[2406.18534] Towards Compositionality in Concept Learning
[2406.18534] Towards Compositionality in Concept Learning - arXiv
Our evaluation shows that CCE finds more compositional concept representations than baselines and yields better accuracy on four downstream classification ...
Towards Compositionality in Concept Learning - arXiv
These concept representations are most useful when they are compositional, meaning that the individual concepts compose to explain the full ...
(PDF) Towards Compositionality in Concept Learning - ResearchGate
These concept representations are most useful when they are compositional, meaning that the individual concepts compose to explain the full ...
[PDF] Towards Compositionality in Concept Learning | Semantic ...
This work identifies two salient properties of compositional concept representations, and proposes Compositional Concept Extraction (CCE) for finding ...
Towards Compositionality in Concept Learning
Towards Compositionality in Concept LearningAdam Stein, Aaditya Naik, Yinjun Wu, Mayur Naik, Eric WongConcept-based interpretability meth...
Towards Compositionality in Concept Learning - Emergent Mind
The paper by Stein et al. proposes Compositional Concept Extraction (CCE) to enhance the interpretability of foundation models by identifying compositional ...
Towards Compositionality in Concept Learning | AI Research Paper ...
Towards Compositionality in Concept Learning. Read original: arXiv:2406.18534 - Published 6/26/2024 by Adam Stein, Aaditya Naik, Yinjun Wu, Mayur Naik, Eric ...
Towards Compositionality in Concept Learning - OpenReview
These concept representations are most useful when they are compositional, meaning that the individual concepts compose to explain the full ...
Towards Compositionality in Concept Learning - Penn Engineering
These concept representations are most useful when they are compositional, meaning that the individual concepts compose to explain the full sample. We show that ...
Adam Stein on X: "Excited to present our ICML 2024 paper ...
Excited to present our ICML 2024 paper: "Towards Compositionality in Concept Learning"! ... Paper: https://arxiv.org/abs/2406.18534 ...
Eric Wong on X: "Traditional concept vectors used to explain deep ...
Excited to present our ICML 2024 paper: "Towards Compositionality in Concept Learning"! ... Paper: https://arxiv.org/abs/2406.18534 Code ...
Towards Compositionality in Concept Learning | alphaXiv
... concepts. These concept representations are most useful when they are compositional, meaning that the individual concepts compose to explain the full sample ...
Towards Compositionality in Concept Learning - Paper Reading
Concept-based interpretability methods offer a lens into the internals of foundation models by decomposing their embeddings into high-level concepts.
Towards Compositionality in Concept Learning - UPenn CIS
These concept representations are most useful when they are compositional, meaning that the individual concepts compose to explain the full sample. We show that ...
Aaditya Naik - Google 学术搜索
Towards compositionality in concept learning. A Stein, A Naik, Y Wu, M Naik, E Wong. arXiv preprint arXiv:2406.18534, 2024. 2, 2024. Torchql: A programming ...
Towards Compositionality in Concept Learning - DebugML
We propose Compositional Concept Extraction (CCE), a new concept learning approach that encourages concepts that linearly compose.
Related papers: Towards Compositionality in Concept Learning
Towards Compositionality in Concept Learning. URL: http://arxiv.org/abs/2406.18534v1; Date: Wed, 26 Jun 2024 17:59:30 GMT; Title: Towards Compositionality ...
Language-Informed Visual Concept Learning - Stanford AI Lab
Compared the text-based image editing methods, our method achieves significantly better compositionality due to the disentangled concept representation, and ...