Curriculum learning
The DOI matches its Crossref record; 35 of the 40 references it cites were found in a registry or an index.
Abstract
Humans and animals learn much better when the examples are not randomly presented but organized in a meaningful order which illustrates gradually more concepts, and gradually more complex ones. Here, we formalize such training strategies in the context of machine learning, and call them "curriculum learning". In the context of recent research studying the difficulty of training in the presence of non-convex training criteria (for deep deterministic and stochastic neural networks), we explore curriculum learning in various set-ups. The experiments show that significant improvements in generalization can be achieved. We hypothesize that curriculum learning has both an effect on the speed of convergence of the training process to a minimum and, in the case of non-convex criteria, on the quality of the local minima obtained: curriculum learning can be seen as a particular form of continuation method (a general strategy for global optimization of non-convex functions).
Overview
Integrity signals
Registry
Cites
Statistics
Verification
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?
In the citing paper
“Moreover, the lack of acoustical signals could even account for spatial disorientation 91,92.”
From The impact of multisensory integration and perceptual load in virtual reality settings on performance, workload and presence (2021) · cites this paper as [92]
In the citing paper
“This suggests that curriculum-style learning (Bengio et al., 2009), where simpler or archetypal problems are presented first to build a repository of valid heuristics, may potentially bootstrap performance.”
From Dynamic Cheatsheet: Test-Time Learning with Adaptive Memory (2025) · cites this paper as Bengio et al., 2009
Citations
Cites
Reducing the dimensionality of data with neural networks
Hinton, G. E., & Salakhutdinov, R. R. (2006) · Science · doi:10.1126/science.1127647 · Cited by 20,928
A fast learning algorithm for deep belief nets
Hinton, G. E., Osindero, S., & Teh, Y.-W. (2006) · Neural Computation · doi:10.1162/neco.2006.18.7.1527 · Cited by 16,336
Learning Deep Architectures for AI
Bengio, Y. (2009) · Foundations and Trends® in Machine Learning · doi:10.1561/2200000006 · Cited by 9,239
Self-citationExtracting and composing robust features with denoising autoencoders
Vincent, P., Larochelle, H., Bengio, Y., & Manzagol, P.-A. (2008) · Proceedings of the 25th international conference on Machine learning - ICML '08 · doi:10.1145/1390156.1390294 · Cited by 7,436
Self-citationA unified architecture for natural language processing: deep neural networks with multitask learning
Collobert, R., & Weston, J. (2008) · Proceedings of the 25th international conference on Machine learning - ICML '08 · doi:10.1145/1390156.1390177 · Cited by 5,058 · cited 6×
“In Collobert and Weston (2008), the main objective is to learn an embedding for the words as a side-e.ect of learning to compute this score.”
BackgroundMethodInfluentialSelf-citationLearning Deep Architectures for AI
Bengio, Y. (2009) · now publishers, Inc. eBooks · doi:10.1561/9781601982957 · Cited by 4,941
Self-citationGreedy layer-wise training of deep networks
Bengio, Y., Lamblin, P., Popovici, D., & Larochelle, H. (2007) · Advances in Neural Information Processing Systems 19 · doi:10.7551/mitpress/7503.003.0024 · Cited by 4,841
Self-citationA Neural Probabilistic Language Model
Bengio, Y., Ducharme, R., Vincent, P., Jauvin, C., Jaz, K., Hofmann, T., Poggio, T., & Shawe-Taylor, J. S. (2000) · Cited by 3,479
Self-citationLearning and development in neural networks: the importance of starting small
Elman, J. L. (1993) · Cognition · doi:10.1016/0010-0277(93)90058-4 · Cited by 1,988 · cited 8×
“The idea of training a learning machine with a curriculum can be traced back at least to Elman (1993).”
BackgroundMethodInfluentialRestricted Boltzmann machines for collaborative filtering
Salakhutdinov, R., Mnih, A., & Hinton, G. (2007) · Proceedings of the 24th international conference on Machine learning · doi:10.1145/1273496.1273596 · Cited by 1,885
Research works printed but not linked: 3
- 10.5555/89603 · doi:10.5555/89603
- 10.5555/866740 · doi:10.5555/866740
- 10.5555/525571 · doi:10.5555/525571
Cited by
Items 3,001–3,100 of 5,240
SAT-MARL: Specification Aware Training in Multi-Agent Reinforcement Learning
Ritz, F., Phan, T., Müller, R., Gabor, T., Sedlmeier, A., Zeller, M., Wieghardt, J., Schmid, R., Sauer, H., Klein, C., et al. (2021) · International Conference on Agents and Artificial Intelligence (ICAART) · doi:10.5220/0010189500280037 · Cited by 5
“Inspired by curriculum learning (Bengio et al., 2009), reward scheme rx only contains β , δ and η in the first part of the training process in order to learn the basic task. ζ , θ and ι are added later during training, so that some constraints are introduced to the agents gradually.”
MethodResultImproving Multimodal Speech Enhancement by Incorporating Self-Supervised and Curriculum Learning
Cheng, Y., He, M., Yu, J., & Feng, R. (2021) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp39728.2021.9413431 · Cited by 6
“We leverage the curriculum learning [26] to address these challenges, starting from easier tasks, and then increasing the difficulty level gradually.”
MethodP-DIFF: Learning Classifier with Noisy Labels based on Probability Difference Distributions
Hu, W., Zhao, Q., Huang, Y., & Zhang, F. (2021) · Proceedings - International Conference on Pattern Recognition/Proceedings/International Conference on Pattern Recognition · doi:10.1109/icpr48806.2021.9412550 · Cited by 10
“Based on Curriculum learning [35], some recent proposed approaches select clean training samples 2020 25th International Conference on Pattern Recognition (ICPR) Milan, Italy, Jan 10-15, 2021”
MethodJaensch, F., Steidle, A., & Verl, A. W. (2021) · International Conference on Transdisciplinary AI (TransAI) · doi:10.1109/transai51903.2021.00031 · Cited by 4
Contrastive Goal Grouping for Policy Generalization in Goal-Conditioned Reinforcement Learning
Zou, Q., & Suzuki, E. (2021) · Lecture notes in computer science · doi:10.1007/978-3-030-92185-9_20 · Cited by 4
Self-Paced Knowledge Distillation for Real-Time Image Guided Depth Completion
Wang, S., Hu, M. H., Li, B., & Gong, X. (2022) · IEEE Signal Processing Letters · doi:10.1109/lsp.2022.3154243 · Cited by 5
Background Matting via Recursive Excitation
Deng, J., Xu, Y., Zhou, Z., et al. (2022) · IEEE International Conference on Multimedia and Expo (ICME) · doi:10.1109/icme52920.2022.9859876 · Cited by 4
“In this paper, inspired by curriculum learning [15,16], we propose a simple yet effective way for further absorbing the guidance from the ground truth (GT).”
BackgroundMethodCoarse-to-Fine Minimization of Some Common Nonconvexities
Mobahi, H., & Fisher, J. S. (2015) · Lecture notes in computer science · doi:10.1007/978-3-319-14612-6_6 · Cited by 7
“Some example applications include semi-supervised kernel machines [27], multiple instance learning [14,18], semi-supervised structured output [11], and language modeling [2].”
MethodWhat And How Other Datasets Can Be Leveraged For Medical Imaging Classification
Shang, H., Sun, Z., Fu, X., et al. (2019) · IEEE International Symposium on Biomedical Imaging (ISBI) · doi:10.1109/isbi.2019.8759148 · Cited by 4
“For physicians, basic and easier knowledge are learned first, like anatomy to distinguish organs, while in machine learning, the strategy of starting small or curriculum learning, has demonstrated benefit [5].”
BackgroundAssisting the training of deep neural networks with applications to computer vision
Romero, A. (2015) · LA Referencia (Red Federada de Repositorios Institucionales de Publicaciones Científicas) · Cited by 5
Planning and exploration in stochastic relational worlds
Lang, T. (2011) · Universitätsbibliothek der FU Berlin Hochschulschriftenstelle u. Dokumentenserver · doi:10.17169/refubium-12370 · Cited by 5
Sequential learning in the form of shaping as a source of cognitive flexibility
Krueger, K. A. (2011) · UCL Discovery (University College London) · Cited by 4
Exploiting Multi-modal Curriculum in Noisy Web Data for Large-scale Concept LearningPreprint
Liang, J., Jiang, L., Meng, D., et al. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1607.04780 · Cited by 4
“The learning framework is motivated by human learning, where people generally start learning easier aspects of a concept, and then gradually take more complex examples into the learning process[1, 23, 19].”
BackgroundTowards Human-Machine Cooperation: Self-supervised Sample Mining for Object DetectionPreprint
Wang, K., Yan, X., Zhang, D., et al. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1803.09867 · Cited by 4
“In our framework, inspired by the recently proposed techniques: Curriculum Learning (CL) [2] and Selfpaced Learning (SPL) [20][18], we formulate the combining of the SSM and the AL as a concise optimization problem.”
BackgroundMethodAnticipating Traffic Accidents with Adaptive Loss and Large-scale Incident DBPreprint
Suzuki, T., Kataoka, H., Aoki, Y., et al. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1804.02675 · Cited by 4
“In our design, the loss adaptively assigns penalty weights depending on how early a model can anticipate a traffic accident at each training epoch, inspired by Curriculum Learning.”
MethodInfluentialUnlearn What You Have Learned: Adaptive Crowd Teaching with Exponentially Decayed Memory LearnersPreprint
Zhou, Y., Nelakurthi, A. R., & He, J. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1804.06481 · Cited by 9
Evolutionarily-Curated Curriculum Learning for Deep Reinforcement Learning AgentsPreprint
Green, M. C., Sergent, B., Shandilya, P., & Kumar, V. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1901.05431 · Cited by 8
Learning from Videos with Deep Convolutional LSTM NetworksPreprint
Courtney, L., & Sreenivas, R. S. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1904.04817 · Cited by 7
One step back, two steps forward: interference and learning in recurrent neural networksPreprint
Beer, C., & Barak, O. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1805.09603 · Cited by 4
“Shaping, or choosing an optimal sequence of training examples, is a topic of importance in both animal training and machine learning research [29,30].”
BackgroundGaining Extra Supervision via Multi-task learning for Multi-Modal Video Question AnsweringPreprint
Kim, J.-Y., Ma, M., Kim, K., et al. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1905.13540 · Cited by 4
“Motivated by curriculum learning [23], multi-task ratio scheduling attempts to learn easier task earlier to set inductive bias at the beginning of the training.”
BackgroundCross-Modal Scene NetworksPreprint
Aytar, Y., Castrejón, L., Vondrick, C., Pirsiavash, H., & Torralba, A. B. (2016) · arXiv (Cornell University) · doi:10.13016/m2r4pl-lmzf · Cited by 5
Optical Flow Requires Multiple Strategies (but only one network)Preprint
Schuster, T., Wolf, L., & Gadot, D. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1611.05607 · Cited by 5
“Curriculum learning [6], inspired by the learning process of humans, was the first method to manipulate the order of samples shown to the model during training.”
MethodShen, Y., & Gao, M. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1808.05744 · Cited by 4
“Curriculum learning is used in our training to stabilize our training process [2].”
MethodLearning a Multi-Domain Curriculum for Neural Machine TranslationPreprint
Wang, W., Tian, Y., Ngiam, J., et al. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1908.10940 · Cited by 4
“Curriculum learning (CL) (Bengio et al., 2009) has been used as a formulation for dynamic data selection.”
MethodOptimizing Data Usage via Differentiable RewardsPreprint
Wang, X., Pham, H., Michel, P., Anastasopoulos, A., Carbonell, J., & Neubig, G. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.10088 · Cited by 39
“…subsets of the data to train the model on (Jiang & Zhai, 2007; Wang et al.; Axelrod et al., 2011; Moore & Lewis, 2010), providing different weights for each example (Sivasankaran et al., 2017; Ren et al., 2018), or changing the presentation order of data (Bengio et al., 2009; Kumar et al., 2019).”
BackgroundMethodLabel Dependent Deep Variational Paraphrase GenerationPreprint
Shakeri, S., & Sethy, A. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1911.11952 · Cited by 6
CIS-Net: A Novel CNN Model for Spatial Image Steganalysis via Cover Image Suppression
Wu, S., Zhong, S.-H., Liu, Y., & Liu, M. (2019) · arXiv (Cornell University) · doi:10.48550/arxiv.1912.06540 · Cited by 3
“To handle the difficulty, we use curriculum learning [48] to detect steganopgraphic algorithms at low payloads.”
MethodExploring the Memorization-Generalization Continuum in Deep LearningPreprint
Jiang, Z., Zhang, C., Talwar, K., & Mozer, M. C. (2020) · arXiv (Cornell University) · Cited by 9
Sequential Cooperative Bayesian InferencePreprint
Wang, J., Wang, P., & Shafto, P. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2002.05706 · Cited by 8
“The literature on curriculum learning considers sequential interactions with a learner by a teacher in which the teacher presents data in an ordered sequence (Bengio et al., 2009), and traces back to various literatures on human and animal learning (Skinner, 1958; Elman, 1993).”
BackgroundZhou, W., & Xu, K. (2020) · AAAI Conference on Artificial Intelligence · doi:10.48550/arxiv.2002.05058 · Cited by 53
“In addition, motivated by the fact that the larger the margin between the quality two selected version of the model, the easier for the comparative evaluator to learn to distinguish the training examples, we propose to use curriculum learning (Bengio et al. 2009) by feeding the comparative evaluator with sample pairs…”
BackgroundLearning About Objects by Learning to Interact with ThemPreprint
Lohmann, M., Salvador, J., Kembhavi, A., & Mottaghi, R. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2006.09306 · Cited by 22
“Sampling from the memory bank combines intuitions from prioritized replay methods [52] (where high-loss, difficult, images have higher probability to be sampled) and self-paced learning [32, 5] (where low-loss, easy images have higher probability).”
BackgroundMollifying NetworksPreprint
Gülçehre, Ç., Moczulski, M., Visin, F., & Bengio, Y. (2016) · arXiv (Cornell University) · doi:10.48550/arxiv.1608.04980 · Cited by 41
“A number of approaches were proposed to alleviate the difficulty of optimization: addressing the problem of the internal covariate shift with Batch Normalization (Ioffe and Szegedy, 2015), learning with a curriculum (Bengio et al., 2009) and recently training with diffusion (Mobahi, 2016) - a form of continuation meth…”
BackgroundMethodInfluentialSelf-citationCurriculum By SmoothingPreprint
Sinha, S., Garg, A., & Larochelle, H. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2003.01367 · Cited by 65
“Curriculum learning was originally defined by Bengio et al. (2009) as a way to train networks by organizing the order in which tasks are learned and incrementally increasing the difficulty of a task, as opposed to regular learning where all tasks are learned at the same time.”
MethodA Distilled Model for Tracking and Tracker Fusion.Preprint
Dunnhofer, M., Martinel, N., & Micheloni, C. (2020) · arXiv (Cornell University) · Cited by 4
Chen, Z., Wang, Y., Lin, D. L., et al. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2008.07032 · Cited by 4
“In curriculum learning [1], prediction variation can be used as a way for estimating example difficulty.”
MethodTowards Automated Anamnesis Summarization: BERT-based Models for Symptom ExtractionPreprint
Schäfer, A., Blach, N., Rausch, O., Warm, M., & Krüger, N. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.01696 · Cited by 6
“ount of possible negative samples, we randomly select one negative sample for each positive sample in a post. Curriculum Learning. To gain further performance, we employ a curriculum learning scheme (Bengio et al. 2009). The hierarchical nature of the symptom collection allows us to form negative samples which are gra…”
MethodTeaching with CommentariesPreprint
Raghu, A., Raghu, M., Kornblith, S., Duvenaud, D., & Hinton, G. E. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.03037 · Cited by 12
“[3], Zhu [35].”
Efficient, Simple and Automated Negative Sampling for Knowledge Graph EmbeddingPreprint
Zhang, Y., Yao, Q., & Chen, L. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2010.14227 · Cited by 4
“The main idea of self-paced learning (or curriculum learning) [3, 31] is to pick up easy samples first, and then”
BackgroundLow-Bandwidth Communication Emerges Naturally in Multi-Agent Learning SystemsPreprint
Grupen, N. A., Lee, D. D., & Selman, B. (2020) · arXiv (Cornell University) · doi:10.48550/arxiv.2011.14890 · Cited by 13
3D Human Pose, Shape and Texture from Low-Resolution Images and VideosPreprint
Xu, X., Chen, H., Moreno-Noguer, F., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2103.06498 · Cited by 4
“However, [53] does not explicitly consider multi-resolution training.”
BackgroundPreview, Attend and Review: Schema-Aware Curriculum Learning for Multi-Domain Dialog State TrackingPreprint
Dai, Y., Li, H., Li, Y., Sun, J. Q., Huang, F., Si, L., & Zhu, X. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.00291 · Cited by 15
“This paradigm is called curriculum learning (CL) (Bengio et al., 2009) and has been shown useful in various other problems (Wang et al.”
BackgroundMethodGradual Domain Adaptation in the Wild:When Intermediate Distributions are AbsentPreprint
Abnar, S., van den Berg, R., Ghiasi, G., Dehghani, M., Kalchbrenner, N., & Sedghi, H. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.06080 · Cited by 33
“[Chang et al., 2019] , 1993, Sanger, 1994, Bengio et al., 2009] has led to better performance in terms of generalization and/or convergence speed in many domains such as computer vision [Pentina et al., 2015, Sarafianos et al., 2017, Guo et al., 2018, Wang et al., 2019b], natural language processing…”
BackgroundMethodInfluentialImproving Human Motion Prediction Through Continual LearningPreprint
Yasar, M. S., & Iqbal, T. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2107.00544 · Cited by 13
“Our choice of continual learning scheme is the curriculum learning setup [3], whereby we first train the network on a comparatively simpler task of representation learning, followed by a relatively dif-”
MethodGao, Z., Shi, J., Zhang, X., et al. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2106.10641 · Cited by 4
“This insight is similar to curriculum learning [11] which has been proven to be effective in histopathology image [12].”
BackgroundAutoSoC: Automating Algorithm-SOC Co-design for Aerial RobotsPreprint
Krishnan, S., Tambe, T., Wan, Z., & Reddi, V. J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2109.05683 · Cited by 9
“This trend is expected because, during the training phase, the model is trained using curriculum learning (Bengio et al., 2009).”
MethodData Quality Matters For Adversarial Training: An Empirical StudyPreprint
Dong, C., Liu, L., & Shang, J. (2021) · arXiv (Cornell University) · doi:10.48550/arxiv.2102.07437 · Cited by 15
“…proposes to dynamically estimate instance hardness during training and encourage the model to focus on those hard examples from a curriculum learning (Bengio et al., 2009) perspective, which can improve both the performance and efficiency for a wide range of datasets and neural architectures.”
BackgroundGan-Based Adaptive Cost Learning for Enhanced Image Steganography SecurityPreprint
Wang, D., Yang, G., Chen, J., et al. (2023) · SSRN Electronic Journal · doi:10.2139/ssrn.4673715 · Cited by 4
Let’s learn step by step: Enhancing in-context learning ability with curriculum learning
Liu, Y., Liu, J., Shi, X., Cheng, Q., Huang, Y., & Lu, W. (2024) · arXiv · doi:10.48550/arxiv.2402.10738 · Cited by 4
“The concept of curriculum learning (Bengio et al., 2009) has inspired numerous research to address various natural language processing tasks.”
BackgroundExternal Data Extraction Attacks Against Retrieval-Augmented Large Language Models
He, Y., Chen, Y., Li, Y., Shao, S., Qi, L., Li, B., Tao, D., & Qin, Z. (2026) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2026.3705326 · Cited by 4
“Inspired by curriculum learning [58], we first optimize p e ∗ against a weaker baseline model ( e . g ., Gemini 2.0 Flash [59]) where non-zero scores are more likely.”
InfluentialGeneralization on the Unseen, Logic Reasoning and Degree Curriculum
Abbe, E., Bengio, S., Lotfi, A., & Rizk, K. (2023) · arXiv · doi:10.48550/arxiv.2301.13105 · Cited by 4
“The bias of neural networks towards min-degree solutions can also be utilized to boost the learning via a curriculum learning (Bengio et al., 2009) algorithm.”
BackgroundFMDL: Enhancing Open-World Object Detection with foundation models and dynamic learning
Huang, Y., Hu, J., & Luo, R. (2025) · Expert Systems with Applications · doi:10.1016/j.eswa.2025.127050 · Cited by 4
One model, two skills: active vision and action learning model for robotic manipulation
Wang, G., Liu, Y., & Liu, H. (2025) · Science China Information Sciences · doi:10.1007/s11432-023-4282-8 · Cited by 7
Ad2Mix: Adversarial and Adaptive Mixup for Unsupervised Domain Adaptation
Zhu, L., Xu, Y., Liu, Y., Goh, R. S. M., & Xu, X. (2025) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv61041.2025.00641 · Cited by 3
“At a deeper level, the intuition entails a curriculum learning [1] strategy which can potentially tackle the large domain gap problem in domain adaptation.”
Sun, C., Wu, X. L., Su, Y., Shi, X., & Sun, C. (2025) · IEEE Transactions on Neural Networks and Learning Systems · doi:10.1109/tnnls.2025.3556438 · Cited by 5
“The first stage is to implement the curriculum training paradigm (CTP) [42] in the 2-”
Risk‐Aware Pedestrian Behavior Using Reinforcement Learning in Mixed Traffic
Cai, C., Wong, S., & Chen, T. (2025) · Computer Animation and Virtual Worlds · doi:10.1002/cav.70031 · Cited by 3
“We use curriculum learning [29–31] to train agents, balancing penalties to provide continuous, informative feedback.”
Silent Drones: A Deep Learning Approach to Suppress Drone Propeller Noise
Rizvi, S. W. F., Ahmed, F. A., Qassmi, N., & Al-Ali, A. K. (2025) · International Conference on Unmanned Aircraft Systems (ICUAS) · doi:10.1109/icuas65942.2025.11007904 · Cited by 4
“Training employs curriculum learning with exponentially growing sequence lengths (0.5 s → 80 s) and gradient clipping (maxnorm = 0.5) to stabilize long-form generation [40].”
InfluentialUnderstanding Model Weaknesses: A Path to Strengthening DNN-Based Android Malware Detection
Li, H., Cheng, X. Y., Zhao, Y., Xu, G., Xu, G., & Wang, H. (2025) · Proceedings of the ACM on software engineering. · doi:10.1145/3728884 · Cited by 5
Enhancing LLMs for Sequential Recommendation With Reversed User History and User Embeddings
Choi, Y. J., Yun, W.-S., & Cho, Y.-S. (2025) · IEEE Access · doi:10.1109/access.2025.3583094 · Cited by 3
“Furthermore, a dynamic learning scheme, curriculum learning [35], is adopted, which progressively guides the model through easy and hard tasks, helping it better capture the complexities of SR.”
NExT-Search: Rebuilding User Feedback Ecosystem for Generative AI Search
Dai, S., Wang, W., Pang, L., Xu, J., Ng, S.-K., Wen, J.-R., & Chua, T. (2025) · International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR) · doi:10.1145/3726302.3730353 · Cited by 7
“Thus, exploring adaptive learning techniques such as multi-task learning [30] or curriculum learning [5] may offer promising avenues for training on different feedback types.”
Liang, Z., Bao, S., Zhang, W., Wang, H., Yan, H., Dai, J., & Xiao, P. (2025) · Remote Sensing · doi:10.3390/rs17152735 · Cited by 3
Hallucination in Large Language Models: From Mechanistic Understanding to Novel Control Frameworks
Zhang, W., Zhang, C., Gu, C., Kou, J., Yuan, H., Fang, X., Duan, X., & Fang, Y. (2024) · International Conference on Universal Village (UV) · doi:10.1109/uv63228.2024.11189226 · Cited by 60
“…Several additional training factors contribute to hallucination through mechanisms that are still being characterized: Curriculum effects : The order and composition of training data can create path dependent biases that influence hallucination patterns in ways not yet fully understood [91].”
Ruiz-Gonzalez, U., Andrés, A., & Ser, J. D. (2025) · Machine Learning and Knowledge Extraction · doi:10.3390/make7040126 · Cited by 4
“Originally conceived as “training from easy to hard”, CL has attracted attention due to its potential to expedite learning processes [11,13].”
Zhang, Z., Bao, X., Jiang, L., Luo, X., Zhang, Z., Yin, J., Zhao, M., Wang, Y., Comai, A., Waldhaus, J., et al. (2025) · Nucleic Acids Research · doi:10.1093/nar/gkaf1269 · Cited by 9
“For task scheduling, we implemented a curriculum learning approach [ 23 ], gradually introducing different genomic modalities into the training based on a pre-defined curriculum ( Supplementary Fig.”
Haddout, Y., & Haddout, S. (2025) · Fluids · doi:10.3390/fluids10120322 · Cited by 3
“Unlike prior studies [23,24] where standard PINNs fail to converge for | R i | > 0 (exhibiting errors above 28%), our approach utilizes a deep residual network combined with curriculum learning and adaptive collocation to mitigate this severe stiffness and achieve machine-precision convergence…”
MILO: A Lightweight Perceptual Quality Metric for Image and Latent-Space Optimization
Çoğalan, U., Bemana, M., Myszkowski, K., Seidel, H., & Groth, C. (2025) · ACM Transactions on Graphics · doi:10.1145/3763340 · Cited by 5
“This enables a curriculum learning strategy [Bengio et al. 2009] in which the network first learns to restore less perceptually critical areas before focusing on regions where distortions are more visible.”
Cetinkaya, H., Ay, F., Tunçel, M., Nounou, H. N., Nounou, M. N., Kurban, H., & Serpedin, E. (2025) · Mathematics · doi:10.3390/math13243996 · Cited by 5
Physics-informed reinforcement learning for air-to-air F16 target aiming
Chen, Y.-H., Chiang, M.-H., & Peng, C.-C. (2025) · Advanced Engineering Informatics · doi:10.1016/j.aei.2025.103988 · Cited by 4
Robust Weakly Supervised Bearing Fault Diagnosis via Dual Learning and Curriculum Strategies
Ding, Y., Miao, Q., Zhao, X., Yan, X., & Lee, C.-G. (2025) · IEEE Transactions on Instrumentation and Measurement · doi:10.1109/tim.2025.3645912 · Cited by 3
“CL [29], a type of continuation method, learns a model by gradually including samples into training from easy to complex, mimicking the progressive learning process observed in human education [30], [31].”
A self-supervised hybrid CNN with uncertainty-aware referral for diabetic retinopathy screening
Kumar, N. A., Madhusudan, D., Ioannou, I. I., Ghantasala, G. P., & Vassiliou, V. (2026) · Biomedical Signal Processing and Control · doi:10.1016/j.bspc.2026.109482 · Cited by 4
Achari, S., Kumar, S., & Beniwal, R. (2026) · IEEE Transactions on Network Science and Engineering · doi:10.1109/tnse.2026.3654671 · Cited by 4
“Curriculum learning, proposed by Bengio et al. [59] as a training strategy that trains machine learning models from easier samples to harder samples.”
Zhang, Z., Zhong, Q., Pan, W., Ai, Y., & Wang, Q. (2026) · Aerospace · doi:10.3390/aerospace13010108 · Cited by 3
“Our work bridges this gap by integrating meta-learning with curriculum learning [47,48] to progressively expose the model to scheduling problems of increasing complexity and structural diversity.”
Spatial-Temporal Transformer with Curriculum Learning for EEG-Based Emotion Recognition
Lin, X., Peng, T., Dai, P., Liang, Y., & Wu, W. (2025) · Conference proceedings/Conference proceedings - IEEE International Conference on Systems, Man, and Cybernetics · doi:10.1109/smc58881.2025.11343476 · Cited by 8
“Curriculum learning [31] provides a systematic solution that plan the training process of models in various domains [32]–[35].”
CLoE: Curriculum Learning on Endoscopic Images for Robust MES Classification
Özdemir, Z. Ö., Keleş, H. Y., & Tanrıöver, Ö. Ö. (2026) · IEEE Access · doi:10.1109/access.2026.3666962 · Cited by 3
“These scores define a curriculum schedule that initially prioritizes high-quality, low-uncertainty samples and gradually incorporates more challenging, noisier cases, following the progressive sampling strategy of curriculum learning [16], [17].”
InfluentialWaveVerify: A Novel Audio Watermarking Framework for Media Authentication and Combatting Deepfakes
Pujari, A., & Rattani, A. (2025) · IEEE International Joint Conference on Biometrics (IJCB) · doi:10.1109/ijcb65343.2025.11411372 · Cited by 6
“Instead of employing fixed augmentation parameters and static selection probabilities, we introduce a Dynamic Effect Scheduler, inspired by curriculum learning principles [9] but specifically adapted for adversarial watermarking.”
Zhang, J., Wang, X., Wang, D., Zheng, Z., Wang, H., & La, M. (2026) · mSystems · doi:10.1128/msystems.00800-25 · Cited by 6
“…adoption of dynamic curriculum learning strategies that progressively increase data complexity (e.g., incremental learning from dominant to rare taxa) (162) can improve the ability of the model to fit long-tailed distributions and training efficiency (163) without changing the model architecture.”
Articulatory and spectrum features integration using generalized distillation framework
Yu, J., Markov, K., & Matsui, T. (2016) · Machine learning for signal processing ... · doi:10.1109/mlsp.2016.7738813 · Cited by 5
“Generalized Distillation is closely related to applications in methods such as semi-supervised learning, domain adaptation, transfer learning, Universum learning [17] and curriculum learning [18].”
MethodInformed Pair Selection for Self-paced Metric Learning in Siamese Neural Networks
Martin, K., Wiratunga, N., Massie, S., et al. (2018) · Lecture notes in computer science · doi:10.1007/978-3-030-04191-5_3 · Cited by 3
“The idea is that by ranking so that the network is initially exposed to simpler examples and then gradually introduced to more complex examples, the network will converge faster [1].”
BackgroundSadr, A. S., Khojasteh, M. S., Malek, H., et al. (2022) · International Conference on Computer and Knowledge Engineering · doi:10.1109/iccke57176.2022.9960115 · Cited by 3
“To avoid being stuck at local optima, gradual learning is applied [15], [16].”
Decanus to Legatus: Synthetic Training for 2D-3D Human Pose Lifting
Zhu, Y., & Picard, D. (2023) · Lecture notes in computer science · doi:10.1007/978-3-031-26316-3_16 · Cited by 3
Human Not in the Loop: Objective Sample Difficulty Measures for Curriculum Learning
Zhou, Z., Luo, J., Arefan, D., Kitamura, G., & Wu, S. (2023) · IEEE International Symposium on Biomedical Imaging (ISBI) · doi:10.1109/isbi53787.2023.10230597 · Cited by 3
“It has been shown that starting training with “easy” samples enables machine learning models to converge faster and shows a significant improvement in the generalization of the models [1].”
BackgroundMethodConfidence-Aware Contrastive Learning for Semantic Segmentation
Lv, L., Liu, Q., Kan, S., & Liang, Y. (2023) · ACM International Conference on Multimedia (ACM MM) · doi:10.1145/3581783.3612037 · Cited by 9
“Motivated by the observation that the harder samples gradually become easier as training progresses [28] and the idea of easy-to-hard in curriculum learning [3], here we propose a confidence-aware marginal anchor sampling (CAMAS) method to effectively select samples for calculation of contrastive…”
MethodSGD Biased towards Early Important Samples for Efficient Training
Quercia, A., Morrison, A., Scharr, H., & Assent, I. (2023) · Proceedings · doi:10.1109/icdm58522.2023.00163 · Cited by 4
“…an unbiased estimator [11, 12, 13] (or approximations [14, 15, 16, 17, 18, 19]), biasing training to hard samples [20, 21, 22, 23, 24, 25, 26, 27, 28], presenting samples through a curriculum [29, 30, 31, 32], or adaptively subsetting training data to approximate the full gradient [33, 34, 35, 36].”
BackgroundMethodA Unified Meta-Learning Framework for Fair Ranking With Curriculum Learning
Wang, Y., Tao, Z., & Fang, Y. (2024) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2024.3377644 · Cited by 7
“To this end, we adopt curriculum learning [26], a method that starts with easier, less biased samples and gradually introduces more complex ones.”
BackgroundMethodExploring the Potential of Pre-Trained Language Models of Code for Automated Program Repair
Hao, S., Shi, X., & Liu, H. (2024) · Electronics · doi:10.3390/electronics13071200 · Cited by 9
Nguyen, K.-B. (2024) · IEEE Winter Conference on Applications of Computer Vision · doi:10.1109/wacv57701.2024.00243 · Cited by 13
“FlexMatch [52] introduces a Curriculum Pseudo Labeling (CPL) method based on the Curriculum Learning (CL) [4].”
MethodDeep Image Clustering Based on Curriculum Learning and Density Information
Zheng, H., Zhang, R., & Wang, H. (2024) · International Conference on Multimedia Retrieval · doi:10.1145/3652583.3658081 · Cited by 3
“In view of this, Bengio et al. [2] propose a curriculum-based training strategy named Curriculum Learning (CL), which aims at training a model from easier data to harder data.”
RefreshNet: learning multiscale dynamics through hierarchical refreshing
Farooq, J., Rafiq, D., Vlachas, P. R., & Bazaz, M. A. (2024) · Nonlinear Dynamics · doi:10.1007/s11071-024-09813-3 · Cited by 5
“…the autoregressive propagation (e.g., Trans-formers [45]), to regularization techniques [46], and novel training methods [40–43 Another factor contributing to error accumulation in autoregressive methods is their training objective, which focuses on minimizing the loss in one-step-aheadpredictions.”
Curriculum adaptation method based on graph neural networks for universal domain adaptation
Fan, C., Liu, P., & Zhao, W. (2024) · Expert Systems with Applications · doi:10.1016/j.eswa.2024.124509 · Cited by 5
Optimizing Crowd Counting in Dense Environments Through Curriculum Learning Training Strategy
Fotia, L., Percannella, G., Saggese, A., & Vento, M. (2024) · SN Computer Science · doi:10.1007/s42979-024-03029-3 · Cited by 5
Dynamic Graph Representation Learning for Passenger Behavior Prediction
Xie, M., Zou, T., Ye, J., Du, B., & Huang, R. (2024) · Future Internet · doi:10.3390/fi16080295 · Cited by 3
“When calculating the loss for gradient descent to update weights [25–27], we can use all the samples to compute the loss; this method is called batch gradient descent (BGD).”
MethodLanzoni, D., Fantasia, A., Bergamaschini, R., Pierre-Louis, O., & Montalenti, F. (2024) · Machine Learning Science and Technology · doi:10.1088/2632-2153/ad8532 · Cited by 11
Dynamic Assessment-Based Curriculum Learning Method for Chinese Grammatical Error Correction
Duan, R., Ma, Z., Zhang, Y., et al. (2024) · Electronics · doi:10.3390/electronics13204079 · Cited by 3
Chang, J., Lee, J., Kwon, D., Lee, J.-H., Lee, M., Jeong, S., Kim, J., Jung, H., & Oh, C. (2024) · Mathematics · doi:10.3390/math12223613 · Cited by 3
Confidence-Based curricula for multi-agent path finding via reinforcement learning
Phan, T., Driscoll, J., Romberg, J. K., & Koenig, S. (2026) · Autonomous Agents and Multi-Agent Systems · doi:10.1007/s10458-026-09747-7 · Cited by 5
“Curriculum learning is a machine learning paradigm inspired by human learning to master complex tasks through stepwise solving of easier (sub-)tasks, which are sorted by difficulty [62, 63 ].”
Mutual Evidential Deep Learning for Semi-supervised Medical Image Segmentation
He, Y., Bi, Y., Li, L., Pun, C., Jiao, W., & Jin, Z. J. (2024) · IEEE International Conference on Bioinformatics and Biomedicine (BIBM) · doi:10.1109/bibm62325.2024.10822008 · Cited by 14
“Drawing inspiration from curriculum learning [22], the proposed approach enhances the learning of complex information, building on a firm grasp of simpler characteristics.”
SPrint: Self-Paced Continual Learning with Adaptive Curriculum and Memory Replay
Kim, M.-S., Liu, L., & Kwon, H.-Y. (2024) · IEEE International Conference on Big Data (BigData) · doi:10.1109/bigdata62323.2024.10825931 · Cited by 4
Self-supervised 3D medical image segmentation by flow-guided mask propagation learning
Bitarafan, A., Mozafari, M., Azampour, M. F., Baghshah, M. S., Navab, N., & Farshad, A. (2025) · Medical Image Analysis · doi:10.1016/j.media.2025.103478 · Cited by 7
Koo, H. (2025) · IEEE International Conference on Acoustics Speech and Signal Processing · doi:10.1109/icassp49660.2025.10890473 · Cited by 9
“Therefore, we pro-pose curriculum learning [30] guided random single-point data erasing [31] to both training and distillation of the teacher.”
Metadata
- Authors
- Bengio, Y., Louradour, J., Collobert, R., & Weston, J.
- Affiliations
- Université de Montréal (CA)
- NEC Laboratories America, Princeton, NJ
- ORCID
- Bengio, Yoshua
- Venue
- Proceedings of the 26th Annual International Conference on Machine Learning
- Publisher
- ACM
- Published
- 2009-06-14
- Pages
- 41-48
- OpenAlex
- W2296073425
- Semantic Scholar
- 8de174ab5419b9d3127695405efd079808e956e8
- Cited by
- 5,366
- References
- 40
- Citations by year
- 2026: 526 · 2025: 636 · 2024: 551 · 2023: 591 · 2022: 536 · 2021: 711 · 2020: 565 · 2019: 442 · 2018: 293 · 2017: 198 · 2016: 103 · 2015: 63 · 2014: 41 · 2013: 40 · 2012: 33
- FWCI
- 16.17
- Type
- conference paper
- Language
- en
- Open access
- closed
- Locations
- International Conference on Machine Learning (publishedVersion)
- Queen Mary Research Online (Queen Mary University of London) (submittedVersion)
- Funders
- Canadian Institute for Advanced Research
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Stochastic Gradient Optimization Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Reinforcement Learning in Robotics
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Keywords
- curriculum learning, training order, non-convex optimization, deep neural network, stochastic neural networks, deterministic neural networks, generalization, convergence rate, numerical continuation, global optimization
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
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.