Research map / Large language models / Language, safety and interpretability
Multitasking Models are Robust to Structural Failure: A Neural Model for Bilingual Cognitive Reserve
Giannis Daras, Negin Raoof, Zoi Gkalitsiou, Alex Dimakis
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 |
|---|---|---|---|
| train bilingual (English+Greek) instead of monolingual GPT-2 | GPT-2 small · GLUE SST2 under random pruning 0.45 | accuracy: 0.651 -> 0.773 | Improved |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · GLUE MRPC under random pruning 0.20 | accuracy: 0.416 -> 0.663 | Improved |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · GLUE CoLA under random pruning 0.30 | MCC: 0.216 -> 0.137 | Hurt |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · GLUE RTE under random pruning 0.30 | accuracy: 0.537 -> 0.536 | No clear effect |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · GLUE QQP, no perturbation | accuracy: 0.876 -> 0.843 | Hurt |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · GLUE MRPC under Gaussian noise std 0.06 | accuracy: 0.316 -> 0.610 | Improved |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · GLUE CoLA under Gaussian noise std 0.04 | MCC: 0.316 -> 0.146 | Hurt |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · GLUE SST2 under Gaussian noise std 0.07 | accuracy: 0.545 -> 0.568 | Improved |
| fine-tune from bilingual checkpoint instead of monolingual | GPT-2 small · GLUE SST2 attention weights | sum of squared weights: 70567.875 -> 60663.121 | Improved |
| train bilingual instead of monolingual GPT-2 | GPT-2 small · IMDB perplexity, no perturbation | perplexity: 36 -> 76 | Hurt |
| delete increasing share of attention weights, bilingual vs monolingual | GPT-2 small · IMDB perplexity under random deletion | perplexity | Improved |
| train bilingual for more epochs (1 vs 8) | bilingual GPT-2 · IMDB perplexity under magnitude pruning | perplexity | Improved |
Closest papers by meaning
- How do Large Language Models Handle Multilingualism? (NeurIPS 2024 · similarity 0.84)
- Using Artificial Populations to Study Psychological Phenomena in Neural Models (AAAI 2024 · similarity 0.84)
- Exploring the LLM Journey from Cognition to Expression with Linear Representations (ICML 2024 · similarity 0.84)
- Language Model Behavioral Phases are Consistent Across Architecture, Training Data, and Scale (NeurIPS 2025 · similarity 0.84)
- Turning large language models into cognitive models (ICLR 2024 · similarity 0.84)
- Spurious Forgetting in Continual Learning of Language Models (ICLR 2025 · similarity 0.83)
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.
