Representation Surgery: Theory and Practice of Affine Steering

• URL: http://arxiv.org/abs/2402.09631v6
• Date: Fri, 5 Jul 2024 08:14:29 GMT
• Title: Representation Surgery: Theory and Practice of Affine Steering
• Authors: Shashwat Singh, Shauli Ravfogel, Jonathan Herzig, Roee Aharoni, Ryan Cotterell, Ponnurangam Kumaraguru,
• Abstract summary: Language models often exhibit undesirable behavior, e.g., generating toxic or gender-biased text. One natural (and common) approach to prevent the model from exhibiting undesirable behavior is to steer the model's representations. This paper investigates the formal and empirical properties of steering functions.
• Score: 72.61363182652853
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Language models often exhibit undesirable behavior, e.g., generating toxic or gender-biased text. In the case of neural language models, an encoding of the undesirable behavior is often present in the model's representations. Thus, one natural (and common) approach to prevent the model from exhibiting undesirable behavior is to steer the model's representations in a manner that reduces the probability of it generating undesirable text. This paper investigates the formal and empirical properties of steering functions, i.e., transformation of the neural language model's representations that alter its behavior. First, we derive two optimal, in the least-squares sense, affine steering functions under different constraints. Our theory provides justification for existing approaches and offers a novel, improved steering approach. Second, we offer a series of experiments that demonstrate the empirical effectiveness of the methods in mitigating bias and reducing toxic generation.

Related papers

• Inverse Decision Modeling: Learning Interpretable Representations of Behavior [72.80902932543474]
  We develop an expressive, unifying perspective on inverse decision modeling. We use this to formalize the inverse problem (as a descriptive model) We illustrate how this structure enables learning (interpretable) representations of (bounded) rationality.
  arXiv  Detail & Related papers  (2023-10-28T05:05:01Z)
• Interpretable Imitation Learning with Dynamic Causal Relations [65.18456572421702]
  We propose to expose captured knowledge in the form of a directed acyclic causal graph. We also design this causal discovery process to be state-dependent, enabling it to model the dynamics in latent causal graphs. The proposed framework is composed of three parts: a dynamic causal discovery module, a causality encoding module, and a prediction module, and is trained in an end-to-end manner.
  arXiv  Detail & Related papers  (2023-09-30T20:59:42Z)
• Learning to Diversify Neural Text Generation via Degenerative Model [39.961572541752005]
  We propose a new approach to prevent degeneration problems by training two models. We first train a model that is designed to amplify undesirable patterns. We then enhance the diversity of the second model by focusing on patterns that the first model fails to learn.
  arXiv  Detail & Related papers  (2023-09-22T04:57:10Z)
• Probabilistic inverse optimal control for non-linear partially observable systems disentangles perceptual uncertainty and behavioral costs [33.690374799743076]
  We introduce a probabilistic approach to inverse optimal control for partially observable non-linear systems with unobserved action signals. We show that our method can disentangle perceptual factors and behavioral costs despite the fact that neuroscience and pragmatic actions are intertwined in sequential decision-making under uncertainty.
  arXiv  Detail & Related papers  (2023-03-29T13:51:06Z)
• NaturalAdversaries: Can Naturalistic Adversaries Be as Effective as Artificial Adversaries? [61.58261351116679]
  We introduce a two-stage adversarial example generation framework (NaturalAdversaries) for natural language understanding tasks. It is adaptable to both black-box and white-box adversarial attacks based on the level of access to the model parameters. Our results indicate these adversaries generalize across domains, and offer insights for future research on improving robustness of neural text classification models.
  arXiv  Detail & Related papers  (2022-11-08T16:37:34Z)
• A Contrastive Framework for Neural Text Generation [46.845997620234265]
  We show that an underlying reason for model degeneration is the anisotropic distribution of token representations. We present a contrastive solution: (i) SimCTG, a contrastive training objective to calibrate the model's representation space, and (ii) a decoding method -- contrastive search -- to encourage diversity while maintaining coherence in the generated text.
  arXiv  Detail & Related papers  (2022-02-13T21:46:14Z)
• Distilling Interpretable Models into Human-Readable Code [71.11328360614479]
  Human-readability is an important and desirable standard for machine-learned model interpretability. We propose to train interpretable models using conventional methods, and then distill them into concise, human-readable code. We describe a piecewise-linear curve-fitting algorithm that produces high-quality results efficiently and reliably across a broad range of use cases.
  arXiv  Detail & Related papers  (2021-01-21T01:46:36Z)
• On the Transferability of Adversarial Attacksagainst Neural Text Classifier [121.6758865857686]
  We investigate the transferability of adversarial examples for text classification models. We propose a genetic algorithm to find an ensemble of models that can induce adversarial examples to fool almost all existing models. We derive word replacement rules that can be used for model diagnostics from these adversarial examples.
  arXiv  Detail & Related papers  (2020-11-17T10:45:05Z)

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.