Universal Sparse Autoencoders: Interpretable Cross-Model Concept Alignment

• URL: http://arxiv.org/abs/2502.03714v1
• Date: Thu, 06 Feb 2025 02:06:16 GMT
• Title: Universal Sparse Autoencoders: Interpretable Cross-Model Concept Alignment
• Authors: Harrish Thasarathan, Julian Forsyth, Thomas Fel, Matthew Kowal, Konstantinos Derpanis,
• Abstract summary: Universal Sparse Autoencoders (USAEs) are a framework for uncovering and aligning interpretable concepts spanning multiple deep neural networks. USAEs learn a universal concept space that can reconstruct and interpret the internal activations of multiple models at once.
• Score: 6.614005142754584
• License:
• Abstract: We present Universal Sparse Autoencoders (USAEs), a framework for uncovering and aligning interpretable concepts spanning multiple pretrained deep neural networks. Unlike existing concept-based interpretability methods, which focus on a single model, USAEs jointly learn a universal concept space that can reconstruct and interpret the internal activations of multiple models at once. Our core insight is to train a single, overcomplete sparse autoencoder (SAE) that ingests activations from any model and decodes them to approximate the activations of any other model under consideration. By optimizing a shared objective, the learned dictionary captures common factors of variation-concepts-across different tasks, architectures, and datasets. We show that USAEs discover semantically coherent and important universal concepts across vision models; ranging from low-level features (e.g., colors and textures) to higher-level structures (e.g., parts and objects). Overall, USAEs provide a powerful new method for interpretable cross-model analysis and offers novel applications, such as coordinated activation maximization, that open avenues for deeper insights in multi-model AI systems

Related papers

• From Unimodal to Multimodal: Scaling up Projectors to Align Modalities [16.733970553781887]
  We propose a novel approach that aligns vision and language modalities using only projection layers on pretrained, frozen unimodal encoders. Our method exploits the high semantic similarity between embedding spaces of well-trained vision and language models. It involves selecting semantically similar encoders in the latent space, curating a concept-rich dataset of image-caption pairs, and training simple projectors.
  arXiv  Detail & Related papers  (2024-09-28T17:57:32Z)
• Composition Vision-Language Understanding via Segment and Depth Anything Model [2.0836143651641033]
  This library synergizes the capabilities of the Depth Anything Model (DAM), Segment Anything Model (SAM), and GPT-4V. Through the fusion of segmentation and depth analysis at the symbolic instance level, our library provides nuanced inputs for language models. Our findings showcase progress in vision-language models through neural-symbolic integration.
  arXiv  Detail & Related papers  (2024-06-07T16:28:06Z)
• Slot Structured World Models [0.0]
  State-of-the-art approaches use a feedforward encoder to extract object embeddings and a latent graph neural network to model the interaction between these object embeddings. We introduce it Slot Structured World Models (SSWM), a class of world models that combines an it object-centric encoder with a latent graph-based dynamics model.
  arXiv  Detail & Related papers  (2024-01-08T21:19:30Z)
• Detecting Any Human-Object Interaction Relationship: Universal HOI Detector with Spatial Prompt Learning on Foundation Models [55.20626448358655]
  This study explores the universal interaction recognition in an open-world setting through the use of Vision-Language (VL) foundation models and large language models (LLMs) Our design includes an HO Prompt-guided Decoder (HOPD), facilitates the association of high-level relation representations in the foundation model with various HO pairs within the image. For open-category interaction recognition, our method supports either of two input types: interaction phrase or interpretive sentence.
  arXiv  Detail & Related papers  (2023-11-07T08:27:32Z)
• Pre-training Contextualized World Models with In-the-wild Videos for Reinforcement Learning [54.67880602409801]
  In this paper, we study the problem of pre-training world models with abundant in-the-wild videos for efficient learning of visual control tasks. We introduce Contextualized World Models (ContextWM) that explicitly separate context and dynamics modeling. Our experiments show that in-the-wild video pre-training equipped with ContextWM can significantly improve the sample efficiency of model-based reinforcement learning.
  arXiv  Detail & Related papers  (2023-05-29T14:29:12Z)
• Concept-Centric Transformers: Enhancing Model Interpretability through Object-Centric Concept Learning within a Shared Global Workspace [1.6574413179773757]
  Concept-Centric Transformers is a simple yet effective configuration of the shared global workspace for interpretability. We show that our model achieves better classification accuracy than all baselines across all problems.
  arXiv  Detail & Related papers  (2023-05-25T06:37:39Z)
• Language Models are General-Purpose Interfaces [109.45478241369655]
  We propose to use language models as a general-purpose interface to various foundation models. A collection of pretrained encoders perceive diverse modalities (such as vision, and language) We propose a semi-causal language modeling objective to jointly pretrain the interface and the modular encoders.
  arXiv  Detail & Related papers  (2022-06-13T17:34:22Z)
• Encoder Fusion Network with Co-Attention Embedding for Referring Image Segmentation [87.01669173673288]
  We propose an encoder fusion network (EFN), which transforms the visual encoder into a multi-modal feature learning network. A co-attention mechanism is embedded in the EFN to realize the parallel update of multi-modal features. The experiment results on four benchmark datasets demonstrate that the proposed approach achieves the state-of-the-art performance without any post-processing.
  arXiv  Detail & Related papers  (2021-05-05T02:27:25Z)
• A Minimalist Dataset for Systematic Generalization of Perception, Syntax, and Semantics [131.93113552146195]
  We present a new dataset, Handwritten arithmetic with INTegers (HINT), to examine machines' capability of learning generalizable concepts. In HINT, machines are tasked with learning how concepts are perceived from raw signals such as images. We undertake extensive experiments with various sequence-to-sequence models, including RNNs, Transformers, and GPT-3.
  arXiv  Detail & Related papers  (2021-03-02T01:32:54Z)
• Neural Entity Linking: A Survey of Models Based on Deep Learning [82.43751915717225]
  This survey presents a comprehensive description of recent neural entity linking (EL) systems developed since 2015. Its goal is to systemize design features of neural entity linking systems and compare their performance to the remarkable classic methods on common benchmarks. The survey touches on applications of entity linking, focusing on the recently emerged use-case of enhancing deep pre-trained masked language models.
  arXiv  Detail & Related papers  (2020-05-31T18:02:26Z)

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.