Cross-Layer Discrete Concept Discovery for Interpreting Language Models

• URL: http://arxiv.org/abs/2506.20040v2
• Date: Wed, 16 Jul 2025 21:35:12 GMT
• Title: Cross-Layer Discrete Concept Discovery for Interpreting Language Models
• Authors: Ankur Garg, Xuemin Yu, Hassan Sajjad, Samira Ebrahimi Kahou,
• Abstract summary: Cross-layer VQ-VAE is a framework that uses vector quantization to map representations across layers.<n>Our approach uniquely combines top-k temperature-based sampling during quantization with EMA codebook updates.
• Score: 13.842670153893977
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Uncovering emergent concepts across transformer layers remains a significant challenge because the residual stream linearly mixes and duplicates information, obscuring how features evolve within large language models. Current research efforts primarily inspect neural representations at single layers, thereby overlooking this cross-layer superposition and the redundancy it introduces. These representations are typically either analyzed directly for activation patterns or passed to probing classifiers that map them to a limited set of predefined concepts. To address these limitations, we propose cross-layer VQ-VAE (CLVQ-VAE), a framework that uses vector quantization to map representations across layers and in the process collapse duplicated residual-stream features into compact, interpretable concept vectors. Our approach uniquely combines top-k temperature-based sampling during quantization with EMA codebook updates, providing controlled exploration of the discrete latent space while maintaining code-book diversity. We further enhance the framework with scaled-spherical k-means++ for codebook initialization, which clusters by directional similarity rather than magnitude, better aligning with semantic structure in word embedding space.

Related papers

• Rethinking VAE: From Continuous to Discrete Representations Without Probabilistic Assumptions [0.0]
  We establish connections between Variational Autoencoders (VAEs) and Vector Quantized-Variational Autoencoders (VQ-VAEs) through a reformulated training framework.<n>We propose a new VAE-like training method that introduces clustering centers to enhance data compactness and ensure well-defined latent spaces.<n> Experimental results on MNIST, CelebA, and FashionMNIST datasets show smooth interpolative transitions, though blurriness persists.
  arXiv  Detail & Related papers  (2025-07-23T06:52:00Z)
• SeqPE: Transformer with Sequential Position Encoding [76.22159277300891]
  SeqPE represents each $n$-dimensional position index as a symbolic sequence and employs a lightweight sequential position encoder to learn their embeddings.<n> Experiments across language modeling, long-context question answering, and 2D image classification demonstrate that SeqPE not only surpasses strong baselines in perplexity, exact match (EM) and accuracy--but also enables seamless generalization to multi-dimensional inputs without requiring manual architectural redesign.
  arXiv  Detail & Related papers  (2025-06-16T09:16:40Z)
• Style Quantization for Data-Efficient GAN Training [18.40243591024141]
  Under limited data setting, GANs often struggle to navigate and effectively exploit the input latent space.<n>We propose textitSQ-GAN, a novel approach that enhances consistency regularization.<n>Experiments demonstrate significant improvements in both discriminator robustness and generation quality.
  arXiv  Detail & Related papers  (2025-03-31T16:28:44Z)
• "Principal Components" Enable A New Language of Images [79.45806370905775]
  We introduce a novel visual tokenization framework that embeds a provable PCA-like structure into the latent token space.<n>Our approach achieves state-of-the-art reconstruction performance and enables better interpretability to align with the human vision system.
  arXiv  Detail & Related papers  (2025-03-11T17:59:41Z)
• Trajectory Forecasting through Low-Rank Adaptation of Discrete Latent Codes [36.12653178844828]
  Trajectory forecasting is crucial for video surveillance analytics, as it enables the anticipation of future movements for a set of agents. We introduce Vector Quantized Variational Autoencoders (VQ-VAEs), which utilize a discrete latent space to tackle the issue of posterior collapse. We show that such a two-fold framework, augmented with instance-level discretization, leads to accurate and diverse forecasts.
  arXiv  Detail & Related papers  (2024-05-31T10:13:17Z)
• Deciphering 'What' and 'Where' Visual Pathways from Spectral Clustering of Layer-Distributed Neural Representations [15.59251297818324]
  We present an approach for analyzing grouping information contained within a neural network's activations. We exploit features from all layers and obviating the need to guess which part of the model contains relevant information.
  arXiv  Detail & Related papers  (2023-12-11T01:20:34Z)
• Disentanglement via Latent Quantization [60.37109712033694]
  In this work, we construct an inductive bias towards encoding to and decoding from an organized latent space. We demonstrate the broad applicability of this approach by adding it to both basic data-re (vanilla autoencoder) and latent-reconstructing (InfoGAN) generative models.
  arXiv  Detail & Related papers  (2023-05-28T06:30:29Z)
• Vector Quantized Wasserstein Auto-Encoder [57.29764749855623]
  We study learning deep discrete representations from the generative viewpoint. We endow discrete distributions over sequences of codewords and learn a deterministic decoder that transports the distribution over the sequences of codewords to the data distribution. We develop further theories to connect it with the clustering viewpoint of WS distance, allowing us to have a better and more controllable clustering solution.
  arXiv  Detail & Related papers  (2023-02-12T13:51:36Z)
• ACSeg: Adaptive Conceptualization for Unsupervised Semantic Segmentation [17.019848796027485]
  Self-supervised visual pre-training models have shown great promise in representing pixel-level semantic relationships. In this work, we investigate the pixel-level semantic aggregation in self-trained models as image encodes and design concepts. We propose the Adaptive Concept Generator (ACG) which adaptively maps these prototypes to informative concepts for each image.
  arXiv  Detail & Related papers  (2022-10-12T06:16:34Z)
• HSVA: Hierarchical Semantic-Visual Adaptation for Zero-Shot Learning [74.76431541169342]
  Zero-shot learning (ZSL) tackles the unseen class recognition problem, transferring semantic knowledge from seen classes to unseen ones. We propose a novel hierarchical semantic-visual adaptation (HSVA) framework to align semantic and visual domains. Experiments on four benchmark datasets demonstrate HSVA achieves superior performance on both conventional and generalized ZSL.
  arXiv  Detail & Related papers  (2021-09-30T14:27:50Z)
• Closed-Form Factorization of Latent Semantics in GANs [65.42778970898534]
  A rich set of interpretable dimensions has been shown to emerge in the latent space of the Generative Adversarial Networks (GANs) trained for synthesizing images. In this work, we examine the internal representation learned by GANs to reveal the underlying variation factors in an unsupervised manner. We propose a closed-form factorization algorithm for latent semantic discovery by directly decomposing the pre-trained weights.
  arXiv  Detail & Related papers  (2020-07-13T18:05:36Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.