Where Computation Lives Inside TabPFN: Causal Localisation of Attention Head Function

• URL: http://arxiv.org/abs/2606.12917v1
• Date: Thu, 11 Jun 2026 05:21:08 GMT
• Title: Where Computation Lives Inside TabPFN: Causal Localisation of Attention Head Function
• Abstract summary: We find that one head's causal necessity dominates that of the others by 2 to 5 times at peak layer.<n>We attribute this result to TabPFN's in context learning mechanism, which encodes task structure through context dependent attention.
• Score: 7.771378647684901
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: We present the first causal mechanistic analysis of a tabular foundation model, investigating how TabPFN 2.5's feature wise attention heads distribute computation across layers. Using activation patching, ablation, and attention entropy across two synthetic regression datasets, we find clear temporal specialisation: one head's causal necessity dominates that of the others by 2 to 5 times at peak layer, with its dominant layer shifting across tasks of different complexity, while the remaining heads exhibit symmetric late layer profiles. Attention entropy and patching provide convergent evidence for the computationally active layers of the dominant head. We additionally investigate inference time steerability via contrastive activation steering, which fails to transfer across samples. We attribute this result to TabPFN's in context learning mechanism, which encodes task structure through context dependent attention rather than the stable parametric directions that make steering tractable in language models.

Related papers

• Topological Signatures of Context-Level Reliability in TabPFN [0.0]
  We study the behavior of TabPFN's internal representation geometry using persistent homology.<n>We find that the topology of TabPFN's internal representation geometry is strongly associated with dataset-level reliability.<n>Our results suggest that zigzag persistence diagnoses the reliability of the inferred in-context task geometry and provides a context-level view of when TabPFN operates in topologically stressed regimes.
  arXiv  Detail & Related papers  (2026-07-20T14:04:01Z)
• Singular perturbations and hierarchical learning in two-layer neural networks [3.899493143383179]
  We study the population gradient flow of an infinitely wide two-layer neural network learning a misspecified single-index model in high dimension.<n>We prove that the constant and linear components of the hidden link function are indeed recovered within the predicted timescales.
  arXiv  Detail & Related papers  (2026-07-12T18:10:23Z)
• How Vision Becomes Language: A Layer-wise Information-Theoretic Analysis of Multimodal Reasoning [3.720343746691839]
  We show that visual-unique information peaks early and decays with depth, language-unique information surges in late layers, and cross-modal synergy remains below 2%.<n>These results provide an information-theoretic, causal account of how vision becomes language in multimodal Transformers.
  arXiv  Detail & Related papers  (2026-02-17T13:49:49Z)
• How Transformers Learn Regular Language Recognition: A Theoretical Study on Training Dynamics and Implicit Bias [48.9399496805422]
  We focus on two representative tasks in the category of regular language recognition, known as even pairs' and parity check'<n>Our goal is to explore how a one-layer transformer, consisting of an attention layer followed by a linear layer, learns to solve these tasks.
  arXiv  Detail & Related papers  (2025-05-02T00:07:35Z)
• A Closer Look at TabPFN v2: Understanding Its Strengths and Extending Its Capabilities [51.08999772842298]
  Tabular Prior-data Fitted Network v2 (TabPFN v2) achieves unprecedented in-context learning performance across diverse downstream datasets.<n>We show that TabPFN v2 can infer attribute relationships even when provided with randomized attribute token inputs.<n>We demonstrate that TabPFN v2's limitations can be addressed through a test-time divide-and-context strategy.
  arXiv  Detail & Related papers  (2025-02-24T17:38:42Z)
• Towards Robust and Realistic Human Pose Estimation via WiFi Signals [85.60557095666934]
  WiFi-based human pose estimation is a challenging task that bridges discrete and subtle WiFi signals to human skeletons.<n>This paper revisits this problem and reveals two critical yet overlooked issues: 1) cross-domain gap, i.e., due to significant variations between source-target domain pose distributions; and 2) structural fidelity gap, i.e., predicted skeletal poses manifest distorted topology.<n>This paper fills these gaps by reformulating the task into a novel two-phase framework dubbed DT-Pose: Domain-consistent representation learning and Topology-constrained Pose decoding.
  arXiv  Detail & Related papers  (2025-01-16T09:38:22Z)
• Unveiling Induction Heads: Provable Training Dynamics and Feature Learning in Transformers [54.20763128054692]
  We study how a two-attention-layer transformer is trained to perform ICL on $n$-gram Markov chain data. We prove that the gradient flow with respect to a cross-entropy ICL loss converges to a limiting model.
  arXiv  Detail & Related papers  (2024-09-09T18:10:26Z)
• The Remarkable Robustness of LLMs: Stages of Inference? [5.346230590800585]
  We investigate the robustness of Large Language Models (LLMs) to structural interventions by deleting and swapping adjacent layers during inference.<n>Surprisingly, models retain 72-95% of their original top-1 prediction accuracy without any fine-tuning.
  arXiv  Detail & Related papers  (2024-06-27T17:57:03Z)
• A phase transition between positional and semantic learning in a solvable model of dot-product attention [30.96921029675713]
  Morelinear model dot-product attention is studied as a non-dimensional self-attention layer with trainable and low-dimensional query and key data. We show that either a positional attention mechanism (with tokens each other based on their respective positions) or a semantic attention mechanism (with tokens tied to each other based their meaning) or a transition from the former to the latter with increasing sample complexity.
  arXiv  Detail & Related papers  (2024-02-06T11:13:54Z)
• In-Context Convergence of Transformers [63.04956160537308]
  We study the learning dynamics of a one-layer transformer with softmax attention trained via gradient descent. For data with imbalanced features, we show that the learning dynamics take a stage-wise convergence process.
  arXiv  Detail & Related papers  (2023-10-08T17:55:33Z)
• Domain Adaptive Synapse Detection with Weak Point Annotations [63.97144211520869]
  We present AdaSyn, a framework for domain adaptive synapse detection with weak point annotations. In the WASPSYN challenge at I SBI 2023, our method ranks the 1st place.
  arXiv  Detail & Related papers  (2023-08-31T05:05:53Z)
• Contrastive Conditional Neural Processes [45.70735205041254]
  Conditional Neural Processes(CNPs) bridge neural networks with probabilistic inference to approximate functions of Processes under meta-learning settings. Two auxiliary contrastive branches are set up hierarchically, namely in-instantiation temporal contrastive learning(tt TCL) and cross-instantiation function contrastive learning(tt FCL) We empirically show that tt TCL captures high-level abstraction of observations, whereas tt FCL helps identify underlying functions, which in turn provides more efficient representations.
  arXiv  Detail & Related papers  (2022-03-08T10:08:45Z)

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.