When Attention Collapses: Residual Evidence Modeling for Compositional Inference

• URL: http://arxiv.org/abs/2605.02323v1
• Date: Mon, 04 May 2026 08:18:02 GMT
• Title: When Attention Collapses: Residual Evidence Modeling for Compositional Inference
• Abstract summary: Under additive superposition, multiple slots converge to the same dominant component while weaker ones remain unrepresented.<n>We introduce residual evidence modeling, instantiated via evidence depletion, a minimal modification combining multiplicative depletion with an attention bias.<n>Across synthetic benchmarks and real-world audio mixtures, evidence depletion reduces slot collapse by up to an order of magnitude.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Compositional inference - the decomposition of observations into an unknown number of latent components - is central to perception and scientific data analysis. Attention-based models perform well when components are approximately separable, as in object-centric vision. Under additive superposition, however - where multiple components contribute to every observation - we identify a structural failure mode we term slot collapse: multiple slots converge to the same dominant component while weaker ones remain unrepresented. We trace this to a general limitation: attention is memoryless with respect to explained evidence. All slots repeatedly operate on the same input without accounting for what has already been explained, so gradients are dominated by the strongest component, inducing shared fixed points across slots. As a result, attention fails to enforce non-redundant allocation under additive superposition. We address this by introducing residual evidence modeling, instantiated via evidence depletion - a minimal modification combining multiplicative depletion with an attention bias. Controlled ablations show that parallel attention, sequential processing alone, and loss-based regularization fail to resolve collapse; evidence depletion, which adds residual state to sequential attention, consistently succeeds. Across synthetic benchmarks and real-world audio mixtures (FUSS), evidence depletion reduces slot collapse by up to an order of magnitude, generalizing beyond synthetic settings. On gravitational-wave source inference for the ESA/NASA LISA mission, under identical architectures, data, and losses, standard attention fails while evidence depletion prevents collapse and enables multi-source posterior estimation. These results show that under additive superposition, residual evidence tracking is the operative ingredient for preventing collapse and enabling compositional inference.

Related papers

• NICE: Scale-Stable Perturbations for Graph Neural Network Explanations via Noise Corruption [67.57304101271096]
  Element-wise Masking (EM) suppresses edge-induced messages toward zero.<n>Noise Corruption (NC) perturbs each message through matched-norm random-direction corruption.<n>NICE, a Noise Corruption-based explanation framework, learns a Restoration Boundary (SRB) under NC-induced uncertainty.
  arXiv  Detail & Related papers  (2026-08-17T03:08:14Z)
• Residual Dominance as a Structural Account of Last-Item Reliance in Causal Self-Attention Recommenders [47.8060666024071]
  We show that SASRec-style models exhibit highly localized last-item reliance.<n>We then find that residual addition sharply shifts the full-block representation toward same-position contributions.<n>Changing the residual strength induces a monotonic trade-off between structural mixing and last-item reliance.
  arXiv  Detail & Related papers  (2026-08-14T07:09:58Z)
• GAUGE: Granularity-Adaptive Counterfactual Gating of Evidence for Incomplete Multimodal Classification [17.595478415811687]
  Multimodal classification typically assumes all modalities are available, yet real-world inputs are often incomplete.<n>We propose GAUGE, a lightweight counterfactual gating framework for incomplete multimodal classification.<n>GAUGE scores the counterfactual effect of replacing every unit with a reference representation through prediction-aware Taylor evidence scores.
  arXiv  Detail & Related papers  (2026-08-06T05:07:14Z)
• Geometric Collapse: When Vision Models Fail to Verify Physical Causality [52.43819179934414]
  We propose Scrambled Edges, a controlled counterfactual that injects salient edge-like cues.<n>With energy-matched and structure-matched controls, we isolate the effect of unsupported edge evidence from high-frequency energy and edge sparsity.<n>Across CNN/ViT/SSL depth predictors, Scrambled Edges induce up to 3.2x larger deviation from clean predictions than energy-matched noise.
  arXiv  Detail & Related papers  (2026-07-08T00:04:02Z)
• The Structural Origin of Attention Sink: Variance Discrepancy, Super Neurons, and Dimension Disparity [22.040532283910522]
  This work provides a textitmechanistic explanation for the phenomenon.<n>We trace its root to the value aggregation process inherent in self-attention.<n>As a proof of concept, we propose textithead-wise RMSNorm, an architectural modification that stabilizes value aggregation outputs during pre-training.
  arXiv  Detail & Related papers  (2026-05-07T17:28:55Z)
• Causal Discovery in Action: Learning Chain-Reaction Mechanisms from Interventions [0.0]
  Causal discovery is challenging in general dynamical systems because the underlying causal graph may not be identifiable even from interventional data.<n>We study causal discovery in such chain-reaction systems and show that the causal structure is uniquely identifiable from blocking interventions.
  arXiv  Detail & Related papers  (2026-03-23T22:45:15Z)
• Neural Uncertainty Principle: A Unified View of Adversarial Fragility and LLM Hallucination [60.197429875410286]
  Adversarial vulnerability in vision and hallucination in large language models are conventionally viewed as separate problems.<n>Formalizing a Neural Uncertainty Principle (NUP) under a loss-induced state, we find that in near-bound regimes, further compression must be accompanied by increased sensitivity dispersion.<n>In vision, masking highly coupled components improves robustness without costly adversarial training.<n>In language, the same prefill-stage probe detects hallucination risk before generating any answer tokens.
  arXiv  Detail & Related papers  (2026-03-20T02:07:10Z)
• Multi-Agent Debate: A Unified Agentic Framework for Tabular Anomaly Detection [7.807411603618987]
  Multi-Agent Debating framework (MAD) treats disagreement as a first-class signal.<n> MAD produces a normalized anomaly score, confidence, and structured evidence, augmented by a large language model (LLM)-based critic.<n>We show how conformal calibration can wrap the debated score to control false positives under exchangeability.
  arXiv  Detail & Related papers  (2026-02-15T17:44:32Z)
• The Alignment Bottleneck in Decomposition-Based Claim Verification [17.197804072440665]
  We introduce a new dataset of real-world complex claims featuring temporally bounded evidence and human-annotated sub-claim evidence spans.<n>We evaluate decomposition under two evidence alignment setups: Sub-claim Aligned Evidence (SAE) and Repeated Claim-level Evidence (SRE)<n>Our results reveal that decomposition brings significant performance improvement only when evidence is granular and strictly aligned.
  arXiv  Detail & Related papers  (2026-02-11T00:02:16Z)
• A Unified View of Attention and Residual Sinks: Outlier-Driven Rescaling is Essential for Transformer Training [86.64715217940274]
  Outliers function jointly with normalization.<n>Outliers serve more as rescale factors rather than contributors.<n>Outliers can be absorbed into learnable parameters or mitigated via explicit gated rescaling.
  arXiv  Detail & Related papers  (2026-01-30T13:29:45Z)
• Random-Matrix-Induced Simplicity Bias in Over-parameterized Variational Quantum Circuits [72.0643009153473]
  We show that expressive variational ansatze enter a Haar-like universality class in which both observable expectation values and parameter gradients concentrate exponentially with system size.<n>As a consequence, the hypothesis class induced by such circuits collapses with high probability to a narrow family of near-constant functions.<n>We further show that this collapse is not unavoidable: tensor-structured VQCs, including tensor-network-based and tensor-hypernetwork parameterizations, lie outside the Haar-like universality class.
  arXiv  Detail & Related papers  (2026-01-05T08:04:33Z)
• Causal Discovery with Mixed Latent Confounding via Precision Decomposition [0.0]
  Differentiable and score-based DAG learners can misinterpret global latent effects as causal edges, while latent-variable graphical models recover only undirected structure.<n>We propose textscDCL-DECOR, a modular, precision-led pipeline that separates these roles.
  arXiv  Detail & Related papers  (2025-12-31T08:03:41Z)
• Analyzing the Mechanism of Attention Collapse in VGGT from a Dynamics Perspective [13.434698786044107]
  Visual Geometry Grounded Transformer (VGGT) delivers state-of-the-art feed-forward 3D reconstruction.<n>Its global self-attention layer suffers from a drastic collapse phenomenon when the input sequence exceeds a few hundred frames.<n>We establish a rigorous mathematical explanation of the collapse by viewing the global-attention as a degenerate diffusion process.
  arXiv  Detail & Related papers  (2025-12-25T14:34:27Z)
• Compose and Fuse: Revisiting the Foundational Bottlenecks in Multimodal Reasoning [49.17801010041155]
  Multimodal large language models (MLLMs) promise enhanced reasoning by integrating diverse inputs such as text, vision, and audio.<n>Yet cross-modal reasoning remains underexplored, with conflicting reports on whether added modalities help or harm performance.<n>We categorize multimodal reasoning into six interaction patterns, varying how facts are distributed across modalities and logically combined.
  arXiv  Detail & Related papers  (2025-09-28T08:46:11Z)
• Neuro-Symbolic Entropy Regularization [78.16196949641079]
  In structured prediction, the goal is to jointly predict many output variables that together encode a structured object. One approach -- entropy regularization -- posits that decision boundaries should lie in low-probability regions. We propose a loss, neuro-symbolic entropy regularization, that encourages the model to confidently predict a valid object.
  arXiv  Detail & Related papers  (2022-01-25T06:23:10Z)

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.