Pith. sign in

REVIEW 6 cited by

On Provable Length and Compositional Generalization

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2402.04875 v6 pith:7MXNVDTE submitted 2024-02-07 cs.LG cs.CLstat.ML

classification cs.LGcs.CLstat.ML
keywords generalizationlengthtrainingcapacitycompositionallimitedarchitecturesdifferent
verification ladder T0 review T1 audit T2 compute T3 formal

Signed reviews

No signed human review yet.

0 comments
read the original abstract

Out-of-distribution generalization capabilities of sequence-to-sequence models can be studied from the lens of two crucial forms of generalization: length generalization -- the ability to generalize to longer sequences than ones seen during training, and compositional generalization: the ability to generalize to token combinations not seen during training. In this work, we provide first provable guarantees on length and compositional generalization for common sequence-to-sequence models -- deep sets, transformers, state space models, and recurrent neural nets -- trained to minimize the prediction error. We show that \emph{limited capacity} versions of these different architectures achieve both length and compositional generalization provided the training distribution is sufficiently diverse. In the first part, we study structured limited capacity variants of different architectures and arrive at the generalization guarantees with limited diversity requirements on the training distribution. In the second part, we study limited capacity variants with less structural assumptions and arrive at generalization guarantees but with more diversity requirements on the training distribution. Further, we also show that chain-of-thought supervision enables length generalization in higher capacity counterparts of the different architectures we study.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 6 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Universal priors: solving empirical Bayes via Bayesian inference and pretraining

    stat.ML 2026-02 conditional novelty 8.0 of 10

    A simple random prior-on-prior lets pretrained transformers achieve near-optimal empirical Bayes regret uniformly over all test priors, and length generalization matches α-posterior inference.

  2. When Does In-Context Search Help? A Sampling-Complexity Theory of Reflection-Driven Reasoning

    cs.AI 2026-07 conditional novelty 7.0 of 10

    When reflections localize early errors, in-context search solves exp-small pass-rate problems with poly sequential attempts; otherwise it offers no asymptotic gain over parallel sampling, and the update is learnable a...

  3. PatchGen: Learning Soft Intra-Image Predictive Subsets for Visual Generalization

    cs.CV 2026-08 conditional novelty 6.0 of 10

    PatchGen learns a sample-dependent soft mask that selects label-predictive image patches, improving visual generalization across domain, category, and combined shifts.

  4. From Isolation to Entanglement: When Do Interpretability Methods Identify and Disentangle Known Concepts?

    cs.LG 2025-12 conditional novelty 6.0 of 10

    Correlational disentanglement scores do not guarantee that steering a feature changes only its target concept; current featurizers entangle concepts in their causal effects.

  5. Extrapolation by Association: Length Generalization Transfer in Transformers

    cs.CL 2025-06 conditional novelty 5.0 of 10

    Length generalization on a short-trained main task can be inherited from a longer-trained related auxiliary task trained jointly with it.

  6. Saving for the future: Enhancing generalization via partial logic regularization

    cs.LG 2025-08 reject novelty 4.0 of 10

    PL-Reg adds a trainable mask and a defined/undefined classification loss to logic-based regularization, improving unknown-class accuracy across GCD, mDG+GCD, and CIL benchmarks.

Pith tools