Where to Place the Query? Unveiling and Mitigating Positional Bias in In-Context Learning for Diffusion LLMs via Decoding Dynamics

• URL: http://arxiv.org/abs/2606.19349v1
• Date: Sun, 26 Apr 2026 20:22:47 GMT
• Title: Where to Place the Query? Unveiling and Mitigating Positional Bias in In-Context Learning for Diffusion LLMs via Decoding Dynamics
• Abstract summary: This paper presents a comprehensive analysis unveiling that query position is actually a first-order variable in dLLMs.<n>We propose Average Confidence ($overlineC$), a novel metric tracking the iterative decoding process.<n>By establishing the foundational spatial ICL baselines, we introduce Auto-ICL, a training-free adaptive routing strategy.
• Score: 5.345984783628041
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: While In-Context Learning (ICL) is extensively studied in Autoregressive (AR) LLMs, its mechanism within Diffusion Large Language Models (dLLMs) remains largely unexplored. Unlike AR models restricted by unidirectional causal masking, dLLMs intrinsically utilize bidirectional attention, offering extensive spatial flexibility for query placement. Unfortunately, current practices conventionally inherit AR-style trailing-query templates, often overlooking the structural paradigm shift. This paper presents a comprehensive analysis unveiling that query position is actually a first-order variable in dLLMs. Through empirical decoupling, we demonstrate that positional variance impacts generation quality on par with example semantic quality. Internally, this positional sensitivity stems from a spatial ``Recency Effect'' in attention flow and task-dependent shifts in decoding trajectories. To mitigate this instability without ground-truth labels, we reveal that traditional single-step confidence ($C_{decoded}$) fails in dLLMs. Instead, we propose Average Confidence ($\overline{C}$), a novel metric tracking the iterative decoding process. By establishing the foundational spatial ICL baselines, we introduce Auto-ICL, a training-free adaptive routing strategy that dynamically optimizes query placement, robustly approaching oracle performance across heterogeneous reasoning and perception tasks.

Related papers

• Analysis-by-Proxy: Localization Signals in VLMs Operating as Condition Encoders [54.06522068041745]
  Vision-Language Models (VLMs) are increasingly utilized as the conditioning backbone for diffusion-based image editing.<n>While standalone VLMs demonstrate strong localization capabilities, editing pipelines frequently struggle to maintain this accuracy.<n>In this work, we investigate this performance gap, hypothesizing that it stems from treating the VLM as a condition encoder.
  arXiv  Detail & Related papers  (2026-07-07T16:11:13Z)
• Closed-Loop LLM Discovery of Non-Standard Channel Priors in Vision Models [48.83701310501069]
  Large Language Models (LLMs) offer a transformative approach to Neural Architecture Search (NAS)<n>We formulate the search as a sequence of conditional code generation tasks, where an LLM refines architectural specifications based on performance telemetry.<n>We generate a vast corpus of valid, shape-consistent architectures via Abstract Syntax Tree (AST) mutations.<n> Experimental results on CIFAR-100 validate the efficacy of this approach, demonstrating that the model yields statistically significant improvements in accuracy.
  arXiv  Detail & Related papers  (2026-01-13T13:00:30Z)
• Grounded Test-Time Adaptation for LLM Agents [75.62784644919803]
  Large language model (LLM)-based agents struggle to generalize to novel and complex environments.<n>We propose two strategies for adapting LLM agents by leveraging environment-specific information available during deployment.
  arXiv  Detail & Related papers  (2025-11-06T22:24:35Z)
• RECALL: REpresentation-aligned Catastrophic-forgetting ALLeviation via Hierarchical Model Merging [33.22889542330089]
  Internal representations in large language models (LLMs) serve as reliable proxies of learned knowledge.<n>We propose RECALL, a representation-aware model merging framework for continual learning without access to historical data.
  arXiv  Detail & Related papers  (2025-10-23T12:17:37Z)
• DAMR: Efficient and Adaptive Context-Aware Knowledge Graph Question Answering with LLM-Guided MCTS [28.828541350757714]
  This paper proposes Dynamically Adaptive MCTS-based Reasoning (DAMR) for Knowledge Graph Question Answering (KGQA)<n>DAMR integrates Monte Carlo Tree Search (MCTS) with adaptive path evaluation to enable context-aware KGQA.<n>Experiments on multiple KGQA benchmarks show DAMR significantly outperforms SOTA methods.
  arXiv  Detail & Related papers  (2025-08-01T15:38:21Z)
• LF-Steering: Latent Feature Activation Steering for Enhancing Semantic Consistency in Large Language Models [16.37602070339033]
  Large Language Models (LLMs) often generate inconsistent responses when prompted with semantically equivalent paraphrased inputs.<n>We propose LF-Steering, a novel activation steering approach to precisely identify latent feature representations responsible for semantic inconsistency.<n>Our method maps the hidden states of the relevant transformer layer into a sparsely activated, high-dimensional feature space based on a sparse autoencoder.
  arXiv  Detail & Related papers  (2025-01-19T13:06:51Z)
• Parallel Key-Value Cache Fusion for Position Invariant RAG [55.9809686190244]
  Large Language Models (LLMs) are sensitive to the position of relevant information within contexts.<n>We introduce a framework that generates consistent outputs for decoder-only models, irrespective of the input context order.
  arXiv  Detail & Related papers  (2025-01-13T17:50:30Z)
• Intuitive or Dependent? Investigating LLMs' Behavior Style to Conflicting Prompts [9.399159332152013]
  This study investigates the behaviors of Large Language Models (LLMs) when faced with conflicting prompts versus their internal memory. This will help to understand LLMs' decision mechanism and also benefit real-world applications, such as retrieval-augmented generation (RAG)
  arXiv  Detail & Related papers  (2023-09-29T17:26:03Z)
• Divide and Contrast: Source-free Domain Adaptation via Adaptive Contrastive Learning [122.62311703151215]
  Divide and Contrast (DaC) aims to connect the good ends of both worlds while bypassing their limitations. DaC divides the target data into source-like and target-specific samples, where either group of samples is treated with tailored goals. We further align the source-like domain with the target-specific samples using a memory bank-based Maximum Mean Discrepancy (MMD) loss to reduce the distribution mismatch.
  arXiv  Detail & Related papers  (2022-11-12T09:21:49Z)
• Semi-supervised Domain Adaptive Structure Learning [72.01544419893628]
  Semi-supervised domain adaptation (SSDA) is a challenging problem requiring methods to overcome both 1) overfitting towards poorly annotated data and 2) distribution shift across domains. We introduce an adaptive structure learning method to regularize the cooperation of SSL and DA.
  arXiv  Detail & Related papers  (2021-12-12T06:11:16Z)

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.