Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Zero-1-to-3: Domain-Level Zero-Shot Cognitive Diagnosis via One Batch of Early-Bird Students towards Three Diagnostic Objectives
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Zero-1-to-3, a domain-level zero-shot cognitive diagnosis framework that decouples student states into domain-shared and domain-specific parts via dual regularizers, then fine-tunes cold-start students on simulated logs derived from early-bird students.
Problem
Domain-level zero-shot cognitive diagnosis (DZCD): estimating cognitive states of unseen students in a newly launched domain with no practice logs, where only a first batch of early-bird students has logs.
Approach
Decouple student embeddings into domain-shared and domain-specific parts with two adversarial-style regularizers so only transferable shared states move to the new domain; then generate simulated practice logs for cold-start students by copying early-bird students' records to behaviorally similar peers, and fine-tune on them.
Key results
- Zero-NeuralCD (Zero-1-to-3 with NeuralCD) achieves best or near-best zero-shot performance on most target domains, e.g., ACC 61.48 on Geometry vs 60.20 for Tech-based baseline.
- On Physics as target, Zero-MIRT reaches AUC 72.06 vs 61.18 for the NLP baseline.
- Performance does not consistently improve as the number of simulated peer students grows, indicating noise from larger simulations.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Zero-NeuralCD instead of the strongest reported baseline | NeuralCD for DZCD · Geometry | AUC: 60.23 -> 60.30 | Helped |
| Use Zero-NeuralCD instead of the strongest reported baseline | NeuralCD for DZCD · Function | ACC: 58.90 -> 59.78 | Helped |
| Use Zero-NeuralCD instead of the strongest reported baseline | NeuralCD for DZCD · Probability | AUC: 53.53 -> 67.43 | Helped |
| Use Zero-NeuralCD instead of the strongest reported baseline | NeuralCD for DZCD · Physics | AUC: 56.45 -> 68.08 | Helped |
| Use Zero-NeuralCD instead of the strongest reported baseline | NeuralCD for DZCD · Arithmetic | ACC: 60.63 -> 60.89 | Helped |
| Use Zero-NeuralCD instead of the strongest reported baseline | NeuralCD for DZCD · English | RMSE: 50.10 -> 48.96 | Helped |
| Increase the number of simulated peer students (ablation) | Zero-NeuralCD · Six DZCD target datasets | diagnostic performance | Mixed |
Surprising findings
- Performance does not consistently improve with more simulated peer students; larger p introduces noise.
- Zero-1-to-3 does not win on every metric; e.g., under IRT on Geometry the NLP baseline has higher ACC.
Limitations
- Simulated logs may be noisy and do not cover all cold-start students; peer selection parameter p is sensitive.
- Only student-side transfer is addressed; question/concept embeddings rely on BERT content encoding.
- Evaluated on one private platform's six datasets, each with one concept per question.
Experimental setup
Datasets: Geometry Function Probability Physics Arithmetic English
Benchmarks: Geometry Function Probability Physics Arithmetic English
Metrics: ACC AUC RMSE
Baselines: Random Oracle NLP-based GCN-based Tech-based
Builds on: NeuralCD IRT MIRT TechCD BERT
Closest papers by meaning
- Multi-Perspective Consolidation Enhanced Cognitive Diagnosis via Conditional Diffusion Model (AAAI 2025 · similarity 0.88)
- Knowledge Is Power: Harnessing Large Language Models for Enhanced Cognitive Diagnosis (AAAI 2025 · similarity 0.88)
- AD4CD: Causal-Guided Anomaly Detection for Enhancing Cognitive Diagnosis (AAAI 2025 · similarity 0.88)
- Collaborative Cognitive Diagnosis with Disentangled Representation Learning for Learner Modeling (NeurIPS 2024 · similarity 0.87)
- Towards Accurate and Fair Cognitive Diagnosis via Monotonic Data Augmentation (NeurIPS 2024 · similarity 0.87)
- Symbolic Cognitive Diagnosis via Hybrid Optimization for Intelligent Education Systems (AAAI 2024 · similarity 0.87)
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.