Interpretable-by-Design Transformers via Architectural Stream Independence

• URL: http://arxiv.org/abs/2603.07482v1
• Date: Sun, 08 Mar 2026 05:44:18 GMT
• Title: Interpretable-by-Design Transformers via Architectural Stream Independence
• Abstract summary: We investigate whether architectural constraints can enforce interpretability by design through architectural stream independence.<n>We validate this principle through the Late Fusion Architecture (LFA), which demonstrates interpretable symbolic heads through all the final layers.<n>We quantify this effect by introducing the Token-Position Dependence Score (PDS), with $PDS_max$ = 0.276 and 0.058 respectively.
• Score: 1.2891210250935148
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: While transformers achieve strong performance, their internal decision-making processes remain opaque. We investigate whether architectural constraints can enforce interpretability by design through architectural stream independence: maintaining a token stream (carrying symbolic structure) and contextual semantics in separated streams that remain independently observable throughout processing, with integration delayed until output. We validate this principle through the Late Fusion Architecture (LFA), which demonstrates interpretable symbolic heads through all the final layers, while standard transformers show dissolution by the third of six layers; we quantify this effect by introducing the Token-Position Dependence Score (PDS), with $PDS_{max}$ = 0.276 and 0.058, respectively. Crucially, intervention experiments demonstrate functional modularity: suppressing LFA's recency heads causes minimal semantic damage (Cohen's d = -0.158) versus catastrophic entanglement in baselines (d = -0.672). LFA demonstrates that architectural constraints improve underlying learning mechanisms, averaging 42% stability versus 19% and 11% for baseline comparisons, with extremes from 50% on LFA's best pairs (6 of 12 heads position-invariant) down to 0% complete collapse in over-constrained cases. By preventing premature entanglement, architectural independence steers models toward semantic understanding over positional heuristics, establishing interpretability as an architectural design criterion enforceable through structural constraints rather than post-hoc analysis.

Related papers

• Feature-Spectral Fragility in Segmentation: Dataset Dependence, Architecture-Specific Localization, and Spectral Correlates [0.0]
  We probe dependence on frequency content within learned feature representations using post-training low-pass interventions.<n>Dice drops by 100%, 73.2%, and 30.9% for CNN, SSM, and Transformer, respectively, compared with 9.4%, 10.3%, and 0.6% on ISIC.<n>Native feature-domain spectral measurements show an inverse association between high-frequency energy and fragility on CVC.
  arXiv  Detail & Related papers  (2026-08-29T09:33:45Z)
• BenthicDINO: Physics-Informed Self-Distillation for View-Invariant Side-Scan Sonar Representations [0.0]
  We introduce a physics-informed self-distillation framework built upon the DINOv3 architecture utilizing a ConvNeXt-v2-Tiny backbone to maximize data efficiency.<n>We propose a dense, hierarchical feature fusion strategy across all four network stages to preserve fine-grained sediment details alongside deep semantic abstractions.
  arXiv  Detail & Related papers  (2026-08-24T13:11:22Z)
• Align, Unify, Suppress, Route: A Coherentist View of Transformer Computation [0.0]
  Coherentist Probabilistic Compositionalism grounds transformer computation in coherentist theories of interpretation.<n>Across 15 models from five architecture families, the suppression, unification, and routing weight-space signatures correlate with held-out activation-level role measures.<n>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.
  arXiv  Detail & Related papers  (2026-08-22T16:29:20Z)
• Architecture-Dependent Causal Transfer of Activation States Across Large Language Models [51.56484100374058]
  Internal language activation states can be transferred causally between different large language models.<n>Injecting projected activations into a target model during generation produces a statistically significant, preregistered causal effect.<n>We conclude that end-to-end activation-state transfer between LLMs, as currently implemented, is architecture-dependent rather than universal.
  arXiv  Detail & Related papers  (2026-08-17T09:53:27Z)
• The Trade-off Between Covariate Dependence and Latent Structure in Representation Learning [0.0]
  Disentangled representation learning seeks latent representations whose indicidual dimensions each align with a distinct covariate.<n>We introduce a unified, supervised framework that couples latent dimension-covariate dependence with constraints on the latent structure.<n>We show that both post-hoc alignment and iFA enable controllability of structured latent representations.
  arXiv  Detail & Related papers  (2026-08-17T08:20:40Z)
• Hierarchical Consistency Learning for Test-time Adaptation in Camouflage Perception [50.278200968044665]
  Camouflaged object detection (COD) aims to localize targets that exhibit minimal perceptual differences from backgrounds through physical attributes.<n>Existing methods, constrained by the static train-then-freeze paradigm, suffer from domain rigidity and dependency annotation.<n>We propose the hierarchical consistency learning framework, which integrates test-time adaptation for dynamic representation recalibration.
  arXiv  Detail & Related papers  (2026-05-25T09:57:46Z)
• Dynamic analysis enhances issue resolution [53.50448142467294]
  DAIRA (Dynamic Analysis-enhanced Issue Resolution Agent) is an automated repair framework that embeds dynamic analysis into the agent's reasoning cycle.<n> Driven by a Test Tracing-Driven methodology, DAIRA utilizes lightweight monitors to extract critical runtime data.<n>When powered by Gemini 3 Flash Preview, DAIRA establishes a new state-of-the-art (SOTA) performance, achieving a 79.4% resolution rate on the SWE-bench Verified dataset.
  arXiv  Detail & Related papers  (2026-03-23T14:48:54Z)
• The Dual-Stream Transformer: Channelized Architecture for Interpretable Language Modeling [1.2891210250935148]
  We introduce the Dual-Stream Transformer, which decomposes the residual stream into two functionally distinct components.<n>We measure this tradeoff on language modeling tasks at 29M parameters.
  arXiv  Detail & Related papers  (2026-03-08T04:50:29Z)
• OrthoFormer: Instrumental Variable Estimation in Transformer Hidden States via Neural Control Functions [0.0]
  Transformer architectures excel at sequential modeling yet remain fundamentally limited by correlational learning.<n>We propose OrthoFormer, a causally grounded architecture that embeds instrumental variable estimation directly into Transformer blocks via neural control functions.
  arXiv  Detail & Related papers  (2026-03-08T03:05:16Z)
• Breaking the Factorization Barrier in Diffusion Language Models [59.946071582340146]
  "Factorization barrier" impedes efficient parallel generation of diffusion language models.<n>We propose Coupled Discrete Diffusion to replace the fully-factorized output distribution.<n>We show that CoDD seamlessly enhances diverse diffusion language model architectures with negligible overhead.
  arXiv  Detail & Related papers  (2026-02-09T08:36:39Z)
• Why Does the LLM Stop Computing: An Empirical Study of User-Reported Failures in Open-Source LLMs [50.075587392477935]
  We conduct the first large-scale empirical study of 705 real-world failures from the open-source DeepSeek, Llama, and Qwen ecosystems.<n>Our analysis reveals a paradigm shift: white-box orchestration relocates the reliability bottleneck from model algorithmic defects to the systemic fragility of the deployment stack.
  arXiv  Detail & Related papers  (2026-01-20T06:42:56Z)
• Decoupling the Effect of Chain-of-Thought Reasoning: A Human Label Variation Perspective [60.45433515408158]
  We show that long Chain-of-Thought (CoT) serves as a decisive decision-maker for the top option but fails to function as a granular distribution calibrator for ambiguous tasks.<n>We observe a distinct "decoupled mechanism": while CoT improves distributional alignment, final accuracy is dictated by CoT content.
  arXiv  Detail & Related papers  (2026-01-06T16:26:40Z)
• From Memorization to Creativity: LLM as a Designer of Novel Neural-Architectures [48.83701310501069]
  Large language models (LLMs) excel in program synthesis, yet their ability to autonomously navigate neural architecture design-balancing reliability, performance, and structural novelty--remains underexplored.<n>We address this by placing a code-oriented LLM within a closed-loop synthesis framework, analyzing its evolution over 22 supervised fine-tuning cycles.
  arXiv  Detail & Related papers  (2026-01-06T13:20:28Z)
• Robust-R1: Degradation-Aware Reasoning for Robust Visual Understanding [54.05243949024302]
  Existing robust MLLMs rely on implicit training/adaptation that focuses solely on visual encoder generalization.<n>We propose Robust-R1, a novel framework that explicitly models visual degradations through structured reasoning chains.<n>Our approach integrates: (i) supervised fine-tuning for degradation-aware reasoning foundations, (ii) reward-driven alignment for accurately perceiving degradation parameters, and (iii) dynamic reasoning depth scaling adapted to degradation intensity.
  arXiv  Detail & Related papers  (2025-12-19T12:56:17Z)
• CoT-X: An Adaptive Framework for Cross-Model Chain-of-Thought Transfer and Optimization [5.857877898558651]
  Chain-of-Thought (CoT) reasoning enhances the problem-solving ability of large language models (LLMs) but leads to substantial inference overhead.<n>This paper investigates efficient CoT transfer across models of different scales and architectures through an adaptive reasoning summarization framework.
  arXiv  Detail & Related papers  (2025-11-07T22:35:31Z)
• ERIS: An Energy-Guided Feature Disentanglement Framework for Out-of-Distribution Time Series Classification [51.07970070817353]
  An ideal time series classification (TSC) should be able to capture invariant representations.<n>Current methods are largely unguided, lacking the semantic direction required to isolate truly universal features.<n>We propose an end-to-end Energy-Regularized Information for Shift-Robustness framework to enable guided and reliable feature disentanglement.
  arXiv  Detail & Related papers  (2025-08-19T12:13:41Z)
• State Stream Transformer (SST) : Emergent Metacognitive Behaviours Through Latent State Persistence [0.0]
  State Stream Transformer (SST) is a novel LLM architecture that reveals emergent reasoning behaviours latent in pretrained weights.<n>SST introduces a sliding window latent state (FFN) cache with weighted decay that maintains and evolves persistent latent processes throughout autoregressive generations.<n>SST achieves substantial performance improvements over the base model on two reasoning benchmarks, reaching 89.01% accuracy on GSM-8K (0-shot) and 91.04% on ARC Challenge (0-shot CoT)
  arXiv  Detail & Related papers  (2025-01-30T14:03:36Z)
• Demystifying Inductive Biases for $\beta$-VAE Based Architectures [19.53632220171481]
  We shed light on the inductive bias responsible for the success of VAE-based architectures. We show that in classical datasets the structure of variance, induced by the generating factors, is conveniently aligned with the latent directions fostered by the VAE objective.
  arXiv  Detail & Related papers  (2021-02-12T23:57:20Z)

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.