Decoding Dark Matter: Specialized Sparse Autoencoders for Interpreting Rare Concepts in Foundation Models

• URL: http://arxiv.org/abs/2411.00743v1
• Date: Fri, 01 Nov 2024 17:09:34 GMT
• Title: Decoding Dark Matter: Specialized Sparse Autoencoders for Interpreting Rare Concepts in Foundation Models
• Authors: Aashiq Muhamed, Mona Diab, Virginia Smith,
• Abstract summary: Specialized Sparse Autoencoders (SSAEs) illuminate elusive dark matter features by focusing on specific. We show that SSAEs effectively capture subdomain tail concepts, exceeding the capabilities of general-purpose SAEs. We showcase the practical utility of SSAEs in a case study on the Bias in Bios dataset, where SSAEs achieve a 12.5% increase in worst-group classification accuracy when applied to remove spurious gender information.
• Score: 26.748765050034876
• License:
• Abstract: Understanding and mitigating the potential risks associated with foundation models (FMs) hinges on developing effective interpretability methods. Sparse Autoencoders (SAEs) have emerged as a promising tool for disentangling FM representations, but they struggle to capture rare, yet crucial concepts in the data. We introduce Specialized Sparse Autoencoders (SSAEs), designed to illuminate these elusive dark matter features by focusing on specific subdomains. We present a practical recipe for training SSAEs, demonstrating the efficacy of dense retrieval for data selection and the benefits of Tilted Empirical Risk Minimization as a training objective to improve concept recall. Our evaluation of SSAEs on standard metrics, such as downstream perplexity and $L_0$ sparsity, show that they effectively capture subdomain tail concepts, exceeding the capabilities of general-purpose SAEs. We showcase the practical utility of SSAEs in a case study on the Bias in Bios dataset, where SSAEs achieve a 12.5\% increase in worst-group classification accuracy when applied to remove spurious gender information. SSAEs provide a powerful new lens for peering into the inner workings of FMs in subdomains.

Related papers

• Interpretability as Compression: Reconsidering SAE Explanations of Neural Activations with MDL-SAEs [0.0]
  We present an information-theoretic framework for interpreting SAEs as lossy compression algorithms. We argue that using MDL rather than sparsity may avoid potential pitfalls with naively maximising sparsity.
  arXiv  Detail & Related papers  (2024-10-15T01:38:03Z)
• IncSAR: A Dual Fusion Incremental Learning Framework for SAR Target Recognition [7.9330990800767385]
  Models' tendency to forget old knowledge when learning new tasks, known as catastrophic forgetting, remains an open challenge. In this paper, an incremental learning framework, called IncSAR, is proposed to mitigate catastrophic forgetting in SAR target recognition. IncSAR comprises a Vision Transformer (ViT) and a custom-designed Convolutional Neural Network (CNN) in individual branches combined through a late-fusion strategy.
  arXiv  Detail & Related papers  (2024-10-08T08:49:47Z)
• Efficient Prompt Tuning of Large Vision-Language Model for Fine-Grained Ship Classification [62.425462136772666]
  Fine-grained ship classification in remote sensing (RS-FGSC) poses a significant challenge due to the high similarity between classes and the limited availability of labeled data. Recent advancements in large pre-trained Vision-Language Models (VLMs) have demonstrated impressive capabilities in few-shot or zero-shot learning. This study delves into harnessing the potential of VLMs to enhance classification accuracy for unseen ship categories.
  arXiv  Detail & Related papers  (2024-03-13T05:48:58Z)
• InfoRM: Mitigating Reward Hacking in RLHF via Information-Theoretic Reward Modeling [66.3072381478251]
  Reward hacking, also termed reward overoptimization, remains a critical challenge. We propose a framework for reward modeling, namely InfoRM, by introducing a variational information bottleneck objective. We show that InfoRM's overoptimization detection mechanism is not only effective but also robust across a broad range of datasets.
  arXiv  Detail & Related papers  (2024-02-14T17:49:07Z)
• Enhancing Few-shot CLIP with Semantic-Aware Fine-Tuning [61.902254546858465]
  Methods based on Contrastive Language-Image Pre-training have exhibited promising performance in few-shot adaptation tasks. We propose fine-tuning the parameters of the attention pooling layer during the training process to encourage the model to focus on task-specific semantics.
  arXiv  Detail & Related papers  (2023-11-08T05:18:57Z)
• Improving Deep Representation Learning via Auxiliary Learnable Target Coding [69.79343510578877]
  This paper introduces a novel learnable target coding as an auxiliary regularization of deep representation learning. Specifically, a margin-based triplet loss and a correlation consistency loss on the proposed target codes are designed to encourage more discriminative representations.
  arXiv  Detail & Related papers  (2023-05-30T01:38:54Z)
• Robust Saliency-Aware Distillation for Few-shot Fine-grained Visual Recognition [57.08108545219043]
  Recognizing novel sub-categories with scarce samples is an essential and challenging research topic in computer vision. Existing literature addresses this challenge by employing local-based representation approaches. This article proposes a novel model, Robust Saliency-aware Distillation (RSaD), for few-shot fine-grained visual recognition.
  arXiv  Detail & Related papers  (2023-05-12T00:13:17Z)
• Uncertainty Estimation by Fisher Information-based Evidential Deep Learning [61.94125052118442]
  Uncertainty estimation is a key factor that makes deep learning reliable in practical applications. We propose a novel method, Fisher Information-based Evidential Deep Learning ($mathcalI$-EDL) In particular, we introduce Fisher Information Matrix (FIM) to measure the informativeness of evidence carried by each sample, according to which we can dynamically reweight the objective loss terms to make the network more focused on the representation learning of uncertain classes.
  arXiv  Detail & Related papers  (2023-03-03T16:12:59Z)
• Boosting the Generalization Capability in Cross-Domain Few-shot Learning via Noise-enhanced Supervised Autoencoder [23.860842627883187]
  We teach the model to capture broader variations of the feature distributions with a novel noise-enhanced supervised autoencoder (NSAE) NSAE trains the model by jointly reconstructing inputs and predicting the labels of inputs as well as their reconstructed pairs. We also take advantage of NSAE structure and propose a two-step fine-tuning procedure that achieves better adaption and improves classification performance in the target domain.
  arXiv  Detail & Related papers  (2021-08-11T04:45:56Z)
• Explaining AI-based Decision Support Systems using Concept Localization Maps [4.9449660544238085]
  Concept Localization Maps (CLMs) is a novel approach towards explainable image classifiers employed as Decision Support Systems (DSS) CLMs extend Concept Activation Vectors (CAVs) by locating significant regions corresponding to a learned concept in the latent space of a trained image classifier. We generated a new synthetic dataset called Simple Concept DataBase (SCDB) that includes annotations for 10 distinguishable concepts, and made it publicly available. We achieved localization recall of above 80% for most relevant concepts and average recall above 60% for all concepts using SE-ResNeXt-50 on SCDB.
  arXiv  Detail & Related papers  (2020-05-04T11:33:00Z)

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.