Position as Probability: Self-Supervised Transformers that Think Past Their Training for Length Extrapolation

• URL: http://arxiv.org/abs/2506.00920v1
• Date: Sun, 01 Jun 2025 09:20:44 GMT
• Title: Position as Probability: Self-Supervised Transformers that Think Past Their Training for Length Extrapolation
• Authors: Philip Heejun Lee,
• Abstract summary: PRISM is a novel positional encoding mechanism that enables Transformers to extrapolate accurately up to 10x beyond their training length.<n>Our analysis demonstrates that PRISM's positional encoding maintains sharp and interpretable internal states, providing a theoretical basis for reliable length generalization.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Deep sequence models typically degrade in accuracy when test sequences significantly exceed their training lengths, yet many critical tasks--such as algorithmic reasoning, multi-step arithmetic, and compositional generalization--require robust length extrapolation. We introduce PRISM, a Probabilistic Relative-position Implicit Superposition Model, a novel positional encoding mechanism that enables Transformers to extrapolate accurately up to 10x beyond their training length. PRISM learns continuous relative positions through a differentiable histogram-filter update, preserving position uncertainty via a probabilistic superposition rather than conventional deterministic embeddings. Empirically, PRISM achieves state-of-the-art length extrapolation, successfully generalizing to previously intractable sequence lengths across algorithmic benchmarks--including arithmetic (addition, multiplication), SCAN compositionality tasks, and complex copy variants derived from DeepMind's recent datasets. Our analysis demonstrates that PRISM's stochastic positional encoding maintains sharp and interpretable internal states, providing a theoretical basis for reliable length generalization. These results advance the goal of neural sequence models that remain algorithmically robust at lengths far exceeding their training horizon.

Related papers

• Transformers Are Universally Consistent [14.904264782690639]
  We show that Transformers equipped with softmax-based nonlinear attention are uniformly consistent when tasked with executing Least Squares regression.<n>We derive upper bounds on the empirical error which, in the regime, decay at a provable rate of $mathcalO(t-1/2d)$, where $t$ denotes the number of input tokens and $d$ the embedding dimensionality.
  arXiv  Detail & Related papers  (2025-05-30T12:39:26Z)
• The Role of Sparsity for Length Generalization in Transformers [58.65997625433689]
  We propose a new theoretical framework to study length generalization for the next-token prediction task.<n>We show that length generalization occurs as long as each predicted token depends on a small (fixed) number of previous tokens.<n>We introduce Predictive Position Coupling, which trains the transformer to predict the position IDs used in a positional coupling approach.
  arXiv  Detail & Related papers  (2025-02-24T03:01:03Z)
• Prompting a Pretrained Transformer Can Be a Universal Approximator [105.59562522323274]
  We show that much smaller pretrained models than previously thought can be universal approximators when prefixed. We also offer Jackson-type bounds on the length of the prefix needed to approximate a function to a desired precision.
  arXiv  Detail & Related papers  (2024-02-22T18:12:48Z)
• Randomized Positional Encodings Boost Length Generalization of Transformers [14.814408238614165]
  Transformers have impressive generalization capabilities on tasks with a fixed context length. They fail to generalize to sequences of arbitrary length, even for seemingly simple tasks such as duplicating a string. We introduce a novel family of positional encodings that can overcome this problem.
  arXiv  Detail & Related papers  (2023-05-26T11:47:52Z)
• DBA: Efficient Transformer with Dynamic Bilinear Low-Rank Attention [53.02648818164273]
  We present an efficient yet effective attention mechanism, namely the Dynamic Bilinear Low-Rank Attention (DBA) DBA compresses the sequence length by input-sensitive dynamic projection matrices and achieves linear time and space complexity. Experiments over tasks with diverse sequence length conditions show that DBA achieves state-of-the-art performance.
  arXiv  Detail & Related papers  (2022-11-24T03:06:36Z)
• Alleviate Exposure Bias in Sequence Prediction \\ with Recurrent Neural Networks [47.52214243454995]
  A popular strategy to train recurrent neural networks (RNNs) is to take the ground truth as input at each time step. We propose a fully differentiable training algorithm for RNNs to better capture long-term dependencies.
  arXiv  Detail & Related papers  (2021-03-22T06:15:22Z)
• MLE-guided parameter search for task loss minimization in neural sequence modeling [83.83249536279239]
  Neural autoregressive sequence models are used to generate sequences in a variety of natural language processing (NLP) tasks. We propose maximum likelihood guided parameter search (MGS), which samples from a distribution over update directions that is a mixture of random search around the current parameters and around the maximum likelihood gradient. Our experiments show that MGS is capable of optimizing sequence-level losses, with substantial reductions in repetition and non-termination in sequence completion, and similar improvements to those of minimum risk training in machine translation.
  arXiv  Detail & Related papers  (2020-06-04T22:21:22Z)
• Scalable Uncertainty for Computer Vision with Functional Variational Inference [18.492485304537134]
  We leverage the formulation of variational inference in function space. We obtain predictive uncertainty estimates at the cost of a single forward pass through any chosen CNN architecture. We propose numerically efficient algorithms which enable fast training in the context of high-dimensional tasks.
  arXiv  Detail & Related papers  (2020-03-06T19:09:42Z)
• Consistency of a Recurrent Language Model With Respect to Incomplete Decoding [67.54760086239514]
  We study the issue of receiving infinite-length sequences from a recurrent language model. We propose two remedies which address inconsistency: consistent variants of top-k and nucleus sampling, and a self-terminating recurrent language model.
  arXiv  Detail & Related papers  (2020-02-06T19:56:15Z)

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.