Retrieval-Warmed Energy-Based Reasoning: A Five-Arm Ablation Methodology for Diffusion-as-Inference on Structured Reasoning Tasks

2026-06-24

AI Summary

Existing diffusion-based inference struggles to disentangle whether performance gains stem from prior shift, stochastic warm starts, or value reuse via structural alignment. This work proposes a retrieval-warmed energy-based inference framework and introduces a novel five-arm ablation study—comprising oracle, best-constant, per-query-random, shuffled, and aligned conditions—to systematically isolate the contribution of each mechanism and precisely identify the dominant performance bottleneck in a given task. Integrating the IRED energy model, Modern Hopfield trajectory memory, task-specific key encoders, and an LLM-RAG-inspired diagnostic decomposition, the approach reveals a 35-percentage-point accuracy gap between aligned and shuffled retrieval on the connectivity-2 task and identifies key quality as the critical bottleneck in Sudoku solving.

Abstract

Warm-started diffusion samplers accelerate iterative inference, but it is rarely clear which part of the pipeline carries the gain. We study \textbf{retrieval-warmed energy-based reasoning (RW-EBR)} -- an IRED energy-based diffusion model \cite{du2024ired} augmented with a Modern Hopfield trajectory memory -- and contribute a \textbf{five-arm ablation methodology} (oracle, best-constant, per-query-random, shuffled, aligned) that separates three confounded effects: class-prior bias shift, stochastic warm-starting, and graph-aligned value reuse. The diagnostic decomposition is adapted from LLM-RAG evaluation \cite{ru2024ragchecker}. On \textbf{connectivity-2} (Erdős--Rényi all-pairs reachability), the aligned-vs-shuffled-oracle swing reaches \textbf{$+35$\,pp} balanced accuracy on a fixed 1{,}000-graph validation-set diagnostic, with value distribution and retrieval mechanics fixed, only per-graph alignment destroyed, while per-query random initialisation falls below cold -- per-graph alignment, not bias shift or stochasticity, dominates. Yet the \emph{deployable} cold-prediction pipeline misses the acceptance gate at stored-value quality. The same diagnostic logic, stopped at the key-quality screen, applied to \textbf{Sudoku} with a task-specific key encoder produces a clean negative at a \emph{different} component -- key quality, under the current setup. The decomposition names the first blocking component on each task. The setting -- graph reachability refined by an iterative diffusion sampler, with explainability of failure modes as the lens -- places the work within structured and spatio-temporal reasoning.

Problem

• diffusion-as-inference
• structured reasoning
• retrieval-warmed reasoning
• ablation methodology
• failure mode analysis

Innovation

• retrieval-warmed reasoning
• energy-based diffusion
• five-arm ablation
• graph-aligned value reuse
• structured reasoning