Architecture Before the Formula: Training-Dependent Computational Semantics in Transformers

Luis F. Rosario Freytes

cs.LGJan 10, 2026

Architecture Before the Formula: Training-Dependent Computational Semantics in Transformers

Authors: Luis F. Rosario Freytes

Abstract

An operation's input-output function does not, by itself, determine its computational role within a represented process. That role also depends on the processes producing its inputs and receiving its outputs, a philosophical stance termed deep learning externalism (DLE). We develop a mathematical account of represented computation that distinguishes functional equivalence, retained predecessor distinctions, restricted receiver capability, and context-dependent effective architectures. The resulting equivalences depend on which computational relations a scientific description must preserve. We prove that two injective producer contexts can preserve every predecessor distinction while changing whether fixed local and attention receiver families are equivalent. We investigate these relations in a dimension-labelled PreNorm Transformer and their evolution under training. Residual bypasses restrict complete-state information identifications to two merges per block, while finer branches can change receiver accessibility. Exact gradient-descent models yield covariance-dependent geometric attenuation under learned producer routing and algebraic $(t^{-1/2})$ attenuation under Q/K-softmax training, with optimal bounded-receiver recovery errors derived in both cases. Coupled producer-attention dynamics establish how learning jointly reorganizes the presented features and receiver allocation. Finally, present functional or receiver-profile equivalence need not determine future behavior under an optimizer. In an input-conditioned attention model, two accessibility measurements form an exact reduced predictive state, whereas one alone does not.

Figures & tables

Panel of Figure 1 — Figure 1: The left column passes from a represented cut to its realized composite and the equivalence relation induced on predecessor states. The right column starts from a complete training state, its transition kernel, and a selected observable, and derives the information required to predict that observable. The two constructions retain different information about the same represented computation.

Panel of Figure 2 — Figure 2: Two injective producers present the same predecessor information differently. With fixed local and attention receiver families, the encoded presentation gives identical distinction profiles. The separated presentation gives the attention family access to $v$ , which is unavailable to the token-1 local receiver. Thus receiver equivalence changes without loss of predecessor information.

Panel of Figure 3 — Figure 3: A represented interface has two external relations. The producer prefix and receiver continuation are not determined by the isolated carrier state.

Panel of Figure 4 — Figure 4: Two residual merges in one PreNorm Transformer block. At a fine cut inside either branch, the carried residual remains separately present. Whole-state identification can occur at the two merges, while normalization, attention routing, hidden FFN computation, and their receiving gradients have finer, receiver-relative semantics. Backpropagation transports covectors through the reversed represented graph.

Figure 5: The derived gradient-flow field for the coupled producer orientation $c=\cos s$ and attention allocation $p$ . The dashed curve is the moving allocation optimum $p_{*}(c)=(1+c)/2$ , not a set of joint equilibria. Trajectories approach aligned producer features and saturated source allocation at different rates. Parameters used for illustration: $\gamma_{s}=\gamma_{\alpha}=\tau=1$ .

Figure 6: Illustrative exact discrete learning trajectory of the producer-dependent Q/K model, showing $p_{+}$ and $p_{-}$ , which equal the optimal unit-Lipschitz recovery errors for the second source on the two input regimes. The same learned attention parameters produce different receiver accessibility because the upstream representations differ.

Explore similar work

Aug 31, 2026

Tracing distinguishability through transformer processing with stochastic LayerNorm

Representational similarity is foundational to analyses of deep networks, yet distances between point-valued representations are not intrinsically tied to downstream function: nearby states may produce different behaviors, while distant states may behave similarly. We instead give representations volume, turning similarity into statistical distinguishability. Overlapping stochastic representations necessarily induce overlapping downstream distributions, grounding latent comparison in model function and bringing it under information-theoretic tools such as the data-processing inequality. We realize this idea in pretrained transformers through a light-touch modification to LayerNorm: at each residual-stream read, we normalize the state, add isotropic Gaussian noise, and renormalize. During distillation fine-tuning, one learned allocation parameter per residual-stream read distributes a fixed global rate budget across the processing stack. The resulting model can be viewed as transformer blocks reading the residual stream with learned finite precision under a shared global rate budget. Using the Bhattacharyya coefficient, we trace which counterfactual distinctions are preserved through MLP blocks or selectively exposed to the query, key, and value computations of individual attention heads. Experiments on ViT-S and GPT-2 small reveal the depthwise propagation of continuous visual perturbations and head-specific sensitivity to token distinctions aligned with known attention motifs. These results establish distinguishability as a functionally grounded lens on transformer computation that complements existing interpretability approaches.

Kieran Murphy

New Jersey Institute of Technology

Transformer Interpretability Layer Normalization

Ask PDF

Aug 22, 2026

Align, Unify, Suppress, Route: A Coherentist View of Transformer Computation

Mechanistic interpretability has identified transformer circuits, but lacks a shared vocabulary for describing how their functions compose across tasks and architectures. We introduce Coherentist Probabilistic Compositionalism (CPC), an interpretive framework that grounds transformer computation in coherentist theories of interpretation and describes it through four operator roles. Alignment identifies candidate relations, unification integrates supporting information, suppression reduces incompatible alternatives, and routing carries selected information to the output. Across 15 models from five architecture families, the suppression, unification, and routing weight-space signatures correlate with held-out activation-level role measures above random baselines. Suppression is more stable across tasks than unification. Ablating alignment heads reduces downstream suppressive activity beyond a random-head control in 10 models, but similar effects on no-conflict prompts indicate a general upstream dependency, not contradiction-specific coupling. Explicit contradictions significantly shift a layerwise coherence proxy in 14 models; after removing shared residual covariance, the gap has the predicted direction in every model. Base and instruction-tuned variants preserve induction-head score structure (

r{\geq}0.98

) without a consistent shift of operator signatures towards later layers. These results support CPC as a shared vocabulary for comparing transformer mechanisms while showing that their depth and geometric expression remain architecture-specific.

Nura Aljaafari, Andre Freitas

University of Manchester, United Kingdom · Idiap Research Institute, Switzerland

Transformer Interpretability Mechanistic Interpretability

Ask PDF

Oct 14, 2026

Training Dynamics of Transformers on Unbounded Inputs: Attention Learning Can Lag Output Layer Learning

The success of transformers is attributed to several factors, including their capacity to express formal languages over variable-length inputs, and ability of the attention mechanism to focus on relevant input features. However, there is limited understanding of the training dynamics of shallow transformers, specifically how to decouple the learning of the attention and the output (feedforward) layer. A natural extension of the theory of one-step gradient feature learning in feedforward networks suggests that one could first learn the attention weights with a single gradient step, and then train the output layer. We show that this fails for a shallow transformer trained to recognize a star-free language from unbounded-length input. We prove that the transformer can be trained to recognize the language with a three-phase gradient descent: the output layer is updated first, followed by an attention update, and the output layer is fine-tuned in the third phase. In contrast, the language cannot be learned by training only the attention or output layer, or by a single attention-layer update followed by finetuning the output layer. We empirically show that ``attention learning lagging output layer learning" is beneficial in other language recognition, regression, and in-context learning tasks as well.

Yaomengxi Han, Debarghya Ghoshdastidar

Technical University of Munich

Neural Network Training Dynamics Transformer Attention Transformer FFNs

Ask PDF