Research map / Continual learning / Online and replay-based
Calibrating CNNs for Lifelong Learning
Pravendra Singh, Vinay Kumar Verma, Pratik Mazumder, Lawrence Carin, Piyush Rai
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add spatial calibration module (ablation) | CCLL<1,1> · Split CIFAR-10/100 | A6 accuracy: 87.85 -> 89.58 | Improved |
| add channel-wise calibration module (ablation) | CCLL<1,1> · Split CIFAR-10/100 | A6 accuracy: 89.06 -> 89.58 | Improved |
| add both calibration modules (SCM and CCM) (ablation) | CCLL base network · Split CIFAR-10/100 | A6 accuracy: 67.06 -> 89.58 | Improved |
| increase alpha (channels per group in SCM) from 1 to 8 (ablation) | CCLL · Split CIFAR-10/100 | A6 accuracy: 89.58 -> 90.99 | Improved |
| train base module on first task instead of leaving it untrained (freeze random base) (ablation) | CCLL · Split CIFAR-10/100 | accuracy | Improved |
| initialize calibration parameters from previous task instead of from scratch (ablation) | CCLL · Split CIFAR-10/100 | accuracy | Improved |
| add batch normalization layers to LeNet (ablation) | CCLL<1,1> · CIFAR-100 10 classes per task | accuracy: 63.71% -> 71.52% | Improved |
| pre-train base module on CIFAR-10 and keep frozen instead of training on first SVHN task (ablation) | CCLL · SVHN (5 tasks) | A5 accuracy: 98.2% -> 97.9% | No clear effect |
| finetune CIFAR-10 pretrained base on first SVHN task (ablation) | CCLL · SVHN (5 tasks) | A5 accuracy: 98.2% -> 98.4% | No clear effect |
| use CCLL instead of the closest replay-based baseline | ResNet-18 continual learner · ImageNet-100/10 | final top-5 accuracy: 74.1 -> 97.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Mitigating Forgetting in Online Continual Learning with Neuron Calibration (NeurIPS 2021 · similarity 0.88)
- CLR: Channel-wise Lightweight Reprogramming for Continual Learning (ICCV 2023 · similarity 0.88)
- An Effective Dynamic Gradient Calibration Method for Continual Learning (ICML 2024 · similarity 0.87)
- PaLoRA: Paced Low-Rank Adaptation for Continual Learning (NeurIPS 2026 · similarity 0.86)
- Adaptive Retention & Correction: Test-Time Training for Continual Learning (ICLR 2025 · similarity 0.86)
- A Benchmark Study on Calibration (ICLR 2024 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
