CORTEXA
← Browse
arxivcs.CLcs.AI2026-07-08

Riemannian Geometry for Pre-trained Language Model Embeddings

Szczepan Konior, Alexandre Quemy, Przemysław Klocek, Bartłomiej Sobieski, Grégoire Cattan

Understanding the geometric structure of pre-trained language model embeddings matters for interpretability and safety. We ask whether sentence-level classification signal lives in the Riemannian geometry of contextual token embeddings, and probe it by extracting per-token pullback metrics from a learned encoder's analytical Jacobian and aggregating them with the Fréchet mean on the symmetric positive definite (SPD) manifold; we call this procedure Riemannian Mean Pooling (RMP). Across three datasets with non-trivial linguistic structure (CoLA, CREAK, RTE), RMP outperforms Euclidean mean pooling, while on FEVER-Symmetric, a benchmark constructed to remove annotation-driven lexical artifacts, the method correctly stays at chance. Ablations show that a randomly initialised encoder combined with Fréchet aggregation already beats Euclidean pooling on two of the three signal-bearing datasets, localising the source of the gain to the geometric aggregation rather than to learned manifold structure; the trained encoder contributes additional signal specifically on CREAK, the most knowledge-heavy of the three signal-bearing datasets.

View free PDFSource page

Related papers

arxivcs.CLcs.AIcs.LG2026-07-16

In-Place Tokenizer Expansion for Pre-trained LLMs

Jimmy T. H. Smith, Tarek Dakhran, Alberto Cabrera, Simon S. Lee, Paul Pak, Aditya Tadimeti, et al.

A tokenizer fixed at the start of pre-training allocates vocabulary in proportion to the pre-training corpus, reflecting the deployment priorities at that time. When those priorities shift, languages added later are split into many more tokens per word, which can raise latency, c…

View free PDFSource page
arxivcs.CLcs.AIcs.PL2026-07-09

TypeProbe: Recovering Type Representations from Hidden States of Pre-trained Code Models

Giuliano Gorgone, Fausto Carcassi

State-of-the-art code models achieve impressive performance, yet the extent to which they internally encode type information remains poorly understood. We probe the residual streams of pretrained code models for internal type representations using a parallel dataset of Java and P…

View free PDFSource page
arxiveess.AScs.AIcs.CLcs.LGcs.SD2026-07-14

The Sound of Absence: Audio-Language Embedding Models Struggle with Negation

Chun-Yi Kuan, Hung-yi Lee

Audio-language embedding models such as CLAP are widely evaluated on matching present sound events, but rarely on negation. We show this affirmation-only evaluation hides a key limitation: these models fail to encode negated sound concepts, mapping affirmative and negated caption…

View free PDFSource page
arxivcs.LGcs.AIcs.CL2026-07-01

CausalMix: Data Mixture as Causal Inference for Language Model Training

Zinan Tang, Yukun Zhang, Shaomian Zheng, Zhuoshi Pan, Qizhi Pei, Dingnan Jin, et al.

In Large Language Model (LLM) training, data mixing plays a pivotal role in determining model performance. Recent methods optimize mixture weights via proxy models, but they rely on the assumption of static data distributions. As a result, when the underlying data pool shifts, th…

View free PDFSource page
arxivcs.CLcs.AI2026-07-09

UltraX: Refining Pre-Training Data at Scale with Adaptive Programmatic Editing

Xinlong Zhao, Dongsheng Liu, Hengyu Zhao, Zixuan Fu, Zheng Wang, Jie Cai, et al.

As available training data approaches its physical limit, gains from Scaling Laws have begun to diminish. Consequently, improving Large Language Models (LLMs) now depends less on data expansion and more on higher-quality data utilization. However, in the context of large-scale co…

View free PDFSource page
arxivcs.LGcs.AIcs.CL2026-07-02

HERMES: A Multi-Granularity Labeling Substrate for Pre-training Data Mixtures

Ziyun Qiao, Yue Min, Ruining Chen, Yujun Li

Most data-mixing methods assume the corpus has already been partitioned into groups, and the choice of those groups determines what a mixer can express. Existing labels, including provenance, topic or format taxonomies, and flat embedding clusters, commit to one semantic axis at…

View free PDFSource page