NeurIPS 2026 Papers

Contemporary large language model (LLM) chat systems treat conversation history as an immutable sequence of turns that defines the model’s working context. However, user intent in real interactions is not static: it evolves through correction, refinement, and shifting constraints. This mismatch between dynamic intent and static transcripts can result in context pollution, where outdated or irrelevant information persists and continues to influence subsequent responses. We introduce mutable transcripts, a new interaction paradigm that enables users to revise prior turns through natural language edit requests, allowing the conversation history itself to be updated rather than appended. This reframes the transcript from a passive record into an editable representation of conversational state. We present a working prototype that integrates transcript-level revision into a standard chat interface and evaluate its feasibility through a controlled user study (n=17) and an illustrative transcript analysis of representative interaction scenarios. Participants significantly preferred mutable transcripts over standard chat across measures of clarity, confidence, and ease of use, with reduced intent to restart conversations. Transcript analysis of representative user study conversations shows that mutable transcripts can reduce conversation length and eliminate obsolete retained context. These findings provide initial evidence that user-driven revision of conversational history can improve interaction quality and help maintain a more current representation of user intent.

Multivariate time series forecasting presents unique challenges because future variables often co-evolve under shared system dynamics. While existing studies mainly focus on cross-variable dependencies in historical observations, dependencies among future values are much less explored. Specifically, modern forecasting models largely follow the Direct Forecasting (DF) paradigm, generating multi-step forecasts with point-wise objectives that do not explicitly constrain cross-variable structure. In this work, we show that the DF objective is mismatched in the presence of cross-variable and lagged dependencies, revealing an objective gap. To address this issue, we propose **C**ross-**V**ariable **Loss** (CvLoss), a plug-in structural regularizer that constrains forecast residuals on a cross-variable graph. CvLoss penalizes inconsistent edge-wise residual differences over forecast patches, encouraging consistency across both synchronous and asynchronous interactions. Our experiments show that CvLoss consistently improves competitive forecasting models, outperforms representative learning objectives, and is compatible with a variety of forecasting backbones. Code is available at: https://anonymous.4open.science/r/CvLoss.

Due to the challenge of uncovering biologically meaningful spatial tissue patterns from standard H\&E-stained whole-slide images (WSIs) without relying on multiplexed staining or fixed-scale modeling, a unified framework is proposed for spatial biomarker discovery that integrates cell classification, adaptive multi-scale microenvironment construction, and colocalization pattern analysis. Cell-level predictions assign discrete, semantically meaningful types that serve as the basis for microenvironment modeling, where physical adjacency initializes local ecological features and a combination of spatial competition-based seed selection with homogeneity-constrained recursive region growing allows microenvironments to expand from locally pure regions in a data-driven manner, capturing heterogeneity across scales without predefined spatial ranges. Microenvironment-level cell composition features are then used to construct a colocalization feature space, explicitly modeling spatial relationships among cell types, and differential biomarkers are identified via a distributional comparison framework that systematically detects group-specific colocalization patterns. Experiments demonstrate that this framework reliably discovers discriminative and biologically interpretable spatial biomarkers from H\&E images alone, exhibiting robustness in tissues with complex structures and heterogeneous scales, and offering a flexible, extensible paradigm for spatial phenotypic biomarker discovery in standard histopathology.

Harmful fine-tuning can rapidly compromise safety alignment in Large Language Models, underscoring the vulnerability of deployed models to post-release modification. While prior work motivates model immunization by inducing ill-conditioned curvature to slow harmful optimization, this approach is not scalable due to the high computational cost of Hessian estimation. We refine this paradigm by extending the theoretical framework to prioritize plateau regions alongside curvature. We further introduce an immunization algorithm based on efficient Hessian estimation and bounds on the extremal singular values of the Hessian. We demonstrate the proposed immunization framework in three applications: (i) Fine-tuning-as-a-service settings where benign user data may be mixed with poisoned harmful data, (ii) open-weight settings where an adversary can fine-tune on harmful-only data with varied training dynamics, and (iii) robust unlearning, where immunized models better resist relearning attacks after unlearning. Extensive experiments confirm that our approach achieves robust defense against adversarial fine-tuning while maintaining general utility and downstream trainability.

Predicting the olfactory perception of scent mixtures remains a fundamental challenge in computational neuroscience. Existing methods derive mixture-level features by encoding individual components with a single-molecule feature encoder pretrained on annotated olfactory data and aggregating these features via mean pooling, concatenation, or self-attention. This paradigm faces two critical limitations: the scarcity of annotated olfactory data and the inability to capture complex interactions among mixture components. We present MixScentNet, a multiscale graph neural network framework, to address these challenges with two key innovations. First, we introduce a self-supervised pretraining strategy that uses molecular graph structures with RDKit features to predict corresponding Mordred descriptors, enabling the model to learn rich physicochemical knowledge without relying on scarce annotated olfactory data. Second, we propose a novel mixture-as-graph paradigm to model the constituent molecules as nodes in the mixture-level graph, aligning with the fact that humans holistically perceive scent mixtures. We further process the graph via a graph attention network V2 (GATv2) to capture the high-order molecular interactions. MixScentNet differs from existing methods in terms of its pretraining strategy and ability to capture mixture-level features, achieving state-of-the-art performance on both mixture-level olfactory label prediction and perceptual distance estimation tasks. We also find that MixScentNet can reproduce olfactory white phenomena, indicating that the model enjoys clear interpretability grounded in psychophysics. The demo code is available in the Anonymous link \url{https://anonymous.4open.science/r/Odor-D2DB}

Test-time alignment aims to adapt large language models (LLMs) to runtime objectives without updating their parameters. This is especially useful when the reward signal is user-specific, time-varying, or available only through a black-box evaluator. We propose a minimally invasive pre-logit steering method that optimizes additive interventions to the hidden states of a frozen LLM in order to improve reward while preserving the base model's behavior. Rather than penalizing the Euclidean norm of the steering vector, we derive an effort penalty from the local KL geometry of the induced token distribution. Specifically, we show that the per-token KL divergence between the steered and reference policies admits a second-order expansion given by a Fisher-quadratic form in the steering vector. This regularizer has an analytic gradient computable through matrix-vector products with the frozen language-model head, making it as inexpensive as a standard quadratic penalty while retaining a principled KL interpretation. We further decompose the sequence-level KL gradient into an analytic Fisher component and a trajectory-dependent score-function component, and introduce a hierarchy of Fisher surrogates with provable first-order equivalence in the small-steering regime. The resulting algorithm provides a training-free, reward-driven test-time alignment procedure that improves reward while controlling deviation from the base policy.

Memory-augmented LLM agents tackle complex long-horizon tasks by recursively summarizing interaction trajectories into compact memory. However, existing approaches typically train these memory policies using outcome-based reinforcement learning, failing to localize where intermediate memory quality degrades. As interactions unfold, ambiguous recursive summaries progressively discard or blur task-relevant information. This exacerbates belief deviation, obscuring the agent’s estimate of the latent task state and ultimately derailing long-horizon reasoning. We therefore argue that memory optimization should focus not merely on trajectory-level success, but on the clarity of the belief induced by intermediate summaries. To this end, we introduce Belief Entropy, a self-supervised proxy that probes how uncertain the model remains about the latent task state given its current memory. Based on this proxy, we propose Metacognitive Memory Policy Optimization (MMPO). Instead of relying only on sparse outcome-based signals, MMPO provides fine-grained, memory-specific supervision by explicitly penalizing summaries that induce high epistemic uncertainty. Experiments show that MMPO consistently outperforms existing methods on diverse long-horizon tasks, maintaining 97.1% performance even when scaled to 1.75M-token contexts.

Associative memory models classically describe retrieval as convergence to stored patterns, but the theory of sequential retrieval remains comparatively limited. We develop a two-timescale input-driven Hopfield network in which fast associative retrieval is modulated by a slow feedback variable evolving over a prescribed memory-transition graph. This yields autonomous, graph-constrained transitions while preserving analytical tractability. Using slow-fast and geometric singular perturbation theory, we characterize memory fixation via memory manifolds and identify the geometric regions that govern stability loss and escape from retrieved memories. We further derive reduced slow dynamics and explicit criteria for transition onset, self-sustained retrieval, and collapse, expressed through escape times, gain thresholds, and fixed points. Together, these results provide a tractable framework for analyzing feedback-driven sequential retrieval in Hopfield networks.

Recent progress in task-optimized neural networks has established encoding models as a powerful tool for predicting brain responses to naturalistic stimuli, yet most existing approaches rely on unimodal representations. The emergence of omni-modal foundation models and rich multimodal neural datasets enables shared encoding models that jointly integrate visual, auditory, and linguistic information across subjects. We introduce MIRAGE, a brain encoding framework that predicts whole-brain fMRI responses to naturalistic audiovisual stimuli with paired transcripts. MIRAGE extracts representations from a single pretrained omni-modal backbone through three modality-specific cross-attention modules whose latent queries adaptively aggregate features across the backbone's 48 layers, and combines them through a transformer-based brain encoder and a subject-specific linear head over the cortical parcels. On the Algonauts benchmark, MIRAGE achieves state-of-the-art results on the out-of-distribution dataset. Controlled comparisons show that native multimodal fusion (features taken from a single jointly trained model) consistently outperforms post-hoc fusion of independently extracted unimodal streams, across architectural levels and backbones. Beyond predictive accuracy, the learned attention weights are directly inspectable: each modality's gating module discovers a distinct depth profile over the backbone, and each modality traces a distinct, anatomically structured pattern across cortex. Together, these results propose adaptive layer-wise aggregation of natively multimodal features as a more generalizable, interpretable, and accurate approach for whole-brain encoding.

Conservative machine learning interatomic potentials (MLIPs) predict potential energy and obtain atomic forces as the negative gradient of energy with respect to atomic positions. Computing forces requires a backward pass from energy to atomic positions, and backpropagating the force loss differentiates through this pass, creating double-backward execution. This execution creates two activation classes with different lifetimes: forward activations and first-backward activations. Heterogeneous atomic graph batches make their memory footprint input dependent, while standard checkpointing mainly targets forward activations and does not directly manage first-backward activations. Therefore, we present MICA, an activation checkpointing planner for double-backward execution in conservative MLIP training. MICA combines a checkpoint primitive for double-backward phases with an input-aware dynamic programming planner that selects checkpoint modes for dynamic atomic graph batches under a specified memory budget. Across EqV3, MatRIS, PET, and UMA, MICA's full-checkpoint mode reduces peak memory by 49\% to 86\%, compared with only 10\% to 24\% from standard PyTorch checkpointing. On an 80GB GPU, training EquiformerV3 without checkpointing peaks at 75.47GB and exceeds 50GB in 76.7\% of 1000 iterations, while MICA keeps all 1000 iterations within a 50GB memory budget with 16.25\% step time overhead. Under the same GPU memory capacity, MICA increases the maximum feasible atoms per batch by $4.2\times$ to $9.2\times$. In distributed training, the reduced peak memory allows MICA to reduce reliance on heavier memory saving techniques, such as Fully Sharded Data Parallel (FSDP) and graph parallelism (GP). On a single H100 node with NVLink, MICA trains a MatRIS-MoE model with 2B parameters, achieving a $1.88\times$ to $2.03\times$ speedup over FSDP and GP baselines.

Large language model agents are increasingly deployed through the Model Context Protocol (MCP), which allows them to discover and invoke tools across multiple servers at runtime. Existing MCP benchmarks primarily measure task completion capability but provide limited insight into why agents fail, reporting aggregate success rates without attributing failures to specific reasoning, execution, or memory breakdowns. We introduce MCPHallu, a benchmark for diagnosing hallucination-prone agent failures in MCP environments along three axes: reasoning, execution, and memory. MCPHallu contains 358 tasks across five capability domains and runs on 36 production MCP servers in isolated containers. Each task is designed with a structural trigger that makes one of four subtypes the dominant failure risk: Branch Collapse, Unreachable Goal, Tool Misuse, or Context Forgetting. This design enables subtype-level analysis rather than aggregate task success alone. We evaluate fourteen frontier LLM agents from eight vendors on 5,012 trajectories. Current MCP agents remain vulnerable to hallucination-inducing tool environments, achieving an average score of 0.642. Failures concentrate on Unreachable Goal tasks, where the average score drops to 0.458 and 22.1\% of runs pass, indicating that agents often continue issuing tool calls or claim successful completion rather than recognizing infeasible goals. Performance across the four subtypes is largely independent: per-model ranks span on average 6.9 of 13 positions, and Branch Collapse is uncorrelated with Unreachable Goal across models ($r = -0.12$). MCPHallu is released at https://anonymous.4open.science/r/mcp_hallu.

While AI weather models excel at short-to-medium range forecasts (up to 15 days), they frequently suffer from ill-defined "instabilities" when rolled out over longer horizons. This work addresses the lack of a formal taxonomy by categorizing these failures into three distinct regimes — blow-up, drift, and loss of seasonality — through year-long rollouts of nine state-of-the-art AI weather models. Our analysis reveals that stability hinges on the treatment of small spatio-temporal scales: unstable models amplify high-frequency energy, while stable models act as denoisers when noise is added to their inputs. Far from reducing these models to mere stochastic parrots, our findings highlight that stable models generate unique weather trajectories, conditioned on the initial state. We verify our findings through ablation studies on architectural design choices, conducted using state-of-the-art Vision Transformer (ViT) AI weather model architectures.

A kernel $k(x,y)$ is *LSHable* if there exists a locality sensitive hashing scheme $\mathcal H$ such that $k(x,y)=\Pr_{h\sim\mathcal H}[h(x)=h(y)]$ for all $x,y$. This notion plays a key role in efficient kernel methods in high dimensions. In this work, we show that the $p$-exponential kernel $k(x,y)=\exp(-\lVert x-y \rVert_p)$ is LSHable in bounded regions for all $1<p\leq2$. Previously, this was known only for $p=1$. Our new mosaic LSH scheme is based on a Poisson hyperplane process with hyperplanes sampled as $\ell_1$-biased $p$-stable vectors, for which we develop efficient sampling procedures. As applications, our results yield new and efficient density estimation methods based on LSHability for those $p$-exponential kernels.

One strategy to scale up ML-driven science is to increase wet lab experiments' information density. We present a method based on a neural extension of compressed sensing to function space. We measure the activity of multiple different molecules simultaneously, rather than individually. Then, we deconvolute the molecule-activity map during model training. Co-design of wet lab experiments and learning algorithms provably leads to orders-of-magnitude gains in information density. We demonstrate on antibodies and cell therapies.

We introduce LibriBrain100, a large-scale MEG dataset for speech decoding designed from the ground up for reproducible, standardised evaluation. LibriBrain100 more than doubles the size of the original LibriBrain release, resulting in over 100 hours of high-quality MEG acquired while subjects listened to naturalistic continuous speech. With ~80 hours from a single subject, LibriBrain100 sets a new record for ‘deep’, within-subject neural data (8x more than the next comparable dataset and roughly 80x more than other datasets). To demonstrate the payoff of this depth-first design, we evaluate on a word-classification benchmark—an increasingly well-established stepping stone towards the open challenge of non-invasive brain-to-text decoding. Using an existing decoding model, we achieve state-of-the-art performance—validating both the quality of the recordings and the value of within-subject data at scale. Because collecting 80 hours of data per user is impractical for real-world applications, we also collected ~40 minutes of additional data from each of 32 subjects. Using the same word-classification benchmark, we demonstrate the value of ’broad’ multi-subject data: supervised fine-tuning of a pre-trained model can substantially compensate for limited per-subject data. We provide standard train, validation, and test splits, all reproducible through an open-sourced Python library that supports easy downloading, optional preprocessing, and data loading for common deep learning frameworks. In addition, the dataset and evaluation infrastructure are being released alongside an open machine-learning competition with a public leaderboard for standardised benchmarking. Ultimately, our hope is that LibriBrain100 will accelerate progress towards practical non-invasive brain-computer interfaces, capable of restoring communication to people living with severe paralysis.

Recurrent neural networks are widely used in neuroscience and machine learning, but their weight matrices often appear dense and unstructured, defying clean interpretation. The Schur decomposition expresses any recurrent weight matrix in quasi-triangular form, allowing network dynamics to be interpreted as a latent circuit of "functionally feedforward" interactions between orthogonal modes. However, its application for interpreting recurrent weight matrices has been limited due to its well-known combinatorial non-uniqueness. We introduce SchurMO (Schur Manifold Optimization), a method for discovering structured Schur forms via Riemannian gradient descent on the manifold of orthogonal similarity transformations, enabling recovery of target motifs. Compared to a greedy baseline, SchurMO more reliably and more efficiently recovers ground-truth latent motifs (chain-like, banded, and modular structure) from noisy weight matrices. Applying SchurMO to nonlinear recurrent networks enables recovery of latent chain-like structure, providing a link between connectivity and the underlying dynamics of sequence production. Our results highlight SchurMO as a scalable, computationally efficient approach to discovering latent structure in the Schur forms of recurrent weights.

Understanding mouse behavior at scale is a foundational problem in neuroscience, with two core tasks: classifying frames into known behaviors and segmenting recordings into recurring patterns without labels. Pose-based methods are the standard approach but face fundamental limitations: pose coordinates capture only skeletal information, discard environmental context, and degrade under frequent occlusions in social scenarios. Mouse videos can provide the visual context that pose lacks, yet not all frames benefit equally, and adding visual features uniformly to every frame is suboptimal. We find that a pose-based motion encoder, trained with auxiliary supervision from videos, can identify where it struggles: with a discrete representation, the encoder learns a codebook of recurring patterns, and frames whose motion does not fit this codebook produce large quantization residuals. We show that these high-residual frames are where visual context tends to help most: residual-guided frame selection consistently outperforms uniform selection across budgets, and often exceeds dense visual processing. We therefore propose ResiFuse, a framework that pretrains a motion encoder on pose with an auxiliary cross-modal loss and integrates vision context only on high-residual frames at fine-tuning time. We evaluate ResiFuse on three mouse behavior datasets and show improvements over pose- and video-based baselines on most tasks, with the largest gains on the most socially complex dataset.

Data collected from the physical world is always a combination of multiple sources: an underlying signal from the physical process of interest and a signal from measurement-dependent artifacts from the sensor or instrument. This secondary signal acts as a confounding factor, limiting our ability to extract information about the underlying physics. Moreover, it poses significant challenges for combining data in heterogeneous or multi-instrument frameworks. To disentangle these factors of variation, we propose a dual-encoder architecture with a counterfactual generation objective that leverages overlapping observations. The resulting representations explicitly separate intrinsic signals from sensor-specific distortions and noise, and can be used for counterfactual view generation, parameter inference, and instrument-independent similarity search---all unconfounded by measurement artifacts. We demonstrate the effectiveness of our approach in a multi-instrument setting on astrophysical galaxy images from the DESI Legacy Imaging Survey (Legacy) and the Hyper Suprime-Cam (HSC) Survey. This framework provides a general recipe for scientific self-supervised pretraining: construct training pairs from overlapping observations of the same physical system, treat sensor- or modality-specific effects as augmentations, and learn invariant representations through counterfactual generation.

Intracortical brain–computer interfaces (iBCIs) can restore movement and communication, but neural drift can degrade performance across recording sessions. Because recalibration adds user burden, there is a need for decoders that adapt from minimal calibration data. Although session-specific input layers are a common way to handle cross-session drift, they can underperform in the extreme few-shot regime because the shared decoder is trained downstream of well-estimated session-specific layers, but must rely at deployment on a new-session layer fit from only a few calibration trials. To avoid this dependence, we propose a two-stage approach: first, train a single GRU decoder on pooled held-in sessions without session-specific input layers, encouraging the backbone to handle cross-session variability directly; second, freeze this backbone and meta-learn a lightweight alignment layer for rapid calibration. We evaluate on FALCON H2, an official human handwriting iBCI benchmark designed for few-shot cross-session decoding, and achieve 7.41\% word error rate using only three released calibration sentences per held-out session. These results suggest that, under extreme calibration limits, learning a transferable backbone before adding session-specific adaptation can be more effective than jointly training flexible session-specific layers.