Sparse Autoencoders Enable Scalable and Reliable Circuit Identification in Language Models

• URL: http://arxiv.org/abs/2405.12522v1
• Date: Tue, 21 May 2024 06:26:10 GMT
• Title: Sparse Autoencoders Enable Scalable and Reliable Circuit Identification in Language Models
• Authors: Charles O'Neill, Thang Bui,
• Abstract summary: This paper introduces an efficient and robust method for discovering interpretable circuits in large language models. We propose training sparse autoencoders on carefully designed positive and negative examples. Our findings highlight the promise of discrete sparse autoencoders for scalable and efficient mechanistic interpretability.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: This paper introduces an efficient and robust method for discovering interpretable circuits in large language models using discrete sparse autoencoders. Our approach addresses key limitations of existing techniques, namely computational complexity and sensitivity to hyperparameters. We propose training sparse autoencoders on carefully designed positive and negative examples, where the model can only correctly predict the next token for the positive examples. We hypothesise that learned representations of attention head outputs will signal when a head is engaged in specific computations. By discretising the learned representations into integer codes and measuring the overlap between codes unique to positive examples for each head, we enable direct identification of attention heads involved in circuits without the need for expensive ablations or architectural modifications. On three well-studied tasks - indirect object identification, greater-than comparisons, and docstring completion - the proposed method achieves higher precision and recall in recovering ground-truth circuits compared to state-of-the-art baselines, while reducing runtime from hours to seconds. Notably, we require only 5-10 text examples for each task to learn robust representations. Our findings highlight the promise of discrete sparse autoencoders for scalable and efficient mechanistic interpretability, offering a new direction for analysing the inner workings of large language models.

Related papers

• On the Query Complexity of Verifier-Assisted Language Generation [35.43462431990329]
  We develop a framework for reasoning about constrained generation using a pre-trained language model generator oracle. Access to a verifier can render an intractable problem (information-theoretically or computationally) to a tractable one. We show even simple algorithms, like tokenwise rejection sampling, can enjoy significant benefits from access to a verifier.
  arXiv  Detail & Related papers  (2025-02-17T18:46:32Z)
• Detecting Memorization in Large Language Models [0.0]
  Large language models (LLMs) have achieved impressive results in natural language processing but are prone to memorizing portions of their training data. Traditional methods for detecting memorization rely on output probabilities or loss functions. We introduce an analytical method that precisely detects memorization by examining neuron activations within the LLM.
  arXiv  Detail & Related papers  (2024-12-02T00:17:43Z)
• Triple-Encoders: Representations That Fire Together, Wire Together [51.15206713482718]
  Contrastive Learning is a representation learning method that encodes relative distances between utterances into the embedding space via a bi-encoder. This study introduces triple-encoders, which efficiently compute distributed utterance mixtures from these independently encoded utterances. We find that triple-encoders lead to a substantial improvement over bi-encoders, and even to better zero-shot generalization than single-vector representation models.
  arXiv  Detail & Related papers  (2024-02-19T18:06:02Z)
• Learning to Skip for Language Modeling [33.51322197222855]
  We argue that in language model pretraining, a variable amount of computation should be assigned to different tokens. In our evaluation across 24 NLP tasks, we demonstrate that the proposed method can significantly improve the 1-shot performance.
  arXiv  Detail & Related papers  (2023-11-26T21:45:53Z)
• Incrementally-Computable Neural Networks: Efficient Inference for Dynamic Inputs [75.40636935415601]
  Deep learning often faces the challenge of efficiently processing dynamic inputs, such as sensor data or user inputs. We take an incremental computing approach, looking to reuse calculations as the inputs change. We apply this approach to the transformers architecture, creating an efficient incremental inference algorithm with complexity proportional to the fraction of modified inputs.
  arXiv  Detail & Related papers  (2023-07-27T16:30:27Z)
• Tram: A Token-level Retrieval-augmented Mechanism for Source Code Summarization [76.57699934689468]
  We propose a fine-grained Token-level retrieval-augmented mechanism (Tram) on the decoder side to enhance the performance of neural models. To overcome the challenge of token-level retrieval in capturing contextual code semantics, we also propose integrating code semantics into individual summary tokens.
  arXiv  Detail & Related papers  (2023-05-18T16:02:04Z)
• Self-Supervised Learning via Maximum Entropy Coding [57.56570417545023]
  We propose Maximum Entropy Coding (MEC) as a principled objective that explicitly optimize on the structure of the representation. MEC learns a more generalizable representation than previous methods based on specific pretext tasks. It achieves state-of-the-art performance consistently on various downstream tasks, including not only ImageNet linear probe, but also semi-supervised classification, object detection, instance segmentation, and object tracking.
  arXiv  Detail & Related papers  (2022-10-20T17:58:30Z)
• Lexically Aware Semi-Supervised Learning for OCR Post-Correction [90.54336622024299]
  Much of the existing linguistic data in many languages of the world is locked away in non-digitized books and documents. Previous work has demonstrated the utility of neural post-correction methods on recognition of less-well-resourced languages. We present a semi-supervised learning method that makes it possible to utilize raw images to improve performance.
  arXiv  Detail & Related papers  (2021-11-04T04:39:02Z)
• Do Encoder Representations of Generative Dialogue Models Encode Sufficient Information about the Task ? [41.36218215755317]
  We showcase evaluating the text generated through human or automatic metrics is not sufficient to appropriately evaluate soundness of the language understanding of dialogue models. We propose a set of probe tasks to evaluate encoder representation of different language encoders commonly used in dialogue models.
  arXiv  Detail & Related papers  (2021-06-20T04:52:37Z)
• Improve Variational Autoencoder for Text Generationwith Discrete Latent Bottleneck [52.08901549360262]
  Variational autoencoders (VAEs) are essential tools in end-to-end representation learning. VAEs tend to ignore latent variables with a strong auto-regressive decoder. We propose a principled approach to enforce an implicit latent feature matching in a more compact latent space.
  arXiv  Detail & Related papers  (2020-04-22T14:41:37Z)

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.