Sparse Autoencoder Features for Classifications and Transferability

• URL: http://arxiv.org/abs/2502.11367v1
• Date: Mon, 17 Feb 2025 02:30:45 GMT
• Title: Sparse Autoencoder Features for Classifications and Transferability
• Authors: Jack Gallifant, Shan Chen, Kuleen Sasse, Hugo Aerts, Thomas Hartvigsen, Danielle S. Bitterman,
• Abstract summary: We analyze Sparse Autoencoders (SAEs) for interpretable feature extraction from Large Language Models (LLMs) Our framework evaluates (1) model-layer selection and scaling properties, (2) SAE architectural configurations, including width and pooling strategies, and (3) the effect of binarizing continuous SAE activations.
• Score: 11.2185030332009
• License:
• Abstract: Sparse Autoencoders (SAEs) provide potentials for uncovering structured, human-interpretable representations in Large Language Models (LLMs), making them a crucial tool for transparent and controllable AI systems. We systematically analyze SAE for interpretable feature extraction from LLMs in safety-critical classification tasks. Our framework evaluates (1) model-layer selection and scaling properties, (2) SAE architectural configurations, including width and pooling strategies, and (3) the effect of binarizing continuous SAE activations. SAE-derived features achieve macro F1 > 0.8, outperforming hidden-state and BoW baselines while demonstrating cross-model transfer from Gemma 2 2B to 9B-IT models. These features generalize in a zero-shot manner to cross-lingual toxicity detection and visual classification tasks. Our analysis highlights the significant impact of pooling strategies and binarization thresholds, showing that binarization offers an efficient alternative to traditional feature selection while maintaining or improving performance. These findings establish new best practices for SAE-based interpretability and enable scalable, transparent deployment of LLMs in real-world applications. Full repo: https://github.com/shan23chen/MOSAIC.

Related papers

• Adaptive Pruning for Large Language Models with Structural Importance Awareness [66.2690963378878]
  Large language models (LLMs) have significantly improved language understanding and generation capabilities. LLMs are difficult to deploy on resource-constrained edge devices due to their high computational and storage resource demands. We propose structurally-aware adaptive pruning (SAAP) to significantly reduce the computational and memory costs while maintaining model performance.
  arXiv  Detail & Related papers  (2024-12-19T18:08:04Z)
• Rethinking Pre-Trained Feature Extractor Selection in Multiple Instance Learning for Whole Slide Image Classification [2.6703221234079946]
  Multiple instance learning (MIL) has become a preferred method for gigapixel whole slide image (WSI) classification without requiring patch-level annotations. This study systematically evaluating MIL feature extractors across three dimensions: pre-training dataset, backbone model, and pre-training method.
  arXiv  Detail & Related papers  (2024-08-02T10:34:23Z)
• Contextualization Distillation from Large Language Model for Knowledge Graph Completion [51.126166442122546]
  We introduce the Contextualization Distillation strategy, a plug-in-and-play approach compatible with both discriminative and generative KGC frameworks. Our method begins by instructing large language models to transform compact, structural triplets into context-rich segments. Comprehensive evaluations across diverse datasets and KGC techniques highlight the efficacy and adaptability of our approach.
  arXiv  Detail & Related papers  (2024-01-28T08:56:49Z)
• Bidirectional Trained Tree-Structured Decoder for Handwritten Mathematical Expression Recognition [51.66383337087724]
  The Handwritten Mathematical Expression Recognition (HMER) task is a critical branch in the field of OCR. Recent studies have demonstrated that incorporating bidirectional context information significantly improves the performance of HMER models. We propose the Mirror-Flipped Symbol Layout Tree (MF-SLT) and Bidirectional Asynchronous Training (BAT) structure.
  arXiv  Detail & Related papers  (2023-12-31T09:24:21Z)
• Scaling In-Context Demonstrations with Structured Attention [75.41845145597875]
  We propose a better architectural design for in-context learning. Structured Attention for In-Context Learning replaces the full-attention by a structured attention mechanism. We show that SAICL achieves comparable or better performance than full attention while obtaining up to 3.4x inference speed-up.
  arXiv  Detail & Related papers  (2023-07-05T23:26:01Z)
• RefSAM: Efficiently Adapting Segmenting Anything Model for Referring Video Object Segmentation [53.4319652364256]
  This paper presents the RefSAM model, which explores the potential of SAM for referring video object segmentation. Our proposed approach adapts the original SAM model to enhance cross-modality learning by employing a lightweight Cross-RValModal. We employ a parameter-efficient tuning strategy to align and fuse the language and vision features effectively.
  arXiv  Detail & Related papers  (2023-07-03T13:21:58Z)
• A Dual-branch Self-supervised Representation Learning Framework for Tumour Segmentation in Whole Slide Images [12.961686610789416]
  Self-supervised learning (SSL) has emerged as an alternative solution to reduce the annotation overheads in whole slide images. These SSL approaches are not designed for handling multi-resolution WSIs, which limits their performance in learning discriminative image features. We propose a Dual-branch SSL Framework for WSI tumour segmentation (DSF-WSI) that can effectively learn image features from multi-resolution WSIs.
  arXiv  Detail & Related papers  (2023-03-20T10:57:28Z)
• USER: Unified Semantic Enhancement with Momentum Contrast for Image-Text Retrieval [115.28586222748478]
  Image-Text Retrieval (ITR) aims at searching for the target instances that are semantically relevant to the given query from the other modality. Existing approaches typically suffer from two major limitations.
  arXiv  Detail & Related papers  (2023-01-17T12:42:58Z)
• i-MAE: Are Latent Representations in Masked Autoencoders Linearly Separable? [26.146459754995597]
  Masked image modeling (MIM) has been recognized as a strong self-supervised pre-training approach in the vision domain. This paper aims to explore an interactive Masked Autoencoders (i-MAE) framework to enhance the representation capability. In addition to qualitatively analyzing the characteristics of the latent representations, we examine the existence of linear separability and the degree of semantics in the latent space.
  arXiv  Detail & Related papers  (2022-10-20T17:59:54Z)

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.