Research map / Label-efficient and robust learning / Self-supervised and contrastive
Augmenting Supervised Neural Networks with Unsupervised Objectives for Large-scale Image Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 an auxiliary reconstructive decoding pathway with known pooling switches and reconstruction losses at all macro-layers | VGGNet · ImageNet ILSVRC-2012 validation, single-crop | Top-1 error: 29.05 -> 27.39 | Helped |
| Add an auxiliary reconstructive decoding pathway with known pooling switches and reconstruction losses at all macro-layers | VGGNet · ImageNet ILSVRC-2012 validation, convolution | Top-1 error: 26.97 -> 25.79 | Helped |
| Add an auxiliary reconstructive decoding pathway with known pooling switches and reconstruction losses at all macro-layers | VGGNet · ImageNet ILSVRC-2012 validation, single-crop | Top-5 error: 10.07 -> 9.06 | Helped |
| Add an auxiliary reconstructive decoding pathway with known pooling switches and reconstruction losses at all macro-layers | VGGNet · ImageNet ILSVRC-2012 validation, convolution | Top-5 error: 8.94 -> 8.13 | Helped |
| Add known pooling-switch unpooling instead of fixed-switch unpooling (ablation) | SWWAE-all versus SAE-all · ImageNet ILSVRC-2012 validation, single-crop | Top-1 error: 27.54 -> 27.39 | Helped |
| Add known pooling-switch unpooling instead of fixed-switch unpooling (ablation) | SWWAE-all versus SAE-all · ImageNet ILSVRC-2012 validation, convolution | Top-1 error: 26.10 -> 25.79 | Helped |
| Use reconstruction losses at all macro-layers rather than only at the image input (ablation) | SWWAE · ImageNet ILSVRC-2012 validation, single-crop | Top-1 error: 27.60 -> 27.39 | Helped |
| Use reconstruction losses at all macro-layers rather than only at the image input (ablation) | SWWAE · ImageNet ILSVRC-2012 validation, convolution | Top-1 error: 25.97 -> 25.79 | Helped |
| Add layer-wise reconstruction loss instead of using reconstruction only at the image input (ablation) | SAE · ImageNet ILSVRC-2012 validation, single-crop | Top-1 error: 27.70 -> 27.54 | Helped |
| Add layer-wise reconstruction loss instead of using reconstruction only at the image input (ablation) | SAE · ImageNet ILSVRC-2012 validation, convolution | Top-1 error: 26.09 -> 26.10 | No clear effect |
| Add the auxiliary reconstructive pathway | VGGNet · ImageNet ILSVRC-2012 training set | training error | Helped |
| Use reconstruction only at the image input rather than layer-wise reconstruction losses (ablation) | SAE/SWWAE · ImageNet ILSVRC-2012 validation | error | Hurt |
Closest papers by meaning
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.89)
- Unsupervised Deep Learning by Neighbourhood Discovery (ICML 2019 · similarity 0.87)
- Unsupervised Deep Learning via Affinity Diffusion (AAAI 2020 · similarity 0.86)
- Unsupervised Representation Learning by Predicting Image Rotations (ICLR 2018 · similarity 0.86)
- Unsupervised Learning by Predicting Noise (ICML 2017 · similarity 0.86)
- URLOST: Unsupervised Representation Learning without Stationarity or Topology (ICLR 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.