Research map / Vision-language and multimodal / Human and embodied modeling
Learning More Universal Representations for Transfer-Learning
Youssef Tamaazousti, Hervé Le Borgne, Céline Hudelot, Mohamed El Amine Seddik, Mohamed Tamaazousti
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 FSFT to MulDiP-Net (ablation) | MulDiP-Net · 8 target datasets (ILSVRC*) | mNRG: +7.9 -> +10.7 | Improved |
| use FSFT instead of SFT (different learning rates for first and last layers) (ablation) | self fine-tuning · 8 target datasets (ILSVRC*) | mNRG: -0.1 -> +4.5 | Improved |
| use FSFT instead of frozen self-training FrST (ablation) | self-training · 8 target datasets (ILSVRC*) | mNRG: -11.6 -> +4.5 | Improved |
| use FSFT as dimensionality reduction (DR) rather than full FSFT (ablation) | FSFT · 8 target datasets (ILSVRC*) | mNRG: +4.5 -> +3.4 | Hurt |
| use FSFT as dimensionality reduction within MulDiP (ablation) | MulDiP+FSFT · 8 target datasets (ILSVRC*) | mNRG: +10.7 -> +8.8 | Hurt |
| MulDiP-Net instead of ensemble of networks | MulDiP-Net · 8 target datasets (ILSVRC*) | mNRG: +2.3 -> +7.9 | Improved |
| MulDiP-Net instead of multi-task learning | MulDiP-Net · 8 target datasets (ILSVRC*) | mNRG: -16.2 -> +7.9 | Improved |
| MulDiP-Net instead of multi-label learning | MulDiP-Net · 8 target datasets (ILSVRC*) | mNRG: -45.0 -> +7.9 | Improved |
| MulDiP-Net instead of recursive learning | MulDiP-Net · 8 target datasets (ILSVRC*) | mNRG: -4.8 -> +7.9 | Improved |
| MulDiP+FSFT instead of MuCaLe-Net | MulDiP+FSFT · 10 target datasets | mNRG: +7.7 -> +9.8 | Improved |
| add MulDiP-Net to Net-S with VGG-16 (ablation) | Net-S VGG-16 · 8 target datasets, full ILSVRC | mNRG: +44.8 -> +55.3 | Improved |
| add MulDiP-Net to Net-S with DarkNet-20 (ablation) | Net-S DarkNet-20 · 8 target datasets, full ILSVRC | mNRG: +38.9 -> +47.1 | Improved |
Closest papers by meaning
- Universal Representations: A Unified Look at Multiple Task and Domain Learning (IJCV 2023 · similarity 0.89)
- Characterizing Universal Object Representations Across Vision Models (NeurIPS 2026 · similarity 0.88)
- Towards Universal Image Embeddings: A Large-Scale Dataset and Challenge for Generic Image Representations (ICCV 2023 · similarity 0.88)
- X-Learner: Learning Cross Sources and Tasks for Universal Visual Representation (ECCV 2022 · similarity 0.87)
- Visual Bridge: Universal Visual Perception Representations Generating (AAAI 2026 · similarity 0.87)
- Universal Sparse Autoencoders: Interpretable Cross-Model Concept Alignment (ICML 2025 · similarity 0.85)
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.
