Loss Landscape Degeneracy Drives Stagewise Development in Transformers

• URL: http://arxiv.org/abs/2402.02364v2
• Date: Thu, 13 Feb 2025 07:29:46 GMT
• Title: Loss Landscape Degeneracy Drives Stagewise Development in Transformers
• Authors: Jesse Hoogland, George Wang, Matthew Farrugia-Roberts, Liam Carroll, Susan Wei, Daniel Murfet,
• Abstract summary: We show that training can be divided into distinct periods of change in loss landscape degeneracy.<n>This finding underscores the potential of a degeneracy-based perspective for understanding modern deep learning.
• Score: 1.947473271879451
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Deep learning involves navigating a high-dimensional loss landscape over the neural network parameter space. Over the course of training, complex computational structures form and re-form inside the neural network, leading to shifts in input/output behavior. It is a priority for the science of deep learning to uncover principles governing the development of neural network structure and behavior. Drawing on the framework of singular learning theory, we propose that model development is deeply linked to degeneracy in the local geometry of the loss landscape. We investigate this link by monitoring loss landscape degeneracy throughout training, as quantified by the local learning coefficient, for a transformer language model and an in-context linear regression transformer. We show that training can be divided into distinct periods of change in loss landscape degeneracy, and that these changes in degeneracy coincide with significant changes in the internal computational structure and the input/output behavior of the transformers. This finding underscores the potential of a degeneracy-based perspective for understanding modern deep learning.

Related papers

• Dynamics of Transient Structure in In-Context Linear Regression Transformers [0.5242869847419834]
  We show that when transformers are trained on in-context linear regression tasks with intermediate task diversity, they behave like ridge regression before specializing to the tasks in their training distribution. This transition from a general solution to a specialized solution is revealed by joint trajectory principal component analysis. We empirically validate this explanation by measuring the model complexity of our transformers as defined by the local learning coefficient.
  arXiv  Detail & Related papers  (2025-01-29T16:32:14Z)
• Gradient Localization Improves Lifelong Pretraining of Language Models [32.29298047707914]
  Large Language Models (LLMs) trained on web-scale text corpora have been shown to capture world knowledge in their parameters. In this work, we examine two types of knowledge relating to temporally sensitive entities and demonstrate that each type is localized to different sets of parameters within the LLMs.
  arXiv  Detail & Related papers  (2024-11-07T05:43:50Z)
• Toward Understanding In-context vs. In-weight Learning [50.24035812301655]
  We identify simplified distributional properties that give rise to the emergence and disappearance of in-context learning. We then extend the study to a full large language model, showing how fine-tuning on various collections of natural language prompts can elicit similar in-context and in-weight learning behaviour.
  arXiv  Detail & Related papers  (2024-10-30T14:09:00Z)
• Dynamical loss functions shape landscape topography and improve learning in artificial neural networks [0.9208007322096533]
  We show how to transform cross-entropy and mean squared error into dynamical loss functions. We show how they significantly improve validation accuracy for networks of varying sizes.
  arXiv  Detail & Related papers  (2024-10-14T16:27:03Z)
• Differentiation and Specialization of Attention Heads via the Refined Local Learning Coefficient [0.49478969093606673]
  We introduce refined variants of the Local Learning Coefficient (LLC), a measure of model complexity grounded in singular learning theory. We study the development of internal structure in transformer language models during training.
  arXiv  Detail & Related papers  (2024-10-03T20:51:02Z)
• Dynamical stability and chaos in artificial neural network trajectories along training [3.379574469735166]
  We study the dynamical properties of this process by analyzing through this lens the network trajectories of a shallow neural network. We find hints of regular and chaotic behavior depending on the learning rate regime. This work also contributes to the cross-fertilization of ideas between dynamical systems theory, network theory and machine learning.
  arXiv  Detail & Related papers  (2024-04-08T17:33:11Z)
• 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)
• Procedural Text Mining with Large Language Models [0.21756081703275998]
  We tackle the problem of extracting procedures from unstructured PDF text in an incremental question-answering fashion. We leverage the current state-of-the-art GPT-4 (Generative Pre-trained Transformer 4) model, accompanied by two variations of in-context learning. The findings highlight both the promise of this approach and the value of the in-context learning customisations.
  arXiv  Detail & Related papers  (2023-10-05T08:27:33Z)
• Emergent learning in physical systems as feedback-based aging in a glassy landscape [0.0]
  We show that the learning dynamics resembles an aging process, where the system relaxes in response to repeated application of the feedback boundary forces. We also observe that the square root of the mean-squared error as a function of epoch takes on a non-exponential form, which is a typical feature of glassy systems.
  arXiv  Detail & Related papers  (2023-09-08T15:24:55Z)
• Unsupervised Learning of Invariance Transformations [105.54048699217668]
  We develop an algorithmic framework for finding approximate graph automorphisms. We discuss how this framework can be used to find approximate automorphisms in weighted graphs in general.
  arXiv  Detail & Related papers  (2023-07-24T17:03:28Z)
• Centered Self-Attention Layers [89.21791761168032]
  The self-attention mechanism in transformers and the message-passing mechanism in graph neural networks are repeatedly applied. We show that this application inevitably leads to oversmoothing, i.e., to similar representations at the deeper layers. We present a correction term to the aggregating operator of these mechanisms.
  arXiv  Detail & Related papers  (2023-06-02T15:19:08Z)
• Latent Traversals in Generative Models as Potential Flows [113.4232528843775]
  We propose to model latent structures with a learned dynamic potential landscape. Inspired by physics, optimal transport, and neuroscience, these potential landscapes are learned as physically realistic partial differential equations. Our method achieves both more qualitatively and quantitatively disentangled trajectories than state-of-the-art baselines.
  arXiv  Detail & Related papers  (2023-04-25T15:53:45Z)
• Transformers learn in-context by gradient descent [58.24152335931036]
  Training Transformers on auto-regressive objectives is closely related to gradient-based meta-learning formulations. We show how trained Transformers become mesa-optimizers i.e. learn models by gradient descent in their forward pass.
  arXiv  Detail & Related papers  (2022-12-15T09:21:21Z)
• The Geometry of Self-supervised Learning Models and its Impact on Transfer Learning [62.601681746034956]
  Self-supervised learning (SSL) has emerged as a desirable paradigm in computer vision. We propose a data-driven geometric strategy to analyze different SSL models using local neighborhoods in the feature space induced by each.
  arXiv  Detail & Related papers  (2022-09-18T18:15:38Z)
• Data-driven emergence of convolutional structure in neural networks [83.4920717252233]
  We show how fully-connected neural networks solving a discrimination task can learn a convolutional structure directly from their inputs. By carefully designing data models, we show that the emergence of this pattern is triggered by the non-Gaussian, higher-order local structure of the inputs.
  arXiv  Detail & Related papers  (2022-02-01T17:11:13Z)
• Transformers Solve the Limited Receptive Field for Monocular Depth Prediction [82.90445525977904]
  We propose TransDepth, an architecture which benefits from both convolutional neural networks and transformers. This is the first paper which applies transformers into pixel-wise prediction problems involving continuous labels.
  arXiv  Detail & Related papers  (2021-03-22T18:00:13Z)
• A Survey on Deep Semi-supervised Learning [51.26862262550445]
  We first present a taxonomy for deep semi-supervised learning that categorizes existing methods. We then offer a detailed comparison of these methods in terms of the type of losses, contributions, and architecture differences.
  arXiv  Detail & Related papers  (2021-02-28T16:22:58Z)
• Going beyond p-convolutions to learn grayscale morphological operators [64.38361575778237]
  We present two new morphological layers based on the same principle as the p-convolutional layer. In this work, we present two new morphological layers based on the same principle as the p-convolutional layer.
  arXiv  Detail & Related papers  (2021-02-19T17:22:16Z)
• Introducing Orthogonal Constraint in Structural Probes [0.2538209532048867]
  We decompose a linear projection of language vector space into isomorphic space rotation and linear scaling directions. We experimentally show that our approach can be performed in a multitask setting.
  arXiv  Detail & Related papers  (2020-12-30T17:14:25Z)
• APo-VAE: Text Generation in Hyperbolic Space [116.11974607497986]
  In this paper, we investigate text generation in a hyperbolic latent space to learn continuous hierarchical representations. An Adrial Poincare Variversaational Autoencoder (APo-VAE) is presented, where both the prior and variational posterior of latent variables are defined over a Poincare ball via wrapped normal distributions. Experiments in language modeling and dialog-response generation tasks demonstrate the winning effectiveness of the proposed APo-VAE model.
  arXiv  Detail & Related papers  (2020-04-30T19:05:41Z)

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.