · see citing works
Learning to Propagate Labels: Transductive Propagation Network for Few-shot LearningPreprint
Liu, Y., Lee, J., Park, M., et al. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1805.10002 · Cited by 517 · see citing works
Meta-Gradient Reinforcement LearningPreprint
Xu, Z., Hasselt, H. V., & Silver, D. (2018) · Neural Information Processing Systems · doi:10.48550/arxiv.1805.09801 · Cited by 388 · see citing works
Learning to Control Fast-Weight Memories: An Alternative to Dynamic Recurrent Networks
Schmidhuber, J. (1992) · Neural Computation · doi:10.1162/neco.1992.4.1.131 · Cited by 357 · see citing works
On the Optimization of a Synaptic Learning Rule
Samy Bengio, A., Bengio, Y., Cloutier, J., & Gescei, J. (2013) · doi:10.4324/9780203773833-22 · Cited by 181 · see citing works
Towards Sample Efficient Reinforcement Learning
Yu, Y. (2018) · International Joint Conference on Artificial Intelligence · doi:10.24963/ijcai.2018/820 · Cited by 161 · see citing works
Hierarchically Structured Meta-learningPreprint
Yao, H., Wei, Y., Huang, J., & Li, Z. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.05301 · Cited by 154 · see citing works
Rapid Adaptation with Conditionally Shifted NeuronsPreprint
Munkhdalai, T., Yuan, X., Mehri, S., et al. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1712.09926 · Cited by 106 · see citing works
Learning to Forget for Meta-Learning
Baik, S., Hong, S., & Lee, K. M. (2020) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr42600.2020.00245 · Cited by 105 · see citing works
Self-citationLearning to Learn: Meta-Critic Networks for Sample Efficient LearningPreprint
Sung, F., Li, Z., Xiang, T., Hospedales, T. M., & Yang, Y. (2017) · arXiv (Cornell University) · doi:10.48550/arxiv.1706.09529 · Cited by 91 · see citing works
Meta Learning via Learned Loss
Bechtle, S., Molchanov, A., Chebotar, Y., Grefenstette, E., Righetti, L., Sukhatme, G. S., & Meier, F. (2021) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr48806.2021.9412010 · Cited by 77 · see citing works
Alpha MAML: Adaptive Model-Agnostic Meta-LearningPreprint
Behl, H. S., Baydin, A. G., & Torr, P. H. S. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.07435 · Cited by 56 · see citing works
Learning to Learn with Conditional Class Dependencies
Jiang, X., Havaei, M., Varno, F., et al. (2018) · International Conference on Learning Representations · Cited by 55 · see citing works
On Modulating the Gradient for Meta-learning
Simon, C., Koniusz, P., Nock, R., & Harandi, M. T. (2020) · Lecture notes in computer science · doi:10.1007/978-3-030-58598-3_33 · Cited by 49 · see citing works
Meta-Gradient Reinforcement Learning with an Objective Discovered OnlinePreprint
Xu, Z., van Hasselt, H. P., Hessel, M., Oh, J., Singh, S. P., & Silver, D. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2007.08433 · Cited by 42 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
McIntosh, T. R., Susnjak, T., Liu, T., Watters, P., Xu, D., Liu, D., & Halgamuge, M. N. (2025) · Technologies · doi:10.3390/technologies13020051 · Cited by 102 · see citing works
“It focuses on few-shot generalization, enabling AI to handle a wide range of tasks with minimal data, underlining its importance in developing versatile and adaptable AI systems [255], [256], [254], [257].”
Background
- Authors
- Baik, S., Choi, J., Kim, H., Cho, D., Min, J., & Lee, K. M.
- Affiliations
- Seoul National University (KR)
- Hyundai Motor Group (South Korea) (KR)
- Venue
- 2021 IEEE/CVF International Conference on Computer Vision (ICCV)
- Publisher
- IEEE
- Published
- 2021-10
- Pages
- 9445-9454
- OpenAlex
- W3202188231
- Cited by
- 163
- References
- 29
- Citations by year
- 2026: 19 · 2025: 42 · 2024: 37 · 2023: 45 · 2022: 20
- FWCI
- 11.78
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- IEEE/CVF International Conference on Computer Vision (ICCV) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Data Classification
- Physical Sciences › Computer Science › Artificial Intelligence › Machine Learning and Algorithms
- Keywords
- meta-learning, few-shot learning, model-agnostic meta-learning, auxiliary loss function, regularization, few-shot classification, generalization
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
