Implicit Geometry of Next-token Prediction: From Language Sparsity Patterns to Model Representations
- URL: http://arxiv.org/abs/2408.15417v1
- Date: Tue, 27 Aug 2024 21:46:47 GMT
- Title: Implicit Geometry of Next-token Prediction: From Language Sparsity Patterns to Model Representations
- Authors: Yize Zhao, Tina Behnia, Vala Vakilian, Christos Thrampoulidis,
- Abstract summary: Next-token prediction (NTP) over large text corpora has become the go-to paradigm to train large language models.
We look at how NTP influences the mapping of linguistic patterns to geometric properties of the resulting model representations.
We validate our findings on synthetic and small-scale real language datasets.
- Score: 24.211603400355756
- License: http://creativecommons.org/licenses/by/4.0/
- Abstract: Next-token prediction (NTP) over large text corpora has become the go-to paradigm to train large language models. Yet, it remains unclear how NTP influences the mapping of linguistic patterns to geometric properties of the resulting model representations. We frame training of large language models as soft-label classification over sparse probabilistic label vectors, coupled with an analytical approximation that allows unrestricted generation of context embeddings. This approach links NTP training to rank-constrained, nuclear-norm regularized optimization in the logit domain, offering a framework for analyzing the geometry of word and context embeddings. In large embedding spaces, we find that NTP implicitly favors learning logits with a sparse plus low-rank structure. While the sparse component captures the co-occurrence frequency of context-word pairs, the orthogonal low-rank component, which becomes dominant as training progresses, depends solely on the sparsity pattern of the co-occurrence matrix. Consequently, when projected onto an appropriate subspace, representations of contexts that are followed by the same set of next-tokens collapse, a phenomenon we term subspace-collapse. We validate our findings on synthetic and small-scale real language datasets. Finally, we outline potential research directions aimed at deepening the understanding of NTP's influence on the learning of linguistic patterns and regularities.
Related papers
- Towards a theory of how the structure of language is acquired by deep neural networks [6.363756171493383]
We use a tree-like generative model that captures many of the hierarchical structures found in natural languages.
We show that token-token correlations can be used to build a representation of the grammar's hidden variables.
We conjecture that the relationship between training set size and effective range of correlations holds beyond our synthetic datasets.
arXiv Detail & Related papers (2024-05-28T17:01:22Z) - Making Pre-trained Language Models Great on Tabular Prediction [50.70574370855663]
The transferability of deep neural networks (DNNs) has made significant progress in image and language processing.
We present TP-BERTa, a specifically pre-trained LM for tabular data prediction.
A novel relative magnitude tokenization converts scalar numerical feature values to finely discrete, high-dimensional tokens, and an intra-feature attention approach integrates feature values with the corresponding feature names.
arXiv Detail & Related papers (2024-03-04T08:38:56Z) - Implicit Optimization Bias of Next-Token Prediction in Linear Models [32.2896512612788]
Next-token prediction (NTP) is the dominant training paradigm for modern language models.
We study the structural properties of the solutions selected by gradient-based generalizations.
arXiv Detail & Related papers (2024-02-28T18:34:53Z) - Constructing Word-Context-Coupled Space Aligned with Associative
Knowledge Relations for Interpretable Language Modeling [0.0]
The black-box structure of the deep neural network in pre-trained language models seriously limits the interpretability of the language modeling process.
A Word-Context-Coupled Space (W2CSpace) is proposed by introducing the alignment processing between uninterpretable neural representation and interpretable statistical logic.
Our language model can achieve better performance and highly credible interpretable ability compared to related state-of-the-art methods.
arXiv Detail & Related papers (2023-05-19T09:26:02Z) - Prototype-based Embedding Network for Scene Graph Generation [105.97836135784794]
Current Scene Graph Generation (SGG) methods explore contextual information to predict relationships among entity pairs.
Due to the diverse visual appearance of numerous possible subject-object combinations, there is a large intra-class variation within each predicate category.
Prototype-based Embedding Network (PE-Net) models entities/predicates with prototype-aligned compact and distinctive representations.
PL is introduced to help PE-Net efficiently learn such entitypredicate matching, and Prototype Regularization (PR) is devised to relieve the ambiguous entity-predicate matching.
arXiv Detail & Related papers (2023-03-13T13:30:59Z) - Better Language Model with Hypernym Class Prediction [101.8517004687825]
Class-based language models (LMs) have been long devised to address context sparsity in $n$-gram LMs.
In this study, we revisit this approach in the context of neural LMs.
arXiv Detail & Related papers (2022-03-21T01:16:44Z) - The Low-Dimensional Linear Geometry of Contextualized Word
Representations [27.50785941238007]
We study the linear geometry of contextualized word representations in ELMO and BERT.
We show that a variety of linguistic features are encoded in low-dimensional subspaces.
arXiv Detail & Related papers (2021-05-15T00:58:08Z) - Rethinking Relational Encoding in Language Model: Pre-Training for
General Sequences [23.806325599416134]
Language model pre-training fails at modeling per-sequence relations in non-natural language domains.
We develop a framework that couples LMPT with deep structure-preserving metric learning to produce richer embeddings.
Our approach offers notable performance improvements on downstream tasks.
arXiv Detail & Related papers (2021-03-18T15:51:04Z) - Infusing Finetuning with Semantic Dependencies [62.37697048781823]
We show that, unlike syntax, semantics is not brought to the surface by today's pretrained models.
We then use convolutional graph encoders to explicitly incorporate semantic parses into task-specific finetuning.
arXiv Detail & Related papers (2020-12-10T01:27:24Z) - Explaining and Improving Model Behavior with k Nearest Neighbor
Representations [107.24850861390196]
We propose using k nearest neighbor representations to identify training examples responsible for a model's predictions.
We show that kNN representations are effective at uncovering learned spurious associations.
Our results indicate that the kNN approach makes the finetuned model more robust to adversarial inputs.
arXiv Detail & Related papers (2020-10-18T16:55:25Z) - Learning Reasoning Strategies in End-to-End Differentiable Proving [50.9791149533921]
Conditional Theorem Provers learn optimal rule selection strategy via gradient-based optimisation.
We show that Conditional Theorem Provers are scalable and yield state-of-the-art results on the CLUTRR dataset.
arXiv Detail & Related papers (2020-07-13T16:22:14Z)
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.