analiza aiqbench
Vector Quantized Latent Concepts (VQLC) offers a scalable alternative for interpreting LLM hidden states by balancing the semantic coherence of hierarchical clustering with the computational efficiency of K-Means. The method achieves this through a discrete codebook learned on frozen hidden states, maintaining competitive faithfulness while significantly reducing memory constraints on larger datasets. This approach provides a practical framework for extracting interpretable, task-relevant concepts without the quadratic cost limitations of traditional clustering techniques.
Vector Quantized Latent Concepts: A Scalable Alternative to Clustering-Based Concept Discovery
arXiv:2602.02726v2 Announce Type: replace Abstract: Large language models (LLMs) encode rich semantic information in their hidden states, yet it remains difficult to understand what information these internal representations capture. Latent concepts extracted from hidden states offer a promising direction for interpreting LLMs, but existing clustering-based methods face a trade-off: hierarchical clustering produces coherent concepts but is limited to small datasets due to its quadratic memory cost, while K-Means scales efficiently but may yield less semantically coherent concepts. We propose Vector Quantized Latent Concept (VQLC), a discrete concept learning framework that learns a codebook of latent concepts on frozen hidden states. Across 12 dataset-model settings, VQLC stays close to K-Means in computational cost, scales better than hierarchical clustering, and remains competitive in faithfulness, with the clearest gains on decoder-only models. LLMs-based evaluation, qualitative analysis, and a Sparse Autoencoder (SAE) comparison demonstrate that the learned concepts are interpretable and task-relevant.
Abstract:Large language models (LLMs) encode rich semantic information in their hidden states, yet it remains difficult to understand what information these internal representations capture. Latent concepts extracted from hidden states offer a promising direction for interpreting LLMs, but existing clustering-based methods face a trade-off: hierarchical clustering produces coherent concepts but is limited to small datasets due to its quadratic memory cost, while K-Means scales efficiently but may yield less semantically coherent concepts. We propose Vector Quantized Latent Concept (VQLC), a discrete concept learning framework that learns a codebook of latent concepts on frozen hidden states. Across 12 dataset-model settings, VQLC stays close to K-Means in computational cost, scales better than hierarchical clustering, and remains competitive in faithfulness, with the clearest gains on decoder-only models. LLMs-based evaluation, qualitative analysis, and a Sparse Autoencoder (SAE) comparison demonstrate that the learned concepts are interpretable and task-relevant.
Machine Learning (cs.LG); Computation and Language (cs.CL)
arXiv:2602.02726 [cs.LG]
(or arXiv:2602.02726v2 [cs.LG] for this version)
https://doi.org/10.48550/arXiv.2602.02726 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 239 cuvinte
Subiecte conexe
PIRAMID: Progress and Plans
LessWrong · 27 iul. 2026
J++ Lens: Jacobian Filtering Enables More Faithful Workspace Lenses
LessWrong · 8 oct. 2026
Cross-Layer Discrete Concept Discovery for Interpreting Language Models
arXiv · 1 oct. 2026
Cross-Layer Discrete Concept Discovery for Interpreting Language Models
arXiv · 12 iun. 2026