Research map / Architectures and efficiency / RNNs and normalization
Dual Input Stream Transformer for Vertical Drift Correction in Eye-Tracking Reading Data
Thomas M. Mercier, Marcin Budka, Martin R. Vasilev, Julie A. Kirkby, Bernhard Angele, Timothy J. Slattery
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 |
|---|---|---|---|
| ensemble six DIST instances over two normalization schemes (E-DIST) | DIST · Mean over 9 eye-tracking datasets (cross-dataset) | accuracy: 95.32 -> 97.67 | Improved |
| add E-DIST votes to classical algorithms via weighted voting (E-WOC) | C-WOC · Mean over 9 eye-tracking datasets (cross-dataset) | accuracy: 96.75 -> 98.17 | Improved |
| combine E-DIST with classical algorithms via E-WOC | E-DIST · OffN dataset (held-out) | accuracy: 92.01 -> 97.00 | Improved |
| ensemble DIST instances (E-DIST) | DIST · OffN dataset (held-out) | accuracy: 75.18 -> 92.01 | Improved |
| add the second input stream (page image through CoAtNet) (ablation) | DIST · 9 datasets | relative accuracy | Improved |
| encode trial info with CoAtNet instead of MLP or BERT on bounding boxes (ablation) | DIST · 9 datasets | relative accuracy | Improved |
| render fixations, characters and bounding boxes together as the image stream (ablation) | DIST · 9 datasets | relative accuracy | Improved |
| use CORN ordinal loss instead of cross-entropy (ablation) | DIST · 9 datasets | accuracy | No clear effect |
| use regression (MSE) loss instead of CORN (ablation) | DIST · 9 datasets | accuracy | Hurt |
| add lw-norm on top of xy-norm (ablation) | DIST · MECOde, CD, OffN | relative accuracy | Mixed |
| add synthetic training trials (at least 2000) (ablation) | DIST · Abbr, Carr, Harry, MECOde | relative accuracy | Improved |
| add synthetic training trials (ablation) | DIST · OffN | relative accuracy | Hurt |
Closest papers by meaning
- Why Do DiT Editors Drift? Plug-and-Play Low Frequency Alignment in VAE Latent Space (NeurIPS 2026 · similarity 0.83)
- Differential Transformer (ICLR 2025 · similarity 0.82)
- Diffusion Instruction Tuning (ICML 2025 · similarity 0.82)
- DiffEye: Diffusion-Based Continuous Eye-Tracking Data Generation Conditioned on Natural Images (NeurIPS 2025 · similarity 0.82)
- Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations (NeurIPS 2025 · similarity 0.81)
- Diff-Aid: Inference-time Adaptive Interaction Denoising for Rectified Text-to-Image Generation (NeurIPS 2026 · similarity 0.81)
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.
