arXiv:2602. 15438v3 Announce Type: replace Abstract: For a broad family of discriminative models that includes autoregressive language models, identifiability results imply that if two models induce the same conditional distributions, then their internal representations are equal up to an invertible linear transformation.
arXiv:2606. 02765v1 Announce Type: cross Abstract: Model dimension ($d_{model}$) is a fundamental hyperparameter in transformer language models, yet its role in setting the geometric limits of feature representation remains under-explored.
arXiv:2506. 08774v2 Announce Type: replace-cross Abstract: Different machine learning models can represent the same underlying concept in different ways.
The paper extends mechanistic interpretability of large language models by modeling concepts as low‑dimensional non‑linear manifolds rather than linear subspaces. It introduces a concept‑based alignment (CBA) score to compare these manifolds across layers and models, revealing block structures in intermediate layers, a shift from syntax‑dominated to mixed syntactic‑semantic concepts, and training‑dependent multilingual sharing. The study also shows that alignment patterns differ across model families and training stages, with adjacent stages aligning more closely than distant ones.
The paper introduces a flow matching framework that unifies various neural representational dissimilarity metrics under a single theoretical umbrella. By interpreting these metrics as Jeffreys divergences with different velocity constraints, the authors demonstrate that flow matching improves distance estimation for complex distributions and continuous variables. The framework also facilitates the principled design of new dissimilarity measures.
arXiv:2608. 15215v1 Announce Type: cross Abstract: Token-level knowledge distillation (KD) matches two conditional distributions per position, yet the standard objectives compare them pointwise: a Kullback-Leibler gradient is blind to which wrong token receives probability mass.
arXiv:2607. 09405v1 Announce Type: new Abstract: Similarity search is a primary application of embedding models trained by contrastive learning.
arXiv:2608. 01263v1 Announce Type: new Abstract: On-policy distillation (OPD) samples trajectories from the current student policy and minimizes token-level divergence between student and teacher next-token distributions at prefixes along those trajectories.
The Platonic Representation Hypothesis (PRH) claims that independently trained models converge on a shared statistical model of reality, yet recent work finds only weak pointwise similarity between mo...
arXiv:2607. 28667v1 Announce Type: cross Abstract: Recent work has shown that classifying large language models (LLMs)' responses can be distinguished by modeling token embeddings as trajectories of a black-box dynamical system (DS) and comparing prediction residuals of two DSs.
arXiv:2409. 10094v3 Announce Type: replace-cross Abstract: Out-of-Distribution (OoD) detection aims to justify whether a given sample is from the training distribution of the classifier-under-protection, i.
arXiv:2503. 08038v2 Announce Type: replace-cross Abstract: In this paper, we delve deeper into the Kullback-Leibler (KL) Divergence loss and mathematically prove that it is equivalent to the Decoupled Kullback-Leibler (DKL) Divergence loss that consists of (1) a weighted Mean Square Error (wMSE) loss and (2) a Cross-Entropy loss incorporating soft labels.
Cross-modal knowledge distillation (CMKD) studies how a (large) teacher model trained on one type of data (e. g.