NeurIPS 2026 Papers

Selective unlearning in large language models is increasingly necessary for safety and compliance, yet harmful or memorized content is often latent, entangled, and hard to specify as a clean ``forget set.'' We aim to close this gap by identifying what to forget from the model’s own failure modes rather than relying solely on explicit supervision. We propose Hallucination-Guided Unlearning (HGU), which treats hallucinations as diagnostic signals: it localizes unsupported spans via evidence and internal consensus, traces a sparse set of high-responsibility MLP neurons, and edits them with compact gates under a utility-preserving KL constraint. Across TOFU, HGU achieves the best MU--FE trade-off with Avg of 0.7221 (forget05) and 0.7699 (forget10), and it also attains the top SepS H-Avg (0.4382/0.4715) under mixed forget/retain prompts; on WMDP auxiliary, it yields the strongest forgetting (e.g., 14.2 on Economics) while improving Retain (52.7) and MMLU (57.2). These results suggest that reframing hallucinations as actionable traces enables targeted, post-hoc unlearning that better balances forgetting efficacy and general utility without full retraining.

Hallucination is often viewed as a direct consequence of missing knowledge: a model answers incorrectly when the correct answer is absent from its generation-time distribution, and correctly when it is present. We test this assumption by introducing a semantic notion of answer availability that aggregates token-level variants expressing the same answer concept, and asks whether the correct concept is already available at the moment the model commits to an answer. Across Qwen and Llama models from 0.8B to 72B in both Instruct and Base variants, 16-47\% of Instruct hallucinations occur with substantial probability mass already on the correct concept, and the rate rises monotonically with scale. Comparing such failures against correct generations with matched semantic support, the distinguishing factor is not whether the correct concept is represented, but how its probability is distributed: correct generations concentrate mass on a single surface form, hallucinations disperse it across alternatives. The same sharpening asymmetry extends across multi-token generation and is detectable in pre-generation hidden states. Together, these results identify a single mechanism: instruction tuning sharpens answer commitment with scale, making helpfulness and confident hallucination two consequences of the same underlying disposition.

Diffusion-based image restoration models produce visually compelling results but systematically hallucinate text---generating characters that appear sharp and legible yet are factually incorrect. As text fidelity emerges as a key requirement for real-world image restoration, this failure mode poses a fundamental barrier to wide deployments of diffusion-based models: a model that replaces degraded-but-faithful text with sharp-but-wrong text misleads users and erodes trust. Recent text-aware models mitigate this by feeding text extracted from intermediate outputs or low-resolution (LR) images into the prompt, but they provide no mechanism to detect or correct the hallucinations that still commonly arise. In this work, we present a systematic study on how to improve text restoration during denoising, demonstrating that standard global conditioning alone is inadequate and that optical character recognition (OCR) confidence scores reliably predict word-level correctness. Grounded in these findings, we propose HalluciText, a training-free framework consisting of two complementary interventions: (i) Confidence-Weighted Classifier-Free Guidance (CW-CFG), which concentrates guidance energy on text regions weighted by OCR confidence to improve text recall; and (ii) VLM-Guided Test-Time Guidance (V-TTG), which leverages a vision-language model (VLM) to identify hallucinated regions and corrects them towards a faithful reference, improving text precision. Both operate solely on the noise prediction or latent representation, making them agnostic to the diffusion backbone and text spotting model. We demonstrate the effectiveness of HalluciText on state-of-the-art (SOTA) text-aware models, achieving up to 5.75 points of improvement in end-to-end text recognition F1 on several datasets without degrading image quality.

Hallucination remains a central failure mode of large language models, but existing benchmarks operationalize it inconsistently across tasks such as summarization, question answering, retrieval-augmented generation, and agentic interaction. This fragmentation makes it unclear whether a mitigation that works in one setting actually reduces hallucinations across contexts. Current hallucination benchmarks either require human annotation and fixed references that may eventually be memorized, or rely on naturalistic observations which are often recorded in settings that are difficult to reproduce or test systematically. To enable further research on the root causes of hallucination, we introduce HalluWorld, an extensible benchmark framework grounded in an explicit reference-world formulation: a model hallucinates when it produces an observable claim that is false with respect to this reference world. Building on this view, we construct a family of synthetic and semi-synthetic benchmark environments in which the reference world is fully specified, the model's observable view is controlled, and hallucination labels can be generated automatically by construction. HalluWorld spans multiple settings that are classically representative for AI, i.e., gridworlds, chess, and realistic terminal tasks. This enables controlled variation of key factors such as world complexity, observability, temporal change, and source-conflict policy, allowing us to disentangle hallucinations into more fine-grained error categories. We evaluate frontier and open-weight language models across these settings and find consistent patterns across domains: perceptual hallucination on directly observed information is near-solved for frontier models, while multi-step state tracking and causal forward simulation are still difficult for frontier models, and are not generally solved by extended thinking. In the terminal setting specifically, models also struggle with when to abstain from answering. The uneven profile of failures across probe types and domains suggest that different hallucinations arise from qualitatively distinct failure modes rather than reflecting a single underlying capability. Our results suggest that controlled reference worlds offer a scalable and reproducible path toward measuring and reducing hallucinations in modern language models.

When a text description is extended with an additional detail, image-text similarity should drop if that detail is wrong. We show that CLIP-style dual encoders often violate this intuition: appending a plausible but incorrect object or relation to an otherwise correct description can increase the similarity score. We call such cases \emph{half-truths}. On the verified COCO evaluation set, CLIP prefers the correct shorter description only 41.0\% of the time, and performance drops to 28.2\% when the added detail is a relation. We trace this vulnerability to weak supervision on caption parts: contrastive training aligns full sentences but does not explicitly enforce that individual entities and relations are grounded. We propose CS-CLIP (Component-Supervised CLIP), which decomposes captions into entity and relation units, constructs a minimally edited foil for each unit, and fine-tunes the model to score the correct unit above its foil while preserving standard dual-encoder inference. CS-CLIP raises half-truth accuracy to 78.2\% and improves average performance on established compositional benchmarks by 5.7 points, suggesting that reducing half-truth errors aligns with broader gains in compositional understanding.

Deep reinforcement learning agents progressively lose representational capacity during training: neurons become dormant, removing active capacity from the network, and effective rank collapses, leaving surviving neurons redundant. Existing remedies such as periodic resets, and special neural network architectures, are largely algorithm- or domain-specific. We propose a simple architectural fix, the Hadamard Representation (HR), which replaces a standard hidden layer with the element-wise product of two independently parameterized layers. HR operates through two complementary mechanisms. First, it reduces the probability of a neuron becoming dormant, which is particularly valuable for continuously differentiable activations such as $\tanh$: unlike dormant ReLU neurons, which are effectively pruned, saturated $\tanh$ neurons silently corrupt downstream layers by turning their outgoing weights into fixed biases. Second, independently of dormancy, the multiplicative structure captures richer feature interactions and increases effective rank without widening the layer. We evaluate HR across five algorithms and three domains: DQN, PPO, and PQN on pixel-based discrete-action Atari, SimbaV2 on state-based continuous control, and MR.Q on visual continuous control. HR consistently improves performance over the strong baselines without any hyperparameter tuning, and gains persist against parameter-matched wider variants, ruling out parameter count as an alternative explanation.

Aligning autonomous agents with human intent remains a central challenge in modern AI. A key manifestation of this challenge is reward hacking, whereby agents appear successful under the evaluation signal while violating the intended objective. Reward hacking has been observed across a wide range of settings, yet methods for reliably measuring it at scale remain lacking. In this work, we introduce a new evaluation paradigm for measuring reward hacking. Whereas prior studies have primarily analyzed it post hoc by inspecting agent trajectories, we instead embed detectable reward hacking opportunities directly into environments. This makes their exploitation verifiable by design, enabling deterministic and automated measurement of whether and how agents exploit such vulnerabilities. We instantiate this approach in TextArena and release Hack-Verifiable TextArena, a testbed in which reward hacking can be measured reliably. Using this benchmark, we analyze reward hacking behavior across language models in diverse environments and settings.

We propose, HaNDF, object-conditional neural distance fields as data-driven priors for modeling the plausible hand–object configurations, in the product manifold of hand articulations. While recent unconditional generative priors showed success in modeling human pose, hands are versatile articulated bodies often found in interaction with other objects. This makes it crucial to develop a generative prior which can model the plausible hand poses conditioned on the object under interaction. To this end, our HaNDF proposes a geometry-aware, object-conditioned neural distance field, representing the plausible articulations in the zero-level set of a conditional field. We leverage tools from Riemannian geometry to (i) represent plausible hands in the zero level-set of an object-induced neural field, and (ii) project a given pose onto the this field. Using HaNDF, we can also optimize for a plausible condition, determining the object under interaction. Our extensive evaluations demonstrate that our HaNDF provides a flexible and powerful prior for hand–object interaction, suitable for applications such as pose refinement, reconstruction under occlusion, and physically plausible manipulation synthesis.

World models enable model-based planning through learned latent dynamics, but imagined rollouts often become unstable as the planning horizon grows or the dynamics distribution shifts. We argue that this instability arises from two missing structures in planner-facing latents: history-conditioned memory for approximate Markov completeness, and geometric organization that separates configuration, momentum, and task semantics. We propose HaM-World, a structured world model that decomposes the latent state into a canonical $(q, p)$ subspace and a context subspace $c$, while incorporating Mamba selective state-space memory as a history-conditioned input to the same latent dynamics. Within this unified interface, $(q, p)$ evolves under a Soft-Hamiltonian dynamics composed of an energy-derived Hamiltonian vector field and learnable residual and control dynamics, while $c$ captures semantic, dissipative, and non-conservative factors. This design provides the planner with a single latent representation shared across dynamics prediction, reward and value estimation, imagined rollouts, and cross-entropy method (CEM) planning. On four DeepMind Control Suite tasks, HaM-World achieves the highest average AUC (+9.5% over strong baselines), reduces long-horizon rollout error to 45% of a competitive model, and wins 11 out of 12 $k \in \{3,5,7\}$ rollout MSE metrics. Under 12 out-of-distribution perturbations, HaM-World consistently attains the highest returns, with average gains of 10.2% on Finger Spin and 13.6% on Reacher Easy. Mechanism diagnostics further demonstrate bounded energy drift under action-free rollouts, structured energy variation under policy control, and coherent control-induced energy transfer, supporting the effectiveness of the proposed Soft-Hamiltonian latent dynamics. Code: https://anonymous.4open.science/r/HaM_World-47CD

Large language models (LLMs) are increasingly adopted for software and hardware design, yet these domains are still evaluated separately. Software benchmarks typically assume fixed hardware targets, while hardware benchmarks focus on component-level optimization without considering the full hardware-software stack. Consequently, no existing benchmark evaluates whether an LLM agent can perform end-to-end, system-level hardware-software co-design. Such a process requires: 1) analyzing applications to identify kernels requiring acceleration, 2) designing and \emph{integrating} heterogeneous accelerators into a System-on-Chip (SoC) under resource constraints, and 3) mapping kernels onto the generated accelerators. We present \hsco{}, an end-to-end hardware-software co-design benchmark for accelerator-rich heterogeneous SoC generation. Built upon an open-source SoC platform with a curated repository structure, HSCO-Bench evaluates the ability of LLMs to jointly optimize software and hardware stacks, producing SoC prototypes deployed on the AMD Virtex-7 FPGA VC707 Evaluation Kit. Experimental results show that end-to-end integration remains challenging for current models, with widespread failures. Among the five frontier models evaluated, only two of them could successfully generate valid SoC prototypes. Yet, even in these successful instances, the generated designs are far from optimal. While we observe a promising peak speedup of 16.22x, the maximum additional resource utilization reaches only 23.67%. This highlights that while state-of-the-art models demonstrate an emerging capability for hardware acceleration, they still heavily underutilize the available hardware capacity, leaving substantial room for future optimization. To the best of our knowledge, HSCO-Bench is the first benchmark targeting this complete co-design flow, enabling LLMs to jointly reason about and modify both the software and hardware stacks of heterogeneous SoCs.

Self-supervised multimodal representation learning has achieved remarkable success across diverse domains, yet capturing synergistic information remains challenging due to the complexity of cross-modal interactions. Unlike the shared information across individual modalities, synergy arises when task-relevant signals emerge only from the joint configuration of multiple modalities and cannot be recovered from any modality in isolation. This work focuses on how to preserve the information capacity for such synergistic signals in multimodal representations. The key observation is that synergistic information is reflected in higher-order statistical dependence among modalities, which provides a principled target for explicitly modeling joint interactions. Motivated by this insight, we propose Higher-order Representation and Information Learning (HRIL), which constructs an empirical cross-moment tensor over modality embeddings to represent multi-way interactions. HRIL employs Tucker decomposition to disentangle non-separable interactions into a core tensor, complemented by a synergy-aware regularizer that prevents energy concentration and preserves higher-order coupling for synergistic information capture. Experiments on synthetic and real-world benchmarks demonstrate consistent improvements over existing multimodal contrastive methods, with notable gains on tasks dominated by synergistic interactions.

Federated self-supervised learning (FSSL) enables collaborative training of self-supervised representation models without sharing raw unlabeled data. While it serves as a crucial paradigm for privacy-preserving learning, its security remains vulnerable to backdoor attacks, where malicious clients manipulate local training to inject targeted backdoors. Existing FSSL attack methods, however, often suffer from low utilization of poisoned samples, limited transferability, and weak persistence. To address these limitations, we propose a new backdoor attack method for FSSL, namely Hallucinated Positive Entanglement (HPE). HPE first employs hallucination-based augmentation using synthetic positive samples to enhance the encoder’s embedding of backdoor features. It then introduces feature entanglement to enforce tight binding between triggers and backdoor samples in the representation space. Finally, selective parameter poisoning and proximity-aware updates constrain the poisoned model within the vicinity of the global model, enhancing its stability and persistence. Experimental results on several FSSL scenarios and datasets show that HPE significantly outperforms existing backdoor attack methods in performance and exhibits strong robustness under various defense mechanisms.

Large Language Models (LLMs) have shown strong capabilities in code generation and reasoning, motivating extensive benchmarks for evaluating functional correctness. However, performance optimization in High-Performance Computing (HPC) poses fundamentally different challenges that are not captured by existing evaluations: HPC performance depends on low-level code, intricate memory access patterns, and computational kernels embedded within sophisticated execution contexts. We introduce HPC-Bench, a performance-grounded benchmark for evaluating LLMs as optimizers of computational kernels in realistic HPC workloads, including $114$ workloads across $16$ HPC computational motifs. HPC-Bench provides a fully automated, end-to-end evaluation pipeline that combines program-level correctness verification with two correctness-gated metrics, $\mathrm{Fast}@k$ and $\mathrm{Speedup}@k$, to characterize reliability and speedup under a fixed sampling budget. Our evaluation of three representative LLMs across serial CPU, OpenMP, and CUDA settings shows that stricter speedup requirements substantially reduce the probability of finding a correct and fast optimization. HPC-Bench is available at https://github.com/neuripspaper2026/NeurIPS_26.

While Graph Neural Networks (GNNs) have proven highly effective at modeling relational data, pairwise connections cannot fully capture multi-way relationships naturally present in complex real-world systems. In response to this, Topological Deep Learning (TDL) leverages more general combinatorial representations--such as simplicial or cellular complexes--to accommodate higher-order interactions. Existing TDL methods often extend GNNs through Higher-Order Message Passing (HOMP), but face critical scalability challenges due to the steep complexity overhead of propagating messages through combinatorial structures. To overcome this limitation, we propose HOPSE (Higher-Order Positional and Structural Encoder), a framework \emph{free of message passing layers} that uses Hasse graph decompositions to derive efficient and expressive encodings over \emph{arbitrary higher-order domains}. Notably, HOPSE scales linearly with the size of combinatorial representations while preserving the expressive power and permutation equivariance of the HOMP approaches. Experiments on molecular and topological benchmarks show that it matches or surpasses state-of-the-art performance while consistently achieving speedups over HOMP-based models, opening a new path for scalable TDL. The code is available at https://anonymous.4open.science/r/TopoBench-BF89.

Estimating physical pressure from vision is essential for understanding contact-rich hand-object interaction. However, prior vision-based pressure estimation methods are largely limited to planar surfaces and single image input, making them difficult to apply to dynamic hand-object interaction with diverse objects. We instead formulate pressure estimation as a hand-centric video prediction problem with monocular video as input. This formulation predicts temporally evolving per-vertex normal pressure and contact directly on the hand mesh, yielding a unified output space independent of object shape and sensor layout. Building on this formulation, we propose HOPE, a framework with two key components. First, we lift tactile-glove pressure, planar-sensor pressure, and distance-based hand-object contact annotations into a shared hand vertex space, allowing bare-hand contact data to regularize pressure learning where metric labels are unavailable. Second, we introduce a vertex-anchored video transformer that treats each vertex as a persistent token, aggregates visual features and hand pose over time, and uses a contact-gated pressure head to enforce that pressure vanishes without contact. Experiments on OpenTouch, PressureVisionDB, and hand-object contact benchmarks demonstrate that HOPE supports a common hand-centric representation across object-pressure, surface-pressure, and contact-supervised HOI settings. Despite using metric pressure supervision primarily from gloved-hand videos, HOPE generalizes to bare-hand egocentric and in-the-wild videos, producing joint contact and pressure predictions beyond the scope of contact-only or planar-pressure baselines.

Human-object-centric video personalization (HOCVP) is a core task within subject-driven video generation. However, existing methods suffer from two key limitations. First, most approaches focusing on $\textit{inter-subject}$ personalization still struggle to strike a balance between high subject fidelity and accurate interaction patterns between humans and diverse objects, especially when objects represent abstract concepts such as logos. Second, while $\textit{intra-subject}$ references ($\textit{e.g.}$, OCR maps, multi-view inputs) are expected to enhance subject fidelity, most existing works lack mechanisms to understand such latent correspondence. To address both challenges, we propose HOMIE, an HOCVP framework that tackles both inter- and intra-subject input settings in a unified manner. Compared to previous approaches, HOMIE proposes a better MLLM integration strategy to extract knowledge of reference-level relationships without compromising the controllability of text encoders or incurring costly re-alignment. Specifically, we introduce global multimodal guidance within self-attention to better align MLLM-derived semantic features with VAE tokens. Furthermore, we propose modality-reference embedding to differentiate tokens from MLLM features and VAE tokens and associate intra-subject reference image tokens. Extensive experiments validate that our method achieves state-of-the-art performance across various HOCVP tasks. Anonymous project page with videos is available at https://homie-demo.github.io/.

Streaming 3D reconstruction enables long-sequence scene understanding by processing frames sequentially, but remains challenging when historical information is compressed into a fixed-size recurrent state. Existing training-free extensions of implicit-memory models mainly improve stability by modulating update magnitude, thereby controlling how strongly each incoming observation modifies the state. However, the geometric intent of the update, namely the direction in which the observation drives the recurrent state, remains tied to the current-frame candidate and can accumulate directional bias over long sequences. We propose HIST3R, a training-free framework that improves recurrent state updating through a direction--magnitude decomposition. HIST3R uses historical recurrent states to rectify the update direction, providing a more reliable geometric intent, and uses observation--history visual discrepancy to modulate the update magnitude, yielding an adaptive update gain. This design makes recurrent updates less dependent on isolated frame-level evidence while preserving the bounded-memory advantage of implicit streaming reconstruction. Across camera pose estimation, 3D reconstruction, and video depth estimation benchmarks, HIST3R consistently improves long-horizon performance without additional training. At 1000 input frames, HIST3R achieves relative ATE reductions of 23.87% on ScanNet and 33.65% on TUM-Dynamics over the strongest baseline; at 500 frames, it improves long-sequence 3D reconstruction over the strongest CUT3R-based variant by 56.52% on 7-Scenes and 33.96% on NRGBD.

Long video understanding with multimodal language models suffers from three compounding bottlenecks: heavy decode cost to obtain dense RGB frames, quadratic token growth with frame count, and weak motion perception under sparse keyframe sampling. Existing remedies either prune visual tokens after decoding, which leaves the expensive RGB pipeline untouched, or they discretise codec motion vectors with a generative tokeniser, which couples motion modelling to a heavy pre-training stage. In this paper we present HIMMEL, a hierarchical video-language framework that allocates semantic and motion capacity along separate paths. A small set of sparse anchor I-frames is routed to the expensive host ViT to ground object identity and scene layout, while the far denser inter-frame intervals are encoded by a lightweight compressed-domain tri-stream adapter that distils motion evidence from motion-vector maps, residual maps, and an I-frame context branch into aligned motion tokens. These tokens are injected into the LLM via a differentiable placeholder mechanism, after a dedicated Stage-1 contrastive alignment that places the motion representation in a geometry compatible with the frozen visual backbone. We further show that an InfoNCE alignment objective beats MSE regression by $+1.5$ pp because it preserves directional motion structure rather than collapsing onto the mean of visual deltas. On Video-MME, HIMMEL surpasses the dense 32-frame baseline by $+2.3$ pp ($61.2 \to 63.5\%$) while using $3.6\times$ fewer context tokens and running end-to-end in $1.03$ s per question instead of $2.75$ s. Switching to a stronger Qwen3-VL-8B host pushes HIMMEL to $64.9\%$, narrowing the gap to much larger proprietary systems while keeping inference on a single consumer GPU. Extensive ablations across stream composition, motion-encoder family, fusion mode, alignment objective, anchor count, LoRA rank, and video duration confirm that the full tri-stream is necessary and sufficient for the observed gains, and that the benefit grows with video length. We release training and evaluation pipelines to support reproducibility.

The discrepancy between controlled synthetic distortions and complex real-world degradations poses a significant domain shift challenge for No-Reference Image Quality Assessment (NR-IQA). While Unsupervised Domain Adaptation (UDA) methods aim to bridge this gap, they heavily suffer from retraining overhead and error propagation during iterative pseudo-labeling. Inspired by the dual-process theory of human cognition, we propose Hierarchical Intuitive and Fuzzy Consensus (HIFC), a completely train-free cross-domain IQA framework. HIFC employs an asymmetric perceptual decomposition to disentangle semantic and distortion features. It derives the final quality score through a synergy of two branches: a memory consensus branch that retrieves a stable Fuzzy Quality Centroid from a synthetic reference gallery, and an intuitive branch that captures aesthetic polarization via zero-shot vision-language alignment. Extensive experiments demonstrate that, without any parameter updates or target-domain adaptation, HIFC achieves highly competitive results against state-of-the-art UDA methods. Most notably, it exhibits exceptional robustness and generalization in the challenging synthetic-to-authentic cross-domain setting. The code will be available upon acceptance.