Mechanistically analyzing the effects of fine-tuning on procedurally defined tasks

• URL: http://arxiv.org/abs/2311.12786v2
• Date: Wed, 21 Aug 2024 16:37:20 GMT
• Title: Mechanistically analyzing the effects of fine-tuning on procedurally defined tasks
• Authors: Samyak Jain, Robert Kirk, Ekdeep Singh Lubana, Robert P. Dick, Hidenori Tanaka, Edward Grefenstette, Tim Rocktäschel, David Scott Krueger,
• Abstract summary: We study how fine-tuning alters the underlying capabilities learned by a model during pretraining. We show that fine-tuning rarely alters the underlying model capabilities. We also show that fine-tuning can unintentionally remove a model's safety wrapper.
• Score: 37.278707106871295
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Fine-tuning large pre-trained models has become the de facto strategy for developing both task-specific and general-purpose machine learning systems, including developing models that are safe to deploy. Despite its clear importance, there has been minimal work that explains how fine-tuning alters the underlying capabilities learned by a model during pretraining: does fine-tuning yield entirely novel capabilities or does it just modulate existing ones? We address this question empirically in synthetic, controlled settings where we can use mechanistic interpretability tools (e.g., network pruning and probing) to understand how the model's underlying capabilities are changing. We perform an extensive analysis of the effects of fine-tuning in these settings, and show that: (i) fine-tuning rarely alters the underlying model capabilities; (ii) a minimal transformation, which we call a 'wrapper', is typically learned on top of the underlying model capabilities, creating the illusion that they have been modified; and (iii) further fine-tuning on a task where such hidden capabilities are relevant leads to sample-efficient 'revival' of the capability, i.e., the model begins reusing these capability after only a few gradient steps. This indicates that practitioners can unintentionally remove a model's safety wrapper merely by fine-tuning it on a, e.g., superficially unrelated, downstream task. We additionally perform analysis on language models trained on the TinyStories dataset to support our claims in a more realistic setup.

Related papers

• Parameter-Efficient and Memory-Efficient Tuning for Vision Transformer: A Disentangled Approach [87.8330887605381]
  We show how to adapt a pre-trained Vision Transformer to downstream recognition tasks with only a few learnable parameters. We synthesize a task-specific query with a learnable and lightweight module, which is independent of the pre-trained model. Our method achieves state-of-the-art performance under memory constraints, showcasing its applicability in real-world situations.
  arXiv  Detail & Related papers  (2024-07-09T15:45:04Z)
• Fine-Tuning Enhances Existing Mechanisms: A Case Study on Entity Tracking [53.66999416757543]
  We study how fine-tuning affects the internal mechanisms implemented in language models. Fine-tuning enhances, rather than alters, the mechanistic operation of the model.
  arXiv  Detail & Related papers  (2024-02-22T18:59:24Z)
• Task-Specific Skill Localization in Fine-tuned Language Models [36.53572616441048]
  This paper introduces the term skill localization for this problem. A simple optimization is used to identify a very small subset of parameters. grafting the fine-tuned values for just this tiny subset onto the pre-trained model gives performance almost as well as the fine-tuned model.
  arXiv  Detail & Related papers  (2023-02-13T18:55:52Z)
• Exploring Strategies for Generalizable Commonsense Reasoning with Pre-trained Models [62.28551903638434]
  We measure the impact of three different adaptation methods on the generalization and accuracy of models. Experiments with two models show that fine-tuning performs best, by learning both the content and the structure of the task, but suffers from overfitting and limited generalization to novel answers. We observe that alternative adaptation methods like prefix-tuning have comparable accuracy, but generalize better to unseen answers and are more robust to adversarial splits.
  arXiv  Detail & Related papers  (2021-09-07T03:13:06Z)
• Sufficiently Accurate Model Learning for Planning [119.80502738709937]
  This paper introduces the constrained Sufficiently Accurate model learning approach. It provides examples of such problems, and presents a theorem on how close some approximate solutions can be. The approximate solution quality will depend on the function parameterization, loss and constraint function smoothness, and the number of samples in model learning.
  arXiv  Detail & Related papers  (2021-02-11T16:27:31Z)
• Goal-Aware Prediction: Learning to Model What Matters [105.43098326577434]
  One of the fundamental challenges in using a learned forward dynamics model is the mismatch between the objective of the learned model and that of the downstream planner or policy. We propose to direct prediction towards task relevant information, enabling the model to be aware of the current task and encouraging it to only model relevant quantities of the state space. We find that our method more effectively models the relevant parts of the scene conditioned on the goal, and as a result outperforms standard task-agnostic dynamics models and model-free reinforcement learning.
  arXiv  Detail & Related papers  (2020-07-14T16:42:59Z)

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.