Research map / Continual learning / Online and replay-based
Batch Model Consolidation: A Multi-Task Model Consolidation Framework
Iordanis Fostiropoulos, Jiaye Zhu, Laurent Itti
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 |
|---|---|---|---|
| Use BMC instead of DMC | Continual learning on Split-CIFAR-100 (10 tasks) · Split-CIFAR-100 (10 tasks) | mean accuracy: 36.2 -> 66.5 | Improved |
| Use BMC instead of GDumb | Continual learning on Split-CIFAR-100 (20 tasks) · Split-CIFAR-100 (20 tasks) | mean accuracy: 22.1 -> 67.4 | Improved |
| Use BMC instead of DER++ | Continual learning on Split-Tiny-ImageNet (10 tasks) · Split-Tiny-ImageNet (10 tasks) | mean accuracy: 39.0 -> 49.4 | Improved |
| Use BMC instead of ER | Continual learning on Stream · Stream (71 tasks) | mean accuracy: 41.4 -> 70.4 | Improved |
| Train tasks in parallel across experts and consolidate them in batches | BMC on Stream · Stream (71 tasks) | relative training time: 100% -> 78% | Improved |
| Increase the number of experts consolidated per step (ablation) | BMC · Permuted-MNIST (128 tasks) | mean accuracy | Improved |
| Replace intermediate-feature distillation with logit KD (ablation) | BMC consolidation loss · Permuted-MNIST (128 tasks) | mean accuracy | Hurt |
| Replace intermediate-feature distillation with penultimate-representation distillation (ablation) | BMC consolidation or stability loss · Permuted-MNIST (128 tasks) | mean accuracy | No clear effect |
| Replace intermediate-feature distillation with EWC (ablation) | BMC stability loss · Permuted-MNIST (128 tasks) | mean accuracy | Hurt |
| Use random sampling instead of gradient-based buffer sampling (ablation) | BMC buffer construction · Permuted-MNIST (128 tasks) | mean accuracy | Improved |
| Increase the stability coefficient (ablation) | BMC · Permuted-MNIST (128 tasks) | mean accuracy: 4.12e-2 | Improved |
| Increase the consolidation coefficient (ablation) | BMC · Permuted-MNIST (128 tasks) | mean accuracy: 9.39e-2 | 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.
- GDumb — GDumb: A Simple Approach that Questions Our Progress in Continual Learning (ECCV 2020) · better on 1 of 1 shared result
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
- Train with Perturbation, Infer after Merging: A Two-Stage Framework for Continual Learning (NeurIPS 2025 · similarity 0.89)
- Enhancing Visual Continual Learning with Language-Guided Supervision (CVPR 2024 · similarity 0.88)
- Model Sensitivity Aware Continual Learning (NeurIPS 2024 · similarity 0.88)
- Merging on the Fly Without Retraining: A Sequential Approach to Scalable Continual Model Merging (NeurIPS 2025 · similarity 0.88)
- LoRanPAC: Low-rank Random Features and Pre-trained Models for Bridging Theory and Practice in Continual Learning (ICLR 2025 · similarity 0.88)
- BECAME: Bayesian Continual Learning with Adaptive Model Merging (ICML 2025 · similarity 0.88)
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.
