Why bother with geometry? On the relevance of linear decompositions of Transformer embeddings

• URL: http://arxiv.org/abs/2310.06977v1
• Date: Tue, 10 Oct 2023 19:56:10 GMT
• Title: Why bother with geometry? On the relevance of linear decompositions of Transformer embeddings
• Authors: Timothee Mickus, Ra\'ul V\'azquez
• Abstract summary: We study representations from machine-translation decoders using two of such embedding decomposition methods. Our results indicate that, while decomposition-derived indicators effectively correlate with model performance, variation across different runs suggests a more nuanced take on this question.
• Score: 5.151529346168568
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: A recent body of work has demonstrated that Transformer embeddings can be linearly decomposed into well-defined sums of factors, that can in turn be related to specific network inputs or components. There is however still a dearth of work studying whether these mathematical reformulations are empirically meaningful. In the present work, we study representations from machine-translation decoders using two of such embedding decomposition methods. Our results indicate that, while decomposition-derived indicators effectively correlate with model performance, variation across different runs suggests a more nuanced take on this question. The high variability of our measurements indicate that geometry reflects model-specific characteristics more than it does sentence-specific computations, and that similar training conditions do not guarantee similar vector spaces.

Related papers

• Knowledge Composition using Task Vectors with Learned Anisotropic Scaling [51.4661186662329]
  We introduce aTLAS, an algorithm that linearly combines parameter blocks with different learned coefficients, resulting in anisotropic scaling at the task vector level. We show that such linear combinations explicitly exploit the low intrinsicity of pre-trained models, with only a few coefficients being the learnable parameters. We demonstrate the effectiveness of our method in task arithmetic, few-shot recognition and test-time adaptation, with supervised or unsupervised objectives.
  arXiv  Detail & Related papers  (2024-07-03T07:54:08Z)
• Optimal Matrix-Mimetic Tensor Algebras via Variable Projection [0.0]
  Matrix mimeticity arises from interpreting tensors as operators that can be multiplied, factorized, and analyzed analogous to matrices. We learn optimal linear mappings and corresponding tensor representations without relying on prior knowledge of the data. We provide original theory of uniqueness of the transformation and convergence analysis of our variable-projection-based algorithm.
  arXiv  Detail & Related papers  (2024-06-11T04:52:23Z)
• When can transformers reason with abstract symbols? [25.63285482210457]
  We prove that for any relational reasoning task in a large family of tasks, transformers learn the abstract relations and generalize to the test set. This is in contrast to classical fully-connected networks, which we prove fail to learn to reason.
  arXiv  Detail & Related papers  (2023-10-15T06:45:38Z)
• From Bricks to Bridges: Product of Invariances to Enhance Latent Space Communication [19.336940758147442]
  It has been observed that representations learned by distinct neural networks conceal structural similarities when the models are trained under similar inductive biases. We introduce a versatile method to directly incorporate a set of invariances into the representations, constructing a product space of invariant components on top of the latent representations. We validate our solution on classification and reconstruction tasks, observing consistent latent similarity and downstream performance improvements in a zero-shot stitching setting.
  arXiv  Detail & Related papers  (2023-10-02T13:55:38Z)
• Equivariant Disentangled Transformation for Domain Generalization under Combination Shift [91.38796390449504]
  Combinations of domains and labels are not observed during training but appear in the test environment. We provide a unique formulation of the combination shift problem based on the concepts of homomorphism, equivariance, and a refined definition of disentanglement.
  arXiv  Detail & Related papers  (2022-08-03T12:31:31Z)
• Curvature-informed multi-task learning for graph networks [56.155331323304]
  State-of-the-art graph neural networks attempt to predict multiple properties simultaneously. We investigate a potential explanation for this phenomenon: the curvature of each property's loss surface significantly varies, leading to inefficient learning.
  arXiv  Detail & Related papers  (2022-08-02T18:18:41Z)
• 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)
• Tensor Component Analysis for Interpreting the Latent Space of GANs [41.020230946351816]
  This paper addresses the problem of finding interpretable directions in the latent space of pre-trained Generative Adversarial Networks (GANs) Our scheme allows for both linear edits corresponding to the individual modes of the tensor, and non-linear ones that model the multiplicative interactions between them. We show experimentally that we can utilise the former to better separate style- from geometry-based transformations, and the latter to generate an extended set of possible transformations.
  arXiv  Detail & Related papers  (2021-11-23T09:14:39Z)
• Equivariant vector field network for many-body system modeling [65.22203086172019]
  Equivariant Vector Field Network (EVFN) is built on a novel equivariant basis and the associated scalarization and vectorization layers. We evaluate our method on predicting trajectories of simulated Newton mechanics systems with both full and partially observed data.
  arXiv  Detail & Related papers  (2021-10-26T14:26:25Z)
• A Differential Geometry Perspective on Orthogonal Recurrent Models [56.09491978954866]
  We employ tools and insights from differential geometry to offer a novel perspective on orthogonal RNNs. We show that orthogonal RNNs may be viewed as optimizing in the space of divergence-free vector fields. Motivated by this observation, we study a new recurrent model, which spans the entire space of vector fields.
  arXiv  Detail & Related papers  (2021-02-18T19:39:22Z)
• Invertible Generative Modeling using Linear Rational Splines [11.510009152620666]
  Normalizing flows attempt to model an arbitrary probability distribution through a set of invertible mappings. The first flow designs used coupling layer mappings built upon affine transformations. Intrepid piecewise functions as a replacement for affine transformations have attracted attention.
  arXiv  Detail & Related papers  (2020-01-15T08:05:55Z)

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.