Research map / Label-efficient and robust learning / Recognition and long-tail
Mutual exclusivity as a challenge for deep neural networks
Kanishk Gandhi, Brenden M Lake
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 feedforward networks with maximum-likelihood loss on one-to-one synthetic mappings | Feedforward classifiers · Synthetic classification | ME score | Hurt |
| Use entropy regularization instead of the standard setup (ablation) | Feedforward classifier · Synthetic classification | ME score | No clear effect |
| Add entropy regularization (ablation) | Feedforward classifier · Synthetic classification | confidence on seen classes | Hurt |
| Train GRU seq2seq models on one-to-one synthetic mappings | Seq2seq models with and without attention · Synthetic sequence mapping | ME score | Hurt |
| Add attention to the seq2seq model (ablation) | GRU seq2seq model · Synthetic sequence mapping | ME score | No clear effect |
| Use an oracle signal and add an ME bias of 5 rather than 0 (ablation) | CNN classifier · Omniglot online classification, new-class instances | classification loss: 5 from 0 | Improved |
| Increase the oracle ME bias beyond 5 (ablation) | CNN classifier · Omniglot online classification, new-class instances | classification loss | No clear effect |
| Use standard online classifiers rather than the dataset's new-class occurrence pattern | Omniglot and ImageNet classifiers · Lifelong image classification | ME score: 8,960 vs 111,872 | Hurt |
| Assume a new source-language word predicts a new target-language word | Lifelong translation learner · IWSLT'14, WMT'14, and WMT'15 | conditional probability of new target word: 0.9: 0.3K–4K; 0.5: 3K–40K; 0.1: 90K–150K | Mixed |
Closest papers by meaning
- A distributional simplicity bias in the learning dynamics of transformers (NeurIPS 2024 · similarity 0.85)
- Data Distributional Properties As Inductive Bias for Systematic Generalization (CVPR 2025 · similarity 0.84)
- Nested Learning: The Illusion of Deep Learning Architectures (NeurIPS 2025 · similarity 0.84)
- The Emergence of Abstract Thought in Large Language Models Beyond Any Language (NeurIPS 2025 · similarity 0.84)
- Separating the 'what' and 'how' of compositional computation to enable reuse and continual learning (NeurIPS 2025 · similarity 0.84)
- Inductive biases of multi-task learning and finetuning: multiple regimes of feature reuse (NeurIPS 2024 · similarity 0.84)
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.
