decomposition

Introduces DECAF, a method that decomposes perturbation responses into evidence, contradiction, and fragility components, improving interpretability over raw response magnitude and achieving strong results across vision benchmarks.

This paper identifies and characterizes Decomposition-Induced Context-Memory Conflict (DI-CC), a failure mode in decompose-then-verify pipelines where decomposition substitutes the model's parametric beliefs for the source text. The authors show it is mechanistically related to classical context-memory conflict, that SelfCheckGPT fails to detect it, and that context-aware decoding suppresses it but introduces severe parsing failures.

This paper introduces TriQua, a framework for LLM factuality evaluation that adaptively represents facts as triples or hyperrelational facts with contextual qualifiers, along with TriQuaScore for fine-grained factuality scoring. It demonstrates strong alignment with human annotations and improved evidence-based verification over existing methods.

This paper shows that per-chunk verification fails for multi-hop RAG because no single chunk is sufficient, and proposes decomposition-based verification to repair it, demonstrating significant improvements across multiple datasets.

Introduces DWM, a framework that decomposes latent world model transitions into action-driven and action-invariant (world effect) components, improving planning success on benchmarks with persistent world effects.

This paper proposes a qubit-efficient quantum framework for hyperdimensional computing decomposition that reduces representation cost from O(D) to O(log D) qubits while preserving the quadratic search advantage, achieving up to 2000x fewer qubits.

Mitchell Hashimoto shares a heuristic for using AI agents: any diff over 1500 lines indicates a need for decomposition, and outlines a pattern for iterative development with agent-generated code.

This paper presents an information-theoretic analysis of multimodal learning, revealing the need to capture sample-specific interactions, and proposes DMIL, a paradigm that explicitly models and learns from these interactions via variational decomposition and fine-tuning, achieving superior performance.

Proposes TD-Grokking, a training-time decomposition framework that recursively breaks down intractable zero-reward problems into verifiable subproblems, enabling LLMs to learn from failed trajectories. Outperforms vanilla GRPO and baselines on mathematical and medical reasoning tasks.

DeSQ is a decomposition-based framework for generating SPARQL queries from natural language questions. It breaks complex questions into atomic constraints, maps them to SPARQL fragments, and assembles them into complete queries, outperforming state-of-the-art on four out of five benchmarks.

The paper decomposes the attention interaction matrix into routing (skew-symmetric) and filtering (symmetric) components, introducing S-D attention to disentangle them. It reveals a spectral cascade in routing that predicts where attention can be simplified, achieving significant parameter reduction with minimal perplexity loss.

Proposes PESD-TSF, a physics-inspired structured decomposition framework for long-term time series forecasting that addresses periodic perception degradation, trend-noise entanglement, and loss of cross-variable dependencies via multiplicative periodic gating, multi-scale structured encoder, and cross-scale collaborative attention.