On the Emergence of Cross-Task Linearity in the Pretraining-Finetuning Paradigm

• URL: http://arxiv.org/abs/2402.03660v2
• Date: Tue, 28 May 2024 08:44:15 GMT
• Title: On the Emergence of Cross-Task Linearity in the Pretraining-Finetuning Paradigm
• Authors: Zhanpeng Zhou, Zijun Chen, Yilan Chen, Bo Zhang, Junchi Yan,
• Abstract summary: We discover an intriguing linear phenomenon in models that are from a common pretrained checkpoint and finetuned on different tasks, termed as Cross-Task Linearity (CTL) We show that if we linearly interpolate the weights of two finetuned models, the features in the weight-interpolated model are often approximately equal to the linearities of features in two finetuned models at each layer. We conjecture that in the pretraining-finetuning paradigm, neural networks approximately function as linear maps, mapping from the parameter space to the feature space.
• Score: 47.55215041326702
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: The pretraining-finetuning paradigm has become the prevailing trend in modern deep learning. In this work, we discover an intriguing linear phenomenon in models that are initialized from a common pretrained checkpoint and finetuned on different tasks, termed as Cross-Task Linearity (CTL). Specifically, we show that if we linearly interpolate the weights of two finetuned models, the features in the weight-interpolated model are often approximately equal to the linear interpolation of features in two finetuned models at each layer. We provide comprehensive empirical evidence supporting that CTL consistently occurs for finetuned models that start from the same pretrained checkpoint. We conjecture that in the pretraining-finetuning paradigm, neural networks approximately function as linear maps, mapping from the parameter space to the feature space. Based on this viewpoint, our study unveils novel insights into explaining model merging/editing, particularly by translating operations from the parameter space to the feature space. Furthermore, we delve deeper into the root cause for the emergence of CTL, highlighting the role of pretraining.

Related papers

• Generalizable Implicit Neural Representation As a Universal Spatiotemporal Traffic Data Learner [46.866240648471894]
  Spatiotemporal Traffic Data (STTD) measures the complex dynamical behaviors of the multiscale transportation system. We present a novel paradigm to address the STTD learning problem by parameterizing STTD as an implicit neural representation. We validate its effectiveness through extensive experiments in real-world scenarios, showcasing applications from corridor to network scales.
  arXiv  Detail & Related papers  (2024-06-13T02:03:22Z)
• Spatiotemporal Implicit Neural Representation as a Generalized Traffic Data Learner [46.866240648471894]
  Spatiotemporal Traffic Data (STTD) measures the complex dynamical behaviors of the multiscale transportation system. We present a novel paradigm to address the STTD learning problem by parameterizing STTD as an implicit neural representation. We validate its effectiveness through extensive experiments in real-world scenarios, showcasing applications from corridor to network scales.
  arXiv  Detail & Related papers  (2024-05-06T06:23:06Z)
• In-Context Convergence of Transformers [63.04956160537308]
  We study the learning dynamics of a one-layer transformer with softmax attention trained via gradient descent. For data with imbalanced features, we show that the learning dynamics take a stage-wise convergence process.
  arXiv  Detail & Related papers  (2023-10-08T17:55:33Z)
• Internal Representations of Vision Models Through the Lens of Frames on Data Manifolds [8.67467876089153]
  We present a new approach to studying such representations inspired by the idea of a frame on the tangent bundle of a manifold. Our construction, which we call a neural frame, is formed by assembling a set of vectors representing specific types of perturbations of a data point. Using neural frames, we make observations about the way that models process, layer-by-layer, specific modes of variation within a small neighborhood of a datapoint.
  arXiv  Detail & Related papers  (2022-11-19T01:48:19Z)
• Designing Universal Causal Deep Learning Models: The Case of Infinite-Dimensional Dynamical Systems from Stochastic Analysis [3.5450828190071655]
  Causal operators (COs) play a central role in contemporary analysis. There is still no canonical framework for designing Deep Learning (DL) models capable of approximating COs. This paper proposes a "geometry-aware" solution to this open problem by introducing a DL model-design framework.
  arXiv  Detail & Related papers  (2022-10-24T14:43:03Z)
• Linear Connectivity Reveals Generalization Strategies [54.947772002394736]
  Some pairs of finetuned models have large barriers of increasing loss on the linear paths between them. We find distinct clusters of models which are linearly connected on the test loss surface, but are disconnected from models outside the cluster. Our work demonstrates how the geometry of the loss surface can guide models towards different functions.
  arXiv  Detail & Related papers  (2022-05-24T23:43:02Z)
• Dynamically-Scaled Deep Canonical Correlation Analysis [77.34726150561087]
  Canonical Correlation Analysis (CCA) is a method for feature extraction of two views by finding maximally correlated linear projections of them. We introduce a novel dynamic scaling method for training an input-dependent canonical correlation model.
  arXiv  Detail & Related papers  (2022-03-23T12:52:49Z)
• A Bayesian Perspective on Training Speed and Model Selection [51.15664724311443]
  We show that a measure of a model's training speed can be used to estimate its marginal likelihood. We verify our results in model selection tasks for linear models and for the infinite-width limit of deep neural networks. Our results suggest a promising new direction towards explaining why neural networks trained with gradient descent are biased towards functions that generalize well.
  arXiv  Detail & Related papers  (2020-10-27T17:56:14Z)
• Kernel and Rich Regimes in Overparametrized Models [69.40899443842443]
  We show that gradient descent on overparametrized multilayer networks can induce rich implicit biases that are not RKHS norms. We also demonstrate this transition empirically for more complex matrix factorization models and multilayer non-linear networks.
  arXiv  Detail & Related papers  (2020-02-20T15:43:02Z)

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.