IDER: IDempotent Experience Replay for Reliable Continual Learning

Zhanwang Liu; Yuting Li; Haoyuan Gao; Yexin Li; Linghe Kong; Lichao Sun; Weiran Huang

cs.LGFeb 28, 2026

IDER: IDempotent Experience Replay for Reliable Continual Learning

Authors: Zhanwang Liu, Yuting Li, Haoyuan Gao, Yexin Li, Linghe Kong, Lichao Sun, Weiran Huang

Organizations: School of Computer Science, Shanghai Jiao Tong University · State Key Laboratory of General Artificial Intelligence, BIGAI · Lehigh University · Shanghai Innovation Institute

Abstract

Catastrophic forgetting, the tendency of neural networks to forget previously learned knowledge when learning new tasks, has been a major challenge in continual learning (CL). To tackle this challenge, CL methods have been proposed and shown to reduce forgetting. Furthermore, CL models deployed in mission-critical settings can benefit from uncertainty awareness by calibrating their predictions to reliably assess their confidences. However, existing uncertainty-aware continual learning methods suffer from high computational overhead and incompatibility with mainstream replay methods. To address this, we propose idempotent experience replay (IDER), a novel approach based on the idempotent property where repeated function applications yield the same output. Specifically, we first adapt the training loss to make model idempotent on current data streams. In addition, we introduce an idempotence distillation loss. We feed the output of the current model back into the old checkpoint and then minimize the distance between this reprocessed output and the original output of the current model. This yields a simple and effective new baseline for building reliable continual learners, which can be seamlessly integrated with other CL approaches. Extensive experiments on different CL benchmarks demonstrate that IDER consistently improves prediction reliability while simultaneously boosting accuracy and reducing forgetting. Our results suggest the potential of idempotence as a promising principle for deploying efficient and trustworthy continual learning systems in real-world applications.Our code is available at https://github.com/YutingLi0606/Idempotent-Continual-Learning.

Figures & tables

Figure 1: We propose the IDER method, which can be directly applied to many recent rehearsal-based continual learning methods, resulting in less calibration error and significant improvements in FAA with less parameter growth compared with NPCL.

Figure 2: Modified Architecture. We modify the architecture of backbone(ResNet) and enable the model to accept two inputs.

Figure 5: Overall framework of Idempotent Experience Replay (IDER). Our method consists of two modules for continual learning: (1) Standard Idempotent Module that trains current model idempotent with data from the current task. (2) Idempotent Distillation Module that enforce the current model to become idempotent with respect to the last task model checkpoint, utilizing data from both the current task and buffer memory. IDER can be integrated into existing CL approaches to make reliable predictions while mitigate catastrophic forgetting.

Method	CIFAR-10		CIFAR-100		Tiny-ImageNet
Method	Buffer 200	Buffer 500	Buffer 500	Buffer 2000	Buffer 500	Buffer 4000
Joint (upper bound)	91.93±0.29		71.15±0.51		59.52±0.33
iCaRL ?	58.37±3.51	62.49±5.42	46.81±0.41	52.51±0.44	22.53±0.62	26.38±0.23
ER ?	44.46±2.87	58.84±3.85	23.41±1.15	40.47±0.95	10.13±0.39	25.12±0.56
BiC ?	52.61±5.37	71.95±1.82	37.82±1.67	47.17±1.17	15.36±1.31	18.67±0.57
LUCIR ?	49.18±7.61	65.26±2.54	37.91±1.18	50.42±0.76	28.79±0.51	31.64±0.51

Table 1: Comparison of Final Average Accuracy (FAA) across different continual learning methods. All experiments are repeated 5 times with different seeds. Results for SARL ? are from our implementation. The best results are highlighted in blue .The second best results are highlighted in green .

Panel of Figure 6 — Figure 6: Results on CIFAR-10 and Tiny-ImageNet with different buffer size. It shows the trend of the average test-set accuracy on the observed tasks.

Method	Uniform				Longtail
Method	Buffer 200	$\Delta$	Buffer 500	$\Delta$	Buffer 200	$\Delta$	Buffer 500	$\Delta$
Joint (upper bound)	58.36±1.02				56.94±1.56
DER++ ?	19.36±0.65		33.66±0.96		27.05±1.11		25.98±0.81
SCoMMER ?	28.56±2.26		35.70±0.86		28.47±1.12		32.99±0.49
ER ?	16.34±0.74		28.76±0.66		19.55±0.69		20.02±1.05
Ours (ER+ID)	26.66±0.63	+10.32	40.54±0.46	+11.78	30.04±0.58	+10.49	35.92±0.35	+15.90

Table 2: Comparison of Final Average Accuracy (FAA) across different continual learning methods on GCIL-CIFAR-100 dataset. All experiments are repeated 5 times with different seeds. Absolute gains are indicated in green .

Method	CIFAR-10				CIFAR-100				Tiny-ImageNet
Method	Buffer 200	$\Delta$	Buffer 500	$\Delta$	Buffer 500	$\Delta$	Buffer 2000	$\Delta$	Buffer 500	$\Delta$	Buffer 4000	$\Delta$
DER ?	29.91		16.20		24.84		10.79		22.80		10.52
NPCL ?	21.03		-		19.95		-		-		-
RC ?	16.39		12.84		19.43		19.31		21.32		16.49
T-CIL ?	22.50		10.51		15.79		8.67		14.50		10.30
ER ?	45.53		32.69		64.59		45.64		67.50		51.37

Table 3: Comparison of Expected Calibration Error (ECE) across different continual learning methods on CIFAR-10 and CIFAR-100 dataset. All experiments are repeated 5 times with different seeds. Results of NPCL are imported from its original work ? . Absolute improvements (lower ECE) are indicated in red .

Table 9

Panel of Figure 7 — Figure 7: Results for model analysis. (a) the training time of different methods on Split TinyImageNet with buffer 500. (b) the Final Forgetting (FF) measures on Split CIFAR-100 with different buffer sizes. (c) the performances on Split CIFAR-100 using different distance metrics for idempotent distillation loss.

Figure 8: Visualizations of the predictions on Split CIFAR-100. It shows top-5 prediction probability produced by ER and ER+ID (ours) after training the first two tasks. The first 10 classes (class ids from 0 to 9) belong to task 1 and the next 10 classes (class ids from 10 to 19) belong to task 2. The ground-truth class is highlighted with red boxes.

Platform	CIFAR-10		CIFAR-100		Tiny-ImageNet
Platform	Buffer 200	Buffer 500	Buffer 500	Buffer 2000	Buffer 4000
RTX 4090	71.02±1.98	74.74±0.42	44.82±0.85	56.59±0.35	43.05±1.40
Ascend 910B	71.13±0.92	74.68±0.37	44.87±0.76	56.44±0.41	43.01±1.26

Table 6: Performance comparison for ER+ID between NVIDIA RTX 4090 and Huawei Ascend 910B. All experiments are repeated 5 times with different seeds.

Appendix figures & tables5 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 9: We perform $t$ -SNE visualization of the features extracted from the first task testing data on CIFAR-100 across training tasks. The figures show how the feature clusters of the 10 classes from the first task change when the model train data from new tasks.

Method	CIFAR-10		CIFAR-100		Tiny-ImageNet
Method	Buffer 200	Buffer 500	Buffer 500	Buffer 2000	Buffer 500	Buffer 4000
Joint	0.00	0.00	0.00	0.00	0.00	0.00
iCaRL ?	36.00±7.29	30.19±3.38	28.58±0.84	24.24±0.58	20.82±1.31	15.29±0.53
ER ?	71.35±7.77	52.12±7.56	71.92±0.74	51.82±0.75	74.79±0.67	57.47±0.57
BiC ?	53.63±7.18	24.87±0.98	48.87±0.91	38.50±1.09	67.57±1.98	63.48±0.28
LUCIR ?	59.79±10.16	37.58±3.80	50.22±1.26	32.48±0.76	35.02±0.56	30.59±0.95

Appendix

Table 7: Comparison of Final Forgetting (FF) across different continual learning methods. All experiments are repeated 5 times with different seeds. The best results (lowest forgetting) are highlighted in blue . The second best results are highlighted in green .

SIM	IDM	ER	FAA
✓	✗	✗	8.23
✓	✗	✓	24.73
✓	✓	✓	44.82

Appendix

Table 8: Ablation study of different components on Split CIFAR-100.

beta	alpha=0.1	alpha=0.2	alpha=0.5	alpha=1
0.1	44.77	44.24	41.74	40.42
0.2	44.41	44.43	44.25	43.02
0.5	43.82	44.33	44.82	43.87
1	41.20	41.58	42.72	40.93

Appendix

Table 9: Impact of hyperparameters

\alpha

and

\beta

for ER+ID on Final Average Accuracy (FAA) on Split CIFAR-100. We use learning rate as 0.03 and training epoch as 50.

P	0.2	0.3	0.4	0.5	0.6	0.7	0.8	0.9	1
FAA	20.9	22.24	27.03	29.19	31.46	37.48	42.24	44.82	43.26

Appendix

Table 10: Impact of the probability

P

on Final Average Accuracy (FAA) on Split CIFAR-100.

Continual Learning

Continual Learning studies how models acquire knowledge from changing data streams while retaining or selectively revising what they learned before.

See all 305 papers

Realistic Continual Learning Approach using Pre-trained Models
Apr 11, 2024Nadia Nasri, Carlos Gutiérrez-Álvarez, Sergio Lafuente-Arroyo +2Class-Incremental Learning Continual Learning
University of Alcala, Department of Signal Theory and Communications, GRAM Research group, Alcala de Henares, 28805, Spain · Univesity of Alicante, Department of Computer Science and Artificial Intelligence, San Vicente del Raspeig, 03690, Spain
Leveraging Complementary Embeddings for Replay Selection in Continual Learning with Small Buffers
Apr 9, 2026Danit Yanowsky, Daphna WeinshallExperience Replay Replay-Based Continual Learning
School of Computer Science and Engineering The Hebrew University of Jerusalem
CARES: Coverage-Aware Replay Exemplar Selection for Continual Learning
Oct 23, 2026Arundhati S. Shanbhag, Brian B. Moser, Federico Raue +2Replay-Based Continual Learning Experience Replay
University of Kaiserslautern-Landau, Germany · German Research Center for Artificial Intelligence, Germany

Apr 11, 2024J/K move · Enter open · S save

Realistic Continual Learning Approach using Pre-trained Models

Nadia Nasri, Carlos Gutiérrez-Álvarez, Sergio Lafuente-Arroyo, Saturnino Maldonado-Bascón+1

University of Alcala, Department of Signal Theory and Communications, GRAM Research group, Alcala de Henares, 28805, Spain · Univesity of Alicante, Department of Computer Science and Artificial Intelligence, San Vicente del Raspeig, 03690, Spain

Class-Incremental Learning Continual Learning

Open paper AskPDF arXiv

Continual learning (CL) evaluates adaptability in learning solutions to retain knowledge. Our research addresses the challenge of catastrophic forgetting, where models lose proficiency in previously learned tasks as they acquire new ones. While numerous solutions have been proposed, existing experimental setups often rely on idealized class-incremental learning scenarios. We introduce Realistic Continual Learning (RealCL), a novel CL paradigm where class distributions across tasks are random. We also present CLARE (Continual Learning Approach with pRE-trained models for RealCL scenarios), a pre-trained model-based solution designed to integrate new knowledge while preserving past learning. Our contributions include pioneering RealCL as a generalization of traditional CL setups, proposing CLARE as an adaptable approach for RealCL tasks, and conducting extensive experiments demonstrating its effectiveness across various RealCL scenarios. Notably, CLARE outperforms existing models on RealCL benchmarks, highlighting its versatility in unpredictable learning environments. Code to reproduce all our experiments can be found at https://github.com/gramuah/clare.

Abstract

Continual Learning

Explore similar work

Realistic Continual Learning Approach using Pre-trained Models

Leveraging Complementary Embeddings for Replay Selection in Continual Learning with Small Buffers

CARES: Coverage-Aware Replay Exemplar Selection for Continual Learning