Research map / Architectures and efficiency / RNNs and normalization
FlowNIB: An Information Bottleneck Analysis of Bidirectional vs. Unidirectional Language Models
Md Kowsher, Nusrat Jahan Prottasha, Shiyun Xu, Shetu Mohanto, Niloofar Yousefi, Ozlem Garibay, Chen Chen
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 masked-token prediction inference instead of pooling | DeBERTa-v3-Base · 9 classification tasks avg | accuracy (avg): 77.90 -> 81.52 | Improved |
| use masked-token prediction inference instead of pooling | DeBERTa-v3-Large · 9 classification tasks avg | accuracy (avg): 80.96 -> 84.73 | Improved |
| use masked-token prediction inference instead of pooling | RoBERTa-Base · 9 classification tasks avg | accuracy (avg): 76.53 -> 79.95 | Improved |
| use masked-token prediction inference instead of pooling | RoBERTa-Large · 9 classification tasks avg | accuracy (avg): 80.14 -> 83.95 | Improved |
| use masked-token prediction inference instead of pooling | ModernBERT-Base · 9 classification tasks avg | accuracy (avg): 76.74 -> 79.73 | Improved |
| use masked-token prediction inference instead of pooling | ModernBERT-Large · 9 classification tasks avg | accuracy (avg): 80.35 -> 83.84 | Improved |
| use generation-based prediction inference instead of pooling | GPT-2 Medium · 9 classification tasks avg | accuracy (avg): 71.02 -> 72.04 | Improved |
| use generation-based prediction inference instead of pooling | GPT-2 Large · 9 classification tasks avg | accuracy (avg): 71.26 -> 72.07 | Improved |
| use generation-based prediction inference instead of pooling | SmolLM2-135M · 9 classification tasks avg | accuracy (avg): 71.37 -> 72.82 | Improved |
| use generation-based prediction inference instead of pooling | SmolLM2-360M · 9 classification tasks avg | accuracy (avg): 72.95 -> 74.40 | Improved |
| use generation-based prediction inference instead of pooling | MobileLLM-125M · 9 classification tasks avg | accuracy (avg): 70.48 -> 71.92 | Improved |
| use generation-based prediction inference instead of pooling | MobileLLM-350M · 9 classification tasks avg | accuracy (avg): 71.89 -> 73.73 | Improved |
Closest papers by meaning
- Talking Heads: Understanding Inter-Layer Communication in Transformer Language Models (NeurIPS 2024 · similarity 0.88)
- Layer by Layer: Uncovering Hidden Representations in Language Models (ICML 2025 · similarity 0.87)
- Taming Knowledge Conflicts in Language Models (ICML 2025 · similarity 0.86)
- Order-Level Attention Similarity Across Language Models: A Latent Commonality (NeurIPS 2025 · similarity 0.86)
- Steering Information Utility in Key-Value Memory for Language Model Post-Training (NeurIPS 2025 · similarity 0.86)
- Understanding the Skill Gap in Recurrent Language Models: The Role of the Gather-and-Aggregate Mechanism (ICML 2025 · 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.
