NeurIPS 2026 论文列表

State-of-the-art pathology foundation models, trained on millions of histology tiles, can fail to preserve tissue similarity when comparisons cross slide or institution boundaries. We show that general-purpose multimodal LLMs, without being trained as pathology foundation models, consistently outperform these specialized models in cross-domain histological similarity judgments. Using a relative similarity framework that we release as the MOSAIC (Model Similarity Assessment across Institutions and Cohorts) benchmark, we evaluate 17 models across 6 datasets and find that pathology encoders often rank same-institution, different-disease tiles as more similar than same-disease, different-institution tiles, a clinically dangerous failure mode invisible to standard within-domain evaluations. LLMs appear less susceptible to this failure, likely because they perform semantic visual comparison of morphology and tissue architecture rather than relying on shortcut features tied to acquisition context. Scaling training data does not resolve the problem for pathology encoders, implicating the learning objective rather than data coverage. Our results expose a fundamental robustness gap in current pathology foundation models and establish multimodal LLMs as a viable alternative for cross-institutional retrieval, dataset harmonization, and multi-site quality control. Code and data will be released upon acceptance.

This position paper argues that multimodal foundation models should use brain recordings as privileged internal-state supervision during training. External behavior alone leaves human-like cognition, robustness, and reasoning weakly constrained. Neural recordings can serve as a training-time modality that shapes representations, latent-state dynamics, private computation policies such as reasoning tokens, and robustness before distillation into agents that operate from ordinary sensory, language, and action streams. We develop this position through Foundational Human Modeling (FHM), a program centered on scaling generative brain models, aligning heterogeneous neural modalities through shared latent geometry, inserting brain tokens into unified multimodal token streams, and testing whether the privileged channel improves distilled agents after removal. The argument is grounded in privileged-information theory and prior work showing that neural or human-derived targets can improve robustness, semantic transfer, data efficiency, and behavioral similarity after auxiliary signals are removed.

Bioassay activity prediction is often data-limited because drug-discovery datasets rely on time-consuming and expensive wet-lab experiments for data generation and evaluation. This challenge has inspired recent research into molecular foundation models (MFMs), which aim to encode general-purpose chemical knowledge into molecular representations that generalize well in data-constrained scenarios. This paper presents Monroe, a new MFM with several innovations over the existing state of the art: increased scale allowing pre-training on over 80 million molecules from the PM6 quantum chemistry dataset; improved graph representation of stereochemistry; improved training losses including conformer denoising and embedding decorrelation; improved multi-task learning; and the use of a prior-data-fitted model (TabPFN) for downstream in-context prediction. Our evaluations use a principled pairwise comparison framework that measures statistically significant performance differences. Across established Polaris benchmarks, Monroe matches or exceeds existing MFMs, while on activity cliff benchmarks, designed to assess utility for molecular discovery, it achieves significant improvements over prior methods. Finally, ablation and transfer experiments show that PFN-based downstream predictors also substantially improve two leading existing models, MiniMol and CheMeleon, yielding new state-of-the-art variants we call MiniMol_PFN and Chemeleon_PFN, suggesting that our downstream adaptation strategy generalizes beyond Monroe. Full source code is at https://anonymous.4open.science/r/monroe-0E0D, and will be openly released.

While autoregressive models dominate the LLM landscape, Discrete Diffusion Language Models have emerged as a compelling alternative due to their advantages in parallel decoding and bidirectional contextual modeling. To enhance scalability and efficiency, recent works have integrated the Mixture-of-Experts architecture into the DLM framework. However, MoE-DLMs remain underexplored. In this work, we present an empirical study revealing that MoE-DLMs exhibit not only task-level expert specialization, but also exhibit specialization across different mask-rate diffusion regimes. We further show that during fine-tuning, gradients induced by mismatched mask rates can interfere with the updates of regime-specialized experts, leading to suboptimal adaptation. Motivated by this, we propose MaGM, an adaptive mask-conditioned gradient masking method for MoE-DLM fine-tuning, which dynamically masks expert parameter updates based on the mask rate of training instance. Experiments demonstrate that MaGM consistently outperforms standard full-parameter fine-tuning for diffusion language models, validating the benefits of regime-aware expert adaptation. Our code will be publicly released.

Quantifying mouse behavior from pose is a foundational tool in neuroscience, ethology, drug discovery, and animal welfare. Self-supervised foundation models are a natural fit, but a foundation model for mouse behavior must handle social behavior - many behaviors that matter (aggression, courtship, mounting, allogrooming) are inherently multi-animal. Existing self-supervised pose models fall short: they encode each animal independently and discard interaction signal, hard-code the animal count, and pretrain one dataset at a time. We introduce MICE (Multi-animal Interaction Context Encoder), a hierarchical foundation model for mouse behavior that pretrains jointly on six mouse-pose corpora spanning one to four mice and 7 to 27 keypoints. MICE combines a hierarchical individual encoder capturing per-mouse kinematics at multiple temporal scales with a Perceiver-style social encoder whose learned latent codebook decouples parameter count from animal count, so a single checkpoint serves any group size without retraining. Across four evaluation protocols on all six datasets, MICE matches or exceeds prior keypoint-based baselines, and layer-wise probing shows the social stage incrementally improves discriminability beyond the individual encoder. Code and weights will be released.

Robot policy pretraining based on human videos is crucial for improving the policy’s generalization ability. One main challenge for this task is the lack of explicit action-relevant representations in such unlabeled data. Recent works tend to estimate 3D hand motion trajectories from videos using 3D foundation models (3D FMs). However, they keep solely the sparse trajectories as supervision, ignoring the rich, fine-grained information produced during the trajectory extraction process. To address this issue, we propose $\textbf{LSA}$, a hierarchical robot policy pretraining framework that includes three phases: $\textbf{L}$ift, $\textbf{S}$ee, and $\textbf{A}$ct. Specifically, the Lift phase elevates 2D video observations into dense 3D representations by adaptively aligning features in the policy's shallow layers with the intermediate representations extracted from multiple 3D FMs. Building upon these lifted representations, the See phase equips the policy with explicit geometric and interactive awareness through dual guidance. It introduces depth-map reconstruction to help the model comprehend 3D spatial layouts and utilizes hand region cues to explicitly supervise the encoder’s attention toward task-relevant interaction hotspots. Empowered by the dense 3D representation learning and precise spatial guidance, our policy achieves robust and accurate robotic manipulation in the Act phase. Extensive experiments on diverse simulation and real-world tasks demonstrate that LSA significantly outperforms current state-of-the-art approaches. The code of this work will be released soon.

Low-Rank Adaptation (LoRA) is the prevailing approach for efficient large language model (LLM) fine-tuning. Building on this paradigm, recent studies have proposed alternative initialization strategies and architectural modifications, reporting substantial improvements over vanilla LoRA. However, these gains are often demonstrated under fixed or narrowly tuned hyperparameter settings, despite the known sensitivity of neural networks to training configurations. In this work, we systematically re-evaluate nine representative LoRA variants alongside vanilla LoRA through extensive hyperparameter searches. Across tasks spanning mathematical reasoning, commonsense reasoning, code generation, and instruction following at diverse model scales, we find that different LoRA methods favor distinct learning rate ranges. Crucially, once learning rates are properly tuned, all methods achieve similar peak performance (within 1-2\%), with only subtle rank-dependent behaviors. These results suggest that vanilla LoRA remains a competitive baseline and that improvements reported under a single training configuration may not reflect consistent methodological advantages. Finally, a second-order analysis attributes the differing optimal learning rate ranges to variations in the largest Hessian eigenvalue, aligning with classical learning theories.

Geometric foundation models, such as the Visual Geometry Grounded Transformer (VGGT), provide strong 3D priors from unposed images. However, such models operate purely in a feed-forward, deterministic regime, \ie~they cannot generate plausible geometry beyond what the input views directly support. Generative models for 3D scenes, on the other hand, must rely on strong geometric priors to produce coherent outputs from sparse inputs. We bridge these two paradigms by performing flow matching directly in VGGT's latent space, leveraging its learned 3D priors without committing to any explicit downstream representation such as Gaussians, meshes, or video-VAE latents. This requires respecting the latent geometry: VGGT tokens occupy a product of high-dimensional hyperspheres on which standard Euclidean flow matching fails. We address this with a Riemannian Flow Matching framework defined on a product manifold of four hyperspheres, aligned with VGGT's multi-scale encoder, which keeps generated tokens on the valid data manifold required by the frozen decoding heads. On RealEstate10K and ScanNet++, our method achieves strong performance against recent scene generation baselines in both per-view 3D geometry and appearance, establishing latent-space flow matching on geometric foundation models as a viable paradigm for 3D generation.

Domain adaptation faces a fundamental paradox in the cold-start regime. When target data is scarce, statistical methods fail to distinguish relevant source domains from irrelevant ones, which often leads to negative transfer. In this paper, we address this challenge by leveraging expert textual descriptions of the target domain, a resource that is often available but overlooked. We propose a probabilistic framework that translates these semantic descriptions into a choice model, namely a Language-Induced Prior (LIP), that learns the preferences from a pre-trained Large Language Model (LLM). The LIP is then integrated into an Expectation-Maximization (EM) algorithm to identify source relevance. Methodologically, this framework is compatible with any parametric model where a likelihood is available. It allows the LIP to guide the selection of sources when target signals are weak, while gradually refining these choices as samples accumulate. Theoretically, we prove that the estimator roughly matches an oracle cold-start MSE under a correct prior, while remaining asymptotically consistent regardless of the quality of the LIP. Empirically, we validated the framework on a descriptive (Gaussian estimation), a predictive (C-MAPSS dataset), and a prescriptive task (MuJoCo Hopper).

Orthogonal parameter-efficient fine-tuning adapts pretrained weights through structure-preserving multiplicative transformations, but existing methods often conflate two distinct design choices: the subspace in which adaptation occurs and the transformation applied within that subspace. This paper introduces LOFT, a low-rank orthogonal fine-tuning framework that explicitly separates these two components. By viewing orthogonal adaptation as a multiplicative subspace rotation, LOFT provides a unified formulation that recovers representative orthogonal PEFT methods, including coordinate-, butterfly-, Householder-, and principal-subspace-based variants. More importantly, this perspective exposes support selection as a central design axis rather than a byproduct of a particular parameterization. We develop a first-order analysis showing that useful adaptation supports should be informed by the downstream training signal, motivating practical gradient-informed support selection strategies. Across language understanding, visual transfer, mathematical reasoning, and multilingual out-of-distribution adaptation, LOFT recovers principal-subspace orthogonal adaptation while gradient-informed supports improve the efficiency–performance trade-off under matched parameter, memory, and transform budgets. These results suggest that principled support selection is an important direction for improving orthogonal PEFT.

We consider the problem of learning to adapt a foundation model in a federated setting, particularly the most realistic and general setting: 1) When the local data sets are sampled from different distributions but we want to learn a globally adapted model, 2) Where local agents enter and leave the federation asynchronously at each time tick which is beyond the control of the learning algorithm, and 3) Where the goal is continuous adaptation so that after each time tick, the learned adapter generalizes accurately for all participants that have been seen during training. We propose a simple idea called federated *library-based adaptation* (*LEAN*) for this setting. In library-based adaptation, the system maintains a pool, or "library" of rank-1 "basis pairs". Groups of these pairs are distributed to local sites to be trained in LoRA fashion, and then separated and *shuffled* at communication. Library-based adaptation is designed to avoid problems with more conventional methods, such as those based on averaging. In particular, we demonstrate *LEAN* outperforms traditional averaging baselines in both communication and computation cost efficiency across a broad range of important settings, including heavy data skewness and high asynchrony.

Inherent temporal heterogeneity, such as varying sampling densities and periodic structures, has posed substantial challenges in zero-shot generalization for Time Series Foundation Models (TSFMs). Existing TSFMs predominantly rely on massive parameterization to absorb such heterogeneity, as their static tokenization and positional encoding schemes entangle diverse temporal patterns into a fixed representation space, encouraging memorization rather than adaptation. To address this limitation, we propose Kairos, a flexible and parameter-efficient TSFM dedicated to forecasting tasks, which decouples temporal heterogeneity from model capacity through a novel tokenization perspective. Kairos introduces a dynamic patching tokenizer and a mixture-of-size encoding that adapt observational granularity to local information density, enabling fine-grained temporal abstraction without increasing model width or depth. In addition, we design a multi-granularity positional embedding based on dynamic rotary encodings, which conditions on instance-level spectral features and temporal structure induced by dynamic patching tokenization, allowing robust modeling of diverse temporal dependencies. Trained on a novel Predictability-Stratified Time-Series (PreSTS) corpus, Kairos achieves superior zero-shot performance with substantially fewer parameters on two mainstream benchmarks, GIFT-Eval and Time-Series-Library.

Wearable motion sensing provides a continuous and scalable window into human behavior and health, making it a natural fit for foundation models, yet its pretraining and scaling principles remain poorly understood. Prior work studies isolated design choices, such as sensor placement or sampling frequency, often under fixed settings and narrow downstream tasks that fail to capture real-world sensing diversity. We introduce Inertia-1, a fully open exploration of wearable motion foundation models. Using massive corpora of accelerometer data from global sources spanning more than 18.2M hours, we build a controlled framework for studying the full lifecycle of wearable motion foundation models, covering data choices such as sensor modality, device placement, sampling rate, window length; model choices such as architectures and model size; and training choices such as pretraining objective and data scale. Extensive evaluations across 15 datasets spanning human activity recognition, freezing-of-gait detection, and disease prediction reveal intriguing findings for building motion foundation models that generalize across tasks and sensing conditions. Collectively, Inertia-1 not only presents state-of-the-art recipes for diverse downstream tasks, but also serves as a comprehensive, practical, and open cookbook for wearable motion representation learning.

We present ImmuVis, a family of efficient foundation models for imaging mass cytometry (IMC), a high-throughput multiplex imaging technology that handles molecular marker measurements as image channels and enables large-scale spatial tissue profiling. Unlike natural images, multiplex imaging lacks a fixed channel space, as real-world marker sets vary across studies, violating a core assumption of standard vision backbones. To address this, ImmuVis introduces marker-adaptive hyperconvolutions that generate convolutional kernels from learned marker embeddings, enabling a single model to operate on arbitrary measured marker subsets without retraining. We pretrain ImmuVis on the largest IMC dataset to date, IMC17M (28 cohorts, 24,405 images, 265 markers, over 17M patches), using self-supervised masked reconstruction. ImmuVis outperforms state-of-the-art baselines and ablations in virtual staining and downstream classification tasks at substantially lower compute cost than transformer-based alternatives, and is the sole model that provides calibrated uncertainty via a heteroscedastic likelihood objective. These results position ImmuVis as a practical framework for real-world IMC modeling.

Self-play fine-tuning enables large language models to improve beyond supervised fine-tuning without additional human annotations by contrasting annotated responses with self-generated ones. Many existing methods rely on a fixed divergence regime. SPIN is closely related to a KL-based regime, SPACE to a Jensen-Shannon-style objective via noise contrastive estimation, and SPIF to $\chi^2$-regularized self-play. Since these divergences exhibit different strengths depending on the distributional gap between model and target, no single choice appears to provide favorable learning dynamics across training stages. We propose IRIS (Interpolative R\'enyi Iterative Self-play), a R\'enyi-based self-play fine-tuning framework with a continuously adjustable objective. IRIS decomposes into two independent tilted risk terms over annotated and synthetic data, with exponential importance weights controlled by the order parameter $\alpha$. We show that several self-play objectives can be interpreted as limiting or representative regimes at particular values of $\alpha$, providing a unified theoretical perspective on these methods. An adaptive order schedule further adjusts $\alpha$ to the distributional gap, shifting from sharper importance weighting early in training to smoother refinement near convergence. Theoretically, we establish the fixed-point property of IRIS and analyze how $\alpha$ controls gradient concentration. Experiments on Zephyr-7B and Qwen2.5-3B across ten benchmarks show that IRIS improves upon baselines, reaching 44.57\% average score with gains across iterations. In our setting, IRIS with only 26$k$ annotated samples surpasses standard supervised fine-tuning trained on the full 200$k$ dataset.

Class-incremental learning aims to learn a stream of new classes while preserving knowledge from prior tasks, but suffers from catastrophic forgetting when the model is adapted across sessions. Recently, approaches leveraging vision-language pretrained models such as CLIP have gained increasing popularity in this setting, due to the strong transferable representations of foundation models. Most CLIP-based approaches resist forgetting by retaining data-derived class memory such as replay samples, visual prototypes, or class distribution information. In contrast, we ask whether stable adaptation can be achieved by constraining adapter-induced feature displacement without retaining any data-derived class memory. We propose \textbf{H}yperbolic \textbf{D}isplacement-\textbf{C}onstrained \textbf{A}daptation (HDCA), which treats forgetting as a consequence of uncontrolled residual displacement induced by sequential adapter updates. HDCA keeps the direction of each adapter update but compresses its radial magnitude, so small updates remain almost unchanged while large or accumulated updates are suppressed. At the per-task level, a hyperbolic low-rank adapter makes each task residual grow sublinearly. At the multi-task level, a cumulative compression stage further suppresses the summed residual across sessions. On the classifier side, HDCA introduces a parameter-free Poincar\'e distance classifier that preserves cosine-based ranking for fixed adapted features but reshapes the training gradient in high-similarity regions. We conduct experiments on five standard class-incremental datasets, and the results comprehensively validate the effectiveness of HDCA without retaining any data-derived class memory during training and inference.

Post-training has become essential for adapting large language models (LLMs) to complex downstream behaviors, including instruction following, preference alignment, and multi-step reasoning. Reinforcement learning with verifiable rewards (RLVR) has recently emerged as a particularly effective post-training paradigm for improving reasoning capabilities, with critic-free algorithms such as GRPO and GSPO enabling scalable optimization. However, RLVR post-training with full fine-tuning (FFT) requires substantial GPU memory and incurs high training costs. Although parameter-efficient fine-tuning (PEFT) methods, such as Low-Rank Adaptation (LoRA), effectively reduce computational costs, they often suffer from a noticeable performance gap compared to full fine-tuning in post-training for complex reasoning tasks. In this paper, we propose Hybrid-LoRA, an efficient hybrid post-training framework that selectively applies full fine-tuning to a small subset of modules less suited to low-rank adaptation, while adapting the remaining components with LoRA. We introduce a novel Hybrid-LoRA Score to rank candidate modules according to their sensitivity to low-rank adaptation under a fixed parameter budget. Experiments show that Hybrid-LoRA closely matches full fine-tuning performance under a 10\% full fine-tuning module budget, with the remaining candidate modules adapted by LoRA, consistently outperforming four state-of-the-art PEFT post-training baselines, achieving improvements of up to 5.65\% and on average 4.36\% over the best baseline.

Large-scale foundation models (FMs) in remote sensing (RS) (denoted as RS FMs) are developed following paradigms established in computer vision (CV), yet the validity of transferring CV scaling laws to RS has not been systematically examined. We hypothesize that RS FMs enter an overparameterized regime at substantially smaller scales than their CV counterparts, with task-relevant information encoded redundantly across model dimensions. To test this hypothesis, we apply post-hoc slimmability, uniform width reduction of pretrained encoder transformer blocks, as a tool to measure representational redundancy across eight state-of-the-art RS FMs on classification, segmentation, and change detection tasks. RS FMs retain 69% to 109% relative accuracy on RS datasets under aggressive width reduction, while masked autoencoder (MAE) and DINOv2 pretrained on natural images (denoted as CV MAE and CV DINOv2) degrade sharply on ImageNet subsets of matched class count over the same range of computational requirements. A CV MAE evaluated directly on the same RS datasets narrows but does not close the gap, indicating that both dataset characteristics and domain-specific pretraining contribute to the differences between the models. Mechanistic analyses such as feature correlation, explained variance, and effective dimensionality indicate that task-relevant variance concentrates in few principal components and is redundantly encoded across model dimensions. We further show that learned slimmable training improves over post-hoc slimmability for contrastive objectives, while reconstruction-based objectives do not benefit from current slimmable training protocols. Our findings establish post-hoc slimming as a practical deployment strategy for resource-constrained RS applications and as a diagnostic tool for representational redundancy in RS FMs. Upon acceptance, we will publish all code.

Low-rank adaptation (LoRA) has become the standard method for parameter-efficient fine-tuning of large pretrained models, yet theoretical explanations for why low-rank updates suffice remain incomplete. Most existing analyses rely on the Neural Tangent Kernel (NTK) regime, which linearizes the network around the pretrained weights. In this work, we take a different approach that requires no linearization: our key observation is that, once the backbone is frozen, the loss sees each adapted weight matrix only through a finite-rank linear map induced by the frozen activations. Trace-norm regularization then forces the optimizer to pick a low-rank matrix among all matrices that produce the same observed output. This yields global minimizers whose ranks are controlled by the rank of this map, which we call the \emph{bottleneck rank}, rather than the sample size. For multi-head attention, we obtain tighter rank upper bounds by exploiting two additional invariances: row-wise softmax is insensitive to additive row constants in the score matrix, and value updates are observable only after attention weighting and frozen output projection. For the factorized LoRA objective with weight decay, we prove a complementary result on the nonconvex side: after an arbitrarily small generic positive semidefinite perturbation, every first-order stationary point is rank-deficient once the LoRA rank exceeds a threshold determined by the bottleneck rank. Finally, we show that global minimizers of the perturbed objective are near-optimal for the original trace-norm problem, and we provide uniform generalization bounds governed by the spectral structure of the frozen features.