NeurIPS 2026 论文列表

Conversational large language models can respond with warmth and reassurance while failing to advance the user's stated goal. We conceptualize this failure as $\textbf{goal displacement}$, the substitution of persona-maintenance content for goal-relevant substance. To measure it, we introduce $\textbf{GDB}$ (Goal Displacement Benchmark), comprising 230 scenarios across nine domains and $4{,}134$ human-scored responses from 18 models. A $2 \times 2$ taxonomy crosses hard constraints with emotional distress, distinguishing failures to confront infeasible goals or combine empathy with practical guidance from reverse displacement, where unsolicited problem-solving replaces requested emotional support. A deletion-test rubric assesses whether substantive help remains after purely supportive sentences are removed. Independent raters achieve weighted $\kappa = 0.82$ against author scores for displacement severity. Trap displacement ranges from 0.0\% to 37.7\% and is strongly negatively correlated with Chatbot Arena Elo ($\rho = -0.706$, $n = 15$). Reverse displacement reaches 41.9\%, including substantial rates in models with negligible trap displacement. A factorial probe supports a distinction from social sycophancy. Human-scored responses in the high-complexity, low-pressure condition exhibit 35.0\% sycophancy but 7.5\% displacement, with little association between the constructs. Matched OLMo-2 checkpoints show that mixed empathy-with-constraints displacement remains at 52.3\% after supervised fine-tuning and falls to 13.8\% after direct preference optimization. GDB provides a reproducible diagnostic of whether responses advance the operationalized user goal, supporting model comparison, post-training evaluation, and regression testing. We release the dataset, annotations, and rubric under CC-BY-4.0 and the evaluation code under MIT.

Discrete graph diffusion models corrupt graph structure through independent edge updates with a Markovian noise model. Combined with the exchangeability requirement on the output distribution, the terminal prior is necessarily an Erd\H{o}s--R\'enyi (E-R) random graph. This has a fundamental consequence for unconditional graph generation without node features: in the large graph limit, any GNN denoiser applied to an E-R graph approximately outputs an E-R graph, making effective denoising difficult. In practice, generated graphs fail to recover meaningful node heterogeneity and higher-order structure. Previous models introduced specialized node features to compensate, framing this as an expressivity problem, while the role of the prior remained unexamined. We instead address the root cause by introducing the \emph{Latent Sociability Prior} (LSP) and a \emph{joint} diffusion process that co-evolves latent node sociability variables and graph structure, removing edge independence while preserving exchangeability. Unlike existing latent graph diffusion methods, the sociability mechanism requires no complex encoder: its sole role is to guide edge updates and preserve node heterogeneity throughout the trajectory. We validate our approach on multiple graph generation datasets and show that the proposed framework more accurately captures graph distributions than baseline methods without node features.

Low-Rank Adaptation (LoRA) constrains weight updates to a low-rank subspace, making the choice of that subspace central to adaptation. We show that the value of LoRA initialization direction is regime-dependent. It helps when the representation dimension is large relative to the labeled set size and the estimated activation subspace is aligned with task structure, and is otherwise inconsequential. We formalize this view with a spiked-covariance Davis--Kahan argument and a first-step LoRA analysis showing that the initial learning signal flows through projected target-token features. From this analysis we derive an alignment coefficient $\rho$, computable from a single forward pass, that closely tracks the empirical gain from direction-aware initialization across tasks. Motivated by this view, we propose PivotLoRA, an initialization that estimates LoRA adapters from target-token PCA and applies a single output-preserving orthogonal pivot after early training drift. Across $30$ BERT-base few-shot settings and $13$ compared methods, PivotLoRA achieves the best average rank and improves over vanilla LoRA on nearly every setting, with the largest gains concentrated on high-alignment tasks. Decoder-scale classification on LLaMA-2-7B and Llama-3.1-8B reproduces the low-shot pattern, while full-data and generation controls confirm the predicted regime specificity.

For $K=2$ regimes, scalar thresholding of the Bayesian posterior is Bayes-optimal. For $K \geq 3$, the posterior evolves on a $(K-1)$-dimensional simplex $\Delta^{K-1}$, yet virtually all deployed detection systems reduce it to a scalar score without theoretical justification. We resolve this question completely. We introduce **Extended Decision Sufficiency (EDS)**—three linear constraints on emission densities and transition dynamics (Rank-One Emissions, Markov Factorisation, Normal Factorisation), verifiable in $O(K^2)$ operations from model parameters alone—and prove the following exact characterisation: **Main result.** EDS holds if and only if the Bayes-optimal stopping value function is constant on every level set of a scalar statistic $\phi : \Delta^{K-1} \to \mathbb{R}$. Sufficiency follows from algebraic closure of the Bellman operator $\mathcal{B}$ under EDS. Necessity is proved by a coupling-based fixed-point contradiction: for each failure mode we exhibit an explicit belief pair with equal $\phi$-values but a strictly positive total-variation gap on predictive densities, which propagates to a value-function gap via the Lipschitz bound on $\mathcal{B}$ and the $\rho$-contraction of $\mathcal{T}$, without assuming any structure on $V$ beyond continuity. Three consequences are sharp and quantitative: * **(i)** When EDS fails, *every* scalar rule—regardless of architecture, capacity, or training data—incurs a strictly positive, computable, algorithm-independent lead-time loss. * **(ii)** When EDS holds, optimal expected lead time is given in closed form by a hitting-time functional of a scalar Markov chain. * **(iii)** Under approximate EDS with deviation $\eta$, performance degrades at rate $O(\eta/(1-\rho))$, making $\eta$ a practical model-diagnostic with direct performance implications. Across 72 synthetic configurations and the CICIDS2017 benchmark, predicted and observed lead times agree within 0.4 steps (MAE). A controlled experiment with matched parameter counts confirms that the lead-time advantage is structural, not a regularisation artefact, thereby validating the theory as the sole causal mechanism.

Online imitation learning (IL), particularly on-policy distillation, has emerged as a strong LLM post-training approach, often outperforming offline supervised fine-tuning (SFT). Yet a principled understanding of when and why online helps remains unclear. In this work, we challenge the view that error accumulation is the main source of online IL's advantage, and instead show that the benefits of online interaction depend critically on whether the setting is realizable, i.e., whether the student policy class can represent the expert policy. Under realizability, we empirically find that offline IL already matches expert performance. In contrast, in non-realizable (misspecified) settings, we prove that offline IL encounters an information-theoretic bottleneck even when horizon $H=1$, and propose a structural characterization of misspecification relative to the reward, under which online IL provably achieves high performance despite large expert-student discrepancy.

Process reward models (PRMs) dramatically outperform outcome reward models (ORMs) for math reasoning and agent tasks, yet provide no benefit for chat and QA. No theory explains why, and practitioners choose between them by expensive trial and error. We introduce state-lift, a diagnostic that predicts whether a PRM is worth training for a new domain — in minutes, from ~500 labeled steps, before committing to expensive reward model training. State-lift measures how much step quality depends on trajectory state versus action text alone, and an accompanying effective gap distinguishes two regimes: state-noise-dominant (PRM essential) and action-gap-dominant (ORM sufficient). Across ten domains and eleven published PRM-ORM comparisons, state-lift correctly predicts PRM advantage, and a minutes-scale linear estimate matches the outcome of full Llama-3.1-8B reward model training — including CaSiNo negotiation (+0.055 lift at SL = 0.22). The accompanying two-branch decision protocol turns state-lift into an actionable PRM-vs-ORM workflow that runs in minutes on a single CPU.

Imitation learning (IL)—training an agent to replicate expert behavior from demonstrations—underpins applications from robotics to language model training. Standard approaches such as Behavior Cloning (BC) are known to suffer from compounding errors and performance plateaus, particularly when the learner cannot perfectly represent the expert’s policy (e.g., as is typical in distillation). Two interventions are widely understood empirically to improve performance: querying the expert interactively along the learner’s own trajectories, and using value function estimation en route to generating a policy. We investigate the nature of these improvements and their potentially surprising interplay. Our main finding is that expert interaction relaxes the representational demands on the learner: one only needs a model capable of realizing the expert’s value function, bypassing the (often stricter) requirement of realizing the expert’s policy itself. Concretely, we introduce OVI, an interactive IL algorithm that is statistically and computationally efficient whenever the learner can represent the expert’s value function. We complement this with a negative result showing that interaction is necessary: without significantly stronger representational assumptions than expert-value realizability alone, a broad class of value-based IL algorithms cannot succeed in the offline setting. These findings bear out empirically: OVI outperforms offline policy-based (BC), interactive policy-based (DAgger), and offline value-based IL methods, with the largest gains when the learner network is substantially less expressive than the expert’s.

Graph retrieval-augmented generation (GraphRAG) is motivated by the idea that relational structure can help a language model find evidence that flat retrieval may miss. Yet retrieval scores and final-answer scores often compress a multi-stage pipeline into a single method row. We study the missing middle of this pipeline through an evidence-to-answer audit that follows each system from candidate availability, through evidence selection and context construction, to generated answers and cost. Across GraphRAG-Benchmark Medical and Novel tasks, we find a consistent conversion gap. Strong reranking, learned evidence scoring, and graph-based retrieval can improve evidence-side signals, but these gains do not yield significant answer-correctness improvements under the same generator and evaluator. Matched attribution controls show that some apparent graph-side gains are driven by reranking rather than graph expansion alone. Gold-evidence and bridge-aware interventions further show that evidence completeness is insufficient on its own; answerability also depends on salience, organization, and support in the generator-facing context. We therefore argue that GraphRAG evaluation should report the full evidence-to-answer conversion chain, including candidate source, selected context, generated answer, metrics, and cost, rather than relying on retrieval or answer leaderboards alone.

Large language models may encounter factual knowledge during pre-training yet fail to reliably use that knowledge after fine-tuning. We study this gap in a stylized one-layer self-attention + MLP transformer trained by next-token prediction and subsequently fine-tuned on question-answering data. We first prove that, under suitable regularity conditions, the model reaches near-optimal pre-training loss while learning structured attention patterns. We further show that fine-tuning turns the Q\&A prompt format into a trigger for pre-trained relation features, enabling the model to extract facts not revisited during fine-tuning. Our analysis reveals a relation-covering characterization for knowledge extraction: fine-tuning need not revisit every stored fact, but it must cover enough latent relation-template directions through which facts were encoded during pre-training. We prove that extraction improves with pre-training multiplicity and fine-tuning coverage, but becomes harder as the relation-template universe grows. Conversely, insufficient coverage yields a failure regime in which facts can be stored but not extracted, providing a stylized mechanism for hallucination. Our analysis covers both full and low-rank fine-tuning, and experiments on synthetic data and PopQA-based GPT-2/Llama models support the predicted trends.

LLM powered multi-agent systems (MAS) have emerged as a promising paradigm for complex tasks. However, their advantages over single-agent systems (SAS) remain unclear, with performance varying inconsistently across settings. Here, we provide an information bottleneck perspective on elucidating the differences between MAS and SAS. Specifically, our key observation is that a SAS accumulates its full reasoning trace in one shared context, while a MAS uses isolated local contexts connected by bounded relay messages. We show that, under infinite relay bandwidth, any SAS can be simulated by a MAS that transmits the full upstream context. Thus, the nontrivial advantage of MAS arises under bounded relays, where compression introduces a fundamental trade-off: reducing redundant context can improve efficiency, but may also incur loss of task-relevant information. We formalize this trade-off as an information bottleneck controlled by an effective parameter $\beta$, which captures how the balance shifts with model capability, and shows that MAS gains arise when context reduction outweighs relay information loss. We conduct 18 controlled experiments across five benchmarks and three model scales to validate our theoretical studies. We observe that MAS consistently helps when relays are near-sufficient, especially for weaker models. In contrast, MAS gains shrink or reverse when relays incur information loss, especially for stronger models that can already extract useful information from redundant context and thus gain little from compression. Our study shows that multi-agent design is fundamentally an information-bottleneck optimization problem. This perspective explains when bounded inter-agent communication helps or hurts.

Causal sensitivity analysis bounds the policy value of off-policy evaluation under unobserved confounding. In offline reinforcement learning, however, sensitivity bounds are typically applied not to raw states but to learned representations. We show that this two-stage pipeline can silently break the coverage guarantee: state aggregation under hidden confounding amplifies the effective sensitivity parameter in the latent space, so bounds computed at the nominal confounding level may exclude the true policy value. We characterise the amplification mechanism, give a sufficient condition for preservation, and prove a continuity result showing the safety condition is not an isolated point: amplification grows continuously with the conditional-independence violation. Neither unsupervised nor standard task-aware representation objectives target the right quantity. We propose Sensitivity-Preserving Representation Learning (SPRL), which augments latent dynamics prediction with a kernel-weighted within-cell propensity homogeneity penalty, an observable surrogate for the unobservable safety condition, and show across synthetic and semi-synthetic benchmarks that SPRL is the only method tested that simultaneously preserves coverage in regimes where unsupervised representations break it and prevents amplification in regimes where task-aware baselines fail.

Cosine nearest-class-mean (NCM) evaluation is a widely used default for closed-set frozen-feature image classification because it is deterministic, train-free, and easy to reproduce. This convenience hides a metric assumption: raw angular geometry alone suffices for benchmark conclusions. We show that the assumption changes model-selection outcomes, not just absolute accuracies: across nine datasets and six frozen backbones, cosine-only reporting changes which backbone ranks first on 2 of 9 datasets and incurs 6.71 percentage points of regret relative to a best-evaluator oracle; on fixed DINOv2-B features, the underlying cosine-to-LDA gap reaches 37.63 points on Aircraft. Motivated by the low-variance spectral directions where class-discriminative signal often appears in such high-gap cases, we call this evaluator dependence the Spectral Decision Gap (SDG). SDG quantifies the gap between raw cosine and train-validated whitening-aware reference evaluators, flagging when cosine-only reporting changes benchmark conclusions. A leave-one-dataset predictive protocol reaches Pearson $r=0.752$ for best-whitening disagreement and incurs only 0.56pp routing regret at a 5pp threshold. This paper provides a compact audit protocol and artifact suite for detecting, predicting, and reporting when cosine prototype conclusions are reliable in this setting. Code and artifacts are available anonymously at https://anonymous.4open.science/r/sdg-ed/.

Vision-language models (VLMs) have achieved significant progress in scene understanding and spatial reasoning. However, existing evaluation benchmarks predominantly assume that all visible content corresponds directly to physical entities in the scene, leaving model performance in reflective scenarios largely unexamined. Reflective surfaces preserve object appearance while altering the mapping between appearance and physical location, causing VLMs to produce systematic reasoning errors in such scenes. In this work, we introduce (i) a grounded RGB-D dataset for mirror-aware spatial reasoning, and (ii) the first comprehensive and systematic evaluation of state-of-the-art VLMs in reflective scenarios. Specifically, we collect 1,011 RGB-D images from 525 real-world scenes, spanning diverse indoor and outdoor environments and six reflective surface types, under varying lighting conditions across both daytime and nighttime settings. All images are annotated with reflective surface region masks and fine-grained, polygon-level instance annotations for both directly observed and mirror-reflected objects, with depth maps serving as both geometric ground truth for annotation and optional model input. We curate a mirror-induced reasoning ambiguity benchmark, named \textbf{MIRA-Bench} (Mirror-Induced Reasoning Ambiguity Benchmark). MIRA-Bench comprises 2,955 unique QA pairs across eight sub-tasks organized into three cognitive levels: \mbox{Reflection-aware} Perception, Spatial Reasoning, and \mbox{Scene-level} \mbox{Decision-making}, with five object-referencing protocols yielding 14,140 QA samples. We evaluate state-of-the-art VLMs on MIRA-Bench and find that these models consistently underperform across tasks, suggesting that current VLMs do not yet understand mirrors. Common failure patterns include misclassifying whether objects are directly observed or seen through reflections, failing to identify that reflected and directly observed instances correspond to the same physical object, producing erroneous cross-space spatial inferences, and tending to adopt conservative default strategies in action-oriented decisions instead of reasoning about the true scene layout. We will release the dataset and evaluation code to support future research.

Multi-agent debate can improve reasoning, yet often fails to beat simple majority voting. Prior martingale-null theory explains such failures as debate without truth-directed signal. We develop a more complete account of when practical LLM debate succeeds: it needs both correct minority proposals and a readout that can recover them. We formalize this view through latent headroom, the gap between majority vote and proposal-oracle performance, which measures unused correct expertise in an expert society. For readout, we developLatent Verification Debate (LVD), a theory that models candidate proposals as receiving latent verified evidence before influencing final generation. For supply, we study neural-thicket expert societies, where nearby model perturbations provide heterogeneous specialists, and introduce a coverage-based construction that increases complementary proposal supply. Across matched-budget reasoning and multi-discipline benchmarks, our construction increases recoverable headroom, and controlled readout experiments show that gains arise from recovering surfaced correct minorities rather than merely adding interaction. Together, these results identify diverse proposal supply and verification-aware utilization as the two mechanisms that determine when debate helps.

Precisely reproducing a target span verbatim within a long context is a fundamental capability test for LLMs: even when the target string is present in context, models often still drift within the span or fail to stop correctly, revealing limitations in source grounding and boundary control. This capability has broad practical relevance. For instance, code generation often requires embedding sensitive strings such as API keys exactly as they appear in context. In this work, we construct a benchmark to test exact span reproduction ability across contracts, scientific literature, code, and random sequences, and find that current models still make surprisingly frequent errors. To address this, we propose CopyGen, a lightweight copy-constrained plugin for frozen LLMs that turns copying from a byproduct of next-token prediction into a controlled process. Our key idea is to decompose copying into three distinct decisions: when to copy, how far to copy, and when to stop. Concretely, CopyGen first determines whether to enter copy mode, then predicts a tentative copy length, and finally verifies the generated span token by token. During verification, we adaptively adjust the logits of copy continuation tokens to incentivize or suppress continuation, and emit only the longest valid prefix. Across three LLM backbones and four benchmark tasks, CopyGen improves average exact match by 35% relative without model finetuning, and integrates seamlessly with SFT to further improve it by 4%, while substantially accelerating generation. These results suggest that reliable exact copying is not something LLMs obtain for free, but a decoding-time control problem that requires explicit modeling. Code is available at https://anonymous.4open.science/r/CopyGen-A65D.

Pretraining noise in foundation models has been shown to significantly impair the generalization performance of downstream tasks, leading to catastrophic inheritance. However, when downstream tasks arrive sequentially, how pretraining noise affects continual adaptation, which involves the forgetting of previously learned knowledge, remains underexplored. This paper is the first to study this problem. Using asymmetric label noise as a realistic setting, we find that pretraining noise persistently degrades performance on new tasks during continual adaptation and may further exacerbate forgetting of previously adapted tasks. Further analysis shows that the persistent effect of pretraining noise mainly manifests as inherited output contraction, where the logit gap between the true class and competing classes is compressed. To mitigate this issue, we propose Ambiguous Boundary Correction (ABC), a method that performs anchor-guided neighborhood reweighting to correct ambiguous regions induced by inherited output contraction. Experiments on synthetic and real-world noisy pretrained models show that ABC improves robustness to pretraining noise under both domain shifts and class shifts.

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, which tracks the remaining explanatory capacity of the input, 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 all fail to resolve collapse; evidence depletion, which adds stateful residual tracking 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 - preventing collapse and enabling compositional inference.

Large language models reliably follow complex instructions in a single turn, yet across long multi-turn interactions they start strong then gradually lose the thread of the instructions, persona, and rules they were given. This degradation has been measured behaviorally but not mechanistically explained. We trace this failure to a transition between two information channels: accessibility to goal-defining tokens through attention, and a residual channel that carries goal information forward through hidden states. We introduce the Goal Accessibility Ratio (GAR), measuring attention from generated tokens to task-defining goal tokens, and combine it with sliding-window ablations and residual-stream probes. When attention to instructions closes, what survives reveals architecture. Across the architectures we test, this transition produces qualitatively different failure modes: some models preserve substantial goal-conditioned behavior at vanishing attention, others fail despite carrying decodable goal information in their residual stream, and the depth at which this encoding emerges varies dramatically by architecture (from layer 2 to layer 27). A within-model causal ablation that closes the attention channel by force on Mistral collapses recall from near-perfect to eleven percent on a 20-fact retention task and raises persona-constraint violations to levels exceeding the adversarial-pressure baseline despite no user pressure, with both effects emerging at the predictable crossover turn. Linear probes on residual representations recover per-episode recall outcomes with AUC up to 0.99 across all four primary architectures (input embedding: chance), evidencing the second channel and showing its depth profile is architecture-specific. Across multiple model architectures and model scales, we show that the attention channel and the residual channel are separable, and that the gap between attention loss and residual capacity determines whether goal-conditioned behavior survives a long conversation. We provide GAR as a metric, the channel transition framework as a mechanism, and a parametric prediction of when multi-turn instruction-following will fail.

Semivalues are widely used to assign credit and guide data curation decisions, yet their outputs depend on a utility function that is not uniquely determined by the task. We study partition-based decisions, which directly model data curation tasks such as selecting high-quality subsets or flagging noisy examples, under two structurally unavoidable sources of utility underspecification: monotone transformations of performance scores and unconstrained small-sample behavior. We formalize outcome robustness in this setting as invariance of the induced partition across admissible utility respecifications. We establish that partition robustness is necessary for guaranteed success of semivalue-based data selection. To assess this condition in practice, we provide algorithms that certify partition robustness or produce a concrete witness utility demonstrating instability, requiring no utility evaluations beyond those already computed for the semivalues themselves, with formal correctness guarantees under both exact and approximate computation. We show that utilities admitting a coalitionally dominant top-$k$ subset, where high-quality points contribute more than low-quality points across every possible coalition, guarantee robust partition decisions, subsuming the modular utility case previously identified as sufficient. Together, these results provide a practical framework for determining when semivalue-based selection is well-defined under utility ambiguity.