NeurIPS 2024 论文列表

Large Vision-Language Models (LVLMs) typically encode an image into a fixed number of visual tokens (e.g., 576) and process these tokens with a language model. Despite their strong performance, LVLMs face challenges in adapting to varying computational constraints. This raises the question: can we achieve flexibility in the number of visual tokens to suit different tasks and computational resources? We answer this with an emphatic yes. Inspired by Matryoshka Representation Learning, we introduce the Matryoshka Query Transformer (MQT), capable of encoding an image into $m$ visual tokens during inference, where $m$ can be any number up to a predefined maximum. This is achieved by employing a query transformer with $M$ latent query tokens to compress the visual embeddings. During each training step, we randomly select $m \leq M$ latent query tokens and train the model using only these first $m$ tokens, discarding the rest. Combining MQT with LLaVA, we train a single model once, and flexibly and drastically reduce the number of inference-time visual tokens while maintaining similar or better performance compared to training independent models for each number of tokens. Our model, MQT-LLaVA, matches LLaVA-1.5 performance across 11 benchmarks using a maximum of 256 tokens instead of LLaVA’s fixed 576. Reducing to 16 tokens (8x less TFLOPs) only sacrifices the performance by 2.4 points on MMBench. On certain tasks such as ScienceQA and MMMU, we can even go down to only 2 visual tokens with performance drops of just 3\% and 6\% each. Our exploration of the trade-off between the accuracy and computational cost brought about by the number of visual tokens facilitates future research to achieve the best of both worlds.

With the proposal of patching technique in time series forecasting, Transformerbased models have achieved compelling performance and gained great interest from the time series community. But at the same time, we observe a new problem that the recent Transformer-based models are overly reliant on patching to achieve ideal performance, which limits their applicability to some forecasting tasks unsuitable for patching. In this paper, we intent to handle this emerging issue. Through diving into the relationship between patching and full attention (the core mechanism in Transformer-based models), we further find out the reason behind this issue is that full attention relies overly on the guidance of patching to focus on the important time points and learn non-trivial temporal representation. Based on this finding, we propose DeformableTST as an effective solution to this emerging issue. Specifically, we propose deformable attention, a sparse attention mechanism that can better focus on the important time points by itself, to get rid of the need of patching. And we also adopt a hierarchical structure to alleviate the efficiency issue caused by the removal of patching. Experimentally, our DeformableTST achieves the consistent state-of-the-art performance in a broader range of time series tasks, especially achieving promising performance in forecasting tasks unsuitable for patching, therefore successfully reducing the reliance on patching and broadening the applicability of Transformer-based models. Code is available at this repository: https://github.com/luodhhh/DeformableTST.

Generalization remains a central challenge in machine learning. In this work, we propose *Learning from Teaching* (**LoT**), a novel regularization technique for deep neural networks to enhance generalization. Inspired by the human ability to capture concise and abstract patterns, we hypothesize that generalizable correlations are expected to be easier to imitate. LoT operationalizes this concept to improve the generalization of the main model with auxiliary student learners. The student learners are trained by the main model and, in turn, provide feedback to help the main model capture more generalizable and imitable correlations. Our experimental results across several domains, including Computer Vision, Natural Language Processing, and methodologies like Reinforcement Learning, demonstrate that the introduction of LoT brings significant benefits compared to training models on the original dataset. The results suggest the effectiveness and efficiency of LoT in identifying generalizable information at the right scales while discarding spurious data correlations, thus making LoT a valuable addition to current machine learning. Code is available at https://github.com/jincan333/LoT.

We introduce MVSplat360, a feed-forward approach for 360° novel view synthesis (NVS) of diverse real-world scenes, using only sparse observations. This setting is inherently ill-posed due to minimal overlap among input views and insufficient visual information provided, making it challenging for conventional methods to achieve high-quality results. Our MVSplat360 addresses this by effectively combining geometry-aware 3D reconstruction with temporally consistent video generation. Specifically, it refactors a feed-forward 3D Gaussian Splatting (3DGS) model to render features directly into the latent space of a pre-trained Stable Video Diffusion (SVD) model, where these features then act as pose and visual cues to guide the denoising process and produce photorealistic 3D-consistent views. Our model is end-to-end trainable and supports rendering arbitrary views with as few as 5 sparse input views. To evaluate MVSplat360's performance, we introduce a new benchmark using the challenging DL3DV-10K dataset, where MVSplat360 achieves superior visual quality compared to state-of-the-art methods on wide-sweeping or even 360° NVS tasks. Experiments on the existing benchmark RealEstate10K also confirm the effectiveness of our model. Readers are highly recommended to view the video results at [donydchen.github.io/mvsplat360](https://donydchen.github.io/mvsplat360).

Large language models (LLMs) are increasingly essential in processing natural languages, yet their application is frequently compromised by biases and inaccuracies originating in their training data. In this study, we introduce \textbf{Cross-Care}, the first benchmark framework dedicated to assessing biases and real world knowledge in LLMs, specifically focusing on the representation of disease prevalence across diverse demographic groups. We systematically evaluate how demographic biases embedded in pre-training corpora like $ThePile$ influence the outputs of LLMs. We expose and quantify discrepancies by juxtaposing these biases against actual disease prevalences in various U.S. demographic groups. Our results highlight substantial misalignment between LLM representation of disease prevalence and real disease prevalence rates across demographic subgroups, indicating a pronounced risk of bias propagation and a lack of real-world grounding for medical applications of LLMs. Furthermore, we observe that various alignment methods minimally resolve inconsistencies in the models' representation of disease prevalence across different languages. For further exploration and analysis, we make all data and a data visualization tool available at: \url{www.crosscare.net}.

To train a deblurring network, an appropriate dataset with paired blurry and sharp images is essential. Existing datasets collect blurry images either synthetically by aggregating consecutive sharp frames or using sophisticated camera systems to capture real blur. However, these methods offer limited diversity in blur types (blur trajectories) or require extensive human effort to reconstruct large-scale datasets, failing to fully reflect real-world blur scenarios. To address this, we propose GS-Blur, a dataset of synthesized realistic blurry images created using a novel approach. To this end, we first reconstruct 3D scenes from multi-view images using 3D Gaussian Splatting~(3DGS), then render blurry images by moving the camera view along the randomly generated motion trajectories. By adopting various camera trajectories in reconstructing our GS-Blur, our dataset contains realistic and diverse types of blur, offering a large-scale dataset that generalizes well to real-world blur. Using GS-Blur with various deblurring methods, we demonstrate its ability to generalize effectively compared to previous synthetic or real blur datasets, showing significant improvements in deblurring performance. We will publicly release our dataset.

While humans effortlessly discern intrinsic dynamics and adapt to new scenarios, modern AI systems often struggle. Current methods for visual grounding of dynamics either use pure neural-network-based simulators (black box), which may violate physical laws, or traditional physical simulators (white box), which rely on expert-defined equations that may not fully capture actual dynamics. We propose the Neural Material Adaptor (NeuMA), which integrates existing physical laws with learned corrections, facilitating accurate learning of actual dynamics while maintaining the generalizability and interpretability of physical priors. Additionally, we propose Particle-GS, a particle-driven 3D Gaussian Splatting variant that bridges simulation and observed images, allowing back-propagate image gradients to optimize the simulator. Comprehensive experiments on various dynamics in terms of grounded particle accuracy, dynamic rendering quality, and generalization ability demonstrate that NeuMA can accurately capture intrinsic dynamics. Project Page: https://xjay18.github.io/projects/neuma.html.

Although modern imaging technologies allow us to study connectivity between two distinct brain regions $\textit{in-vivo}$, an in-depth understanding of how anatomical structure supports brain function and how spontaneous functional fluctuations emerge remarkable cognition is still elusive. Meanwhile, tremendous efforts have been made in the realm of machine learning to establish the nonlinear mapping between neuroimaging data and phenotypic traits. However, the absence of neuroscience insight in the current approaches poses significant challenges in understanding cognitive behavior from transient neural activities. To address this challenge, we put the spotlight on the coupling mechanism of structural connectivity (SC) and functional connectivity (FC) by formulating such network neuroscience question into an expressive graph representation learning problem for high-order topology. Specifically, we introduce the concept of $\textit{topological detour}$ to characterize how a ubiquitous instance of FC (direct link) is supported by neural pathways (detour) physically wired by SC, which forms a cyclic loop interacted by brain structure and function. In the clich\'e of machine learning, the multi-hop detour pathway underlying SC-FC coupling allows us to devise a novel multi-head self-attention mechanism within Transformer to capture multi-modal feature representation from paired graphs of SC and FC. Taken together, we propose a biological-inspired deep model, coined as $\textit{NeuroPath}$, to find putative connectomic feature representations from the unprecedented amount of neuroimages, which can be plugged into various downstream applications such as task recognition and disease diagnosis. We have evaluated $\textit{NeuroPath}$ on large-scale public datasets including Human Connectome Project (HCP) and UK Biobank (UKB) under different experiment settings of supervised and zero-shot learning, where the state-of-the-art performance by our $\textit{NeuroPath}$ indicates great potential in network neuroscience.

Intent learning, which aims to learn users' intents for user understanding and item recommendation, has become a hot research spot in recent years. However, existing methods suffer from complex and cumbersome alternating optimization, limiting performance and scalability. To this end, we propose a novel intent learning method termed \underline{ELCRec}, by unifying behavior representation learning into an \underline{E}nd-to-end \underline{L}earnable \underline{C}lustering framework, for effective and efficient \underline{Rec}ommendation. Concretely, we encode user behavior sequences and initialize the cluster centers (latent intents) as learnable neurons. Then, we design a novel learnable clustering module to separate different cluster centers, thus decoupling users' complex intents. Meanwhile, it guides the network to learn intents from behaviors by forcing behavior embeddings close to cluster centers. This allows simultaneous optimization of recommendation and clustering via mini-batch data. Moreover, we propose intent-assisted contrastive learning by using cluster centers as self-supervision signals, further enhancing mutual promotion. Both experimental results and theoretical analyses demonstrate the superiority of ELCRec from six perspectives. Compared to the runner-up, ELCRec improves NDCG@5 by 8.9\% and reduces computational costs by 22.5\% on the Beauty dataset. Furthermore, due to the scalability and universal applicability, we deploy this method on the industrial recommendation system with 130 million page views and achieve promising results. The codes are available on GitHub\footnote{https://github.com/yueliu1999/ELCRec}. A collection (papers, codes, datasets) of deep group recommendation/intent learning methods is available on GitHub\footnote{https://github.com/yueliu1999/Awesome-Deep-Group-Recommendation}.

Recent 3D large reconstruction models (LRMs) can generate high-quality 3D content in sub-seconds by integrating multi-view diffusion models with scalable multi-view reconstructors. Current works further leverage 3D Gaussian Splatting as 3D representation for improved visual quality and rendering efficiency. However, we observe that existing Gaussian reconstruction models often suffer from multi-view inconsistency and blurred textures. We attribute this to the compromise of multi-view information propagation in favor of adopting powerful yet computationally intensive architectures (\eg, Transformers). To address this issue, we introduce MVGamba, a general and lightweight Gaussian reconstruction model featuring a multi-view Gaussian reconstructor based on the RNN-like State Space Model (SSM). Our Gaussian reconstructor propagates causal context containing multi-view information for cross-view self-refinement while generating a long sequence of Gaussians for fine-detail modeling with linear complexity. With off-the-shelf multi-view diffusion models integrated, MVGamba unifies 3D generation tasks from a single image, sparse images, or text prompts. Extensive experiments demonstrate that MVGamba outperforms state-of-the-art baselines in all 3D content generation scenarios with approximately only $0.1\times$ of the model size. The codes are available at \url{https://github.com/SkyworkAI/MVGamba}.

Estimating the density of a distribution from samples is a fundamental problem in statistics. In many practical settings, the Wasserstein distance is an appropriate error metric for density estimation. For example, when estimating population densities in a geographic region, a small Wasserstein distance means that the estimate is able to capture roughly where the population mass is. In this work we study differentially private density estimation in the Wasserstein distance. We design and analyze instance-optimal algorithms for this problem that can adapt to easy instances. For distributions $P$ over $\mathbb{R}$, we consider a strong notion of instance-optimality: an algorithm that uniformly achieves the instance-optimal estimation rate is competitive with an algorithm that is told that the distribution is either $P$ or $Q_P$ for some distribution $Q_P$ whose probability density function (pdf) is within a factor of 2 of the pdf of $P$. For distributions over $\mathbb{R}^2$, we use a slightly different notion of instance optimality. We say that an algorithm is instance-optimal if it is competitive with an algorithm that is given a constant multiplicative approximation of the density of the distribution. We characterize the instance-optimal estimation rates in both these settings and show that they are uniformly achievable (up to polylogarithmic factors). Our approach for $\mathbb{R}^2$ extends to arbitrary metric spaces as it goes via hierarchically separated trees. As a special case our results lead to instance-optimal learning in TV distance for discrete distributions.

Anomaly detection methods typically require fully observed data for model training and inference and cannot handle incomplete data, while the missing data problem is pervasive in science and engineering, leading to challenges in many important applications such as abnormal user detection in recommendation systems and novel or anomalous cell detection in bioinformatics, where the missing rates can be higher than 30\% or even 80\%. In this work, first, we construct and evaluate a straightforward strategy, ''impute-then-detect'', via combining state-of-the-art imputation methods with unsupervised anomaly detection methods, where the training data are composed of normal samples only. We observe that such two-stage methods frequently yield imputation bias from normal data, namely, the imputation methods are inclined to make incomplete samples ''normal", where the fundamental reason is that the imputation models learned only on normal data and cannot generalize well to abnormal data in the inference stage. To address this challenge, we propose an end-to-end method that integrates data imputation with anomaly detection into a unified optimization problem. The proposed model learns to generate well-designed pseudo-abnormal samples to mitigate the imputation bias and ensure the discrimination ability of both the imputation and detection processes. Furthermore, we provide theoretical guarantees for the effectiveness of the proposed method, proving that the proposed method can correctly detect anomalies with high probability. Experimental results on datasets with manually constructed missing values and inherent missing values demonstrate that our proposed method effectively mitigates the imputation bias and surpasses the baseline methods significantly. The source code of our method is available at https://github.com/jicongfan/ImAD-Anomaly-Detection-With-Missing-Data.

Diffusion-based generative models have demonstrated their powerful performance across various tasks, but this comes at a cost of the slow sampling speed. To achieve both efficient and high-quality synthesis, various distillation-based accelerated sampling methods have been developed recently. However, they generally require time-consuming fine tuning with elaborate designs to achieve satisfactory performance in a specific number of function evaluation (NFE), making them difficult to employ in practice. To address this issue, we propose **S**imple and **F**ast **D**istillation (SFD) of diffusion models, which simplifies the paradigm used in existing methods and largely shortens their fine-tuning time up to $1000\times$. We begin with a vanilla distillation-based sampling method and boost its performance to state of the art by identifying and addressing several small yet vital factors affecting the synthesis efficiency and quality. Our method can also achieve sampling with variable NFEs using a single distilled model. Extensive experiments demonstrate that SFD strikes a good balance between the sample quality and fine-tuning costs in few-step image generation task. For example, SFD achieves 4.53 FID (NFE=2) on CIFAR-10 with only **0.64 hours** of fine-tuning on a single NVIDIA A100 GPU.

This paper examines two extensions of multi-armed bandit problems: multi-armed bandits with expert advice and contextual linear bandits. For the former problem, multi-armed bandits with expert advice, the previously known best upper and lower bounds have been $O(\sqrt{KT \log \frac{N}{K} })$ and $\Omega( \sqrt{KT \frac{ \log N }{\log K }} )$, respectively. Here, $K$, $N$, and $T$ represent the numbers of arms, experts, and rounds, respectively. We provide a lower bound of $\Omega( \sqrt{KT \log \frac{N}{K}} )$ for the setup in which the player chooses an expert before observing the advices in each round. For the latter problem, contextual linear bandits, we provide an algorithm that achieves $O ( \sqrt{d T \log ( K \min\{ 1, \frac{S}{d} \} )} )$ together with a matching lower bound, where $d$ and $S$ represent the dimensionality of feature vectors and the size of the context space, respectively.

We present AROMA (Attentive Reduced Order Model with Attention), a framework designed to enhance the modeling of partial differential equations (PDEs) using local neural fields. Our flexible encoder-decoder architecture can obtain smooth latent representations of spatial physical fields from a variety of data types, including irregular-grid inputs and point clouds. This versatility eliminates the need for patching and allows efficient processing of diverse geometries. The sequential nature of our latent representation can be interpreted spatially and permits the use of a conditional transformer for modeling the temporal dynamics of PDEs. By employing a diffusion-based formulation, we achieve greater stability and enable longer rollouts compared to conventional MSE training. AROMA's superior performance in simulating 1D and 2D equations underscores the efficacy of our approach in capturing complex dynamical behaviors.

How can diffusion models process 3D geometries in a coarse-to-fine manner, akin to our multiscale view of the world? In this paper, we address the question by focusing on a fundamental biochemical problem of generating 3D molecular conformers conditioned on molecular graphs in a multiscale manner. Our approach consists of two hierarchical stages: i) generation of coarse-grained fragment-level 3D structure from the molecular graph, and ii) generation of fine atomic details from the coarse-grained approximated structure while allowing the latter to be adjusted simultaneously. For the challenging second stage, which demands preserving coarse-grained information while ensuring SE(3) equivariance, we introduce a novel generative model termed Equivariant Blurring Diffusion (EBD), which defines a forward process that moves towards the fragment-level coarse-grained structure by blurring the fine atomic details of conformers, and a reverse process that performs the opposite operation using equivariant networks. We demonstrate the effectiveness of EBD by geometric and chemical comparison to state-of-the-art denoising diffusion models on a benchmark of drug-like molecules. Ablation studies draw insights on the design of EBD by thoroughly analyzing its architecture, which includes the design of the loss function and the data corruption process. Codes are released at https://github.com/Shen-Lab/EBD.

Models trained with self-supervised learning objectives have recently matched or surpassed models trained with traditional supervised object recognition in their ability to predict neural responses of object-selective neurons in the primate visual system. A self-supervised learning objective is arguably a more biologically plausible organizing principle, as the optimization does not require a large number of labeled examples. However, typical self-supervised objectives may result in network representations that are overly invariant to changes in the input. Here, we show that a representation with structured variability to the input transformations is better aligned with known features of visual perception and neural computation. We introduce a novel framework for converting standard invariant SSL losses into "contrastive-equivariant" versions that encourage preserving aspects of the input transformation without supervised access to the transformation parameters. We further demonstrate that our proposed method systematically increases models' ability to predict responses in macaque inferior temporal cortex. Our results demonstrate the promise of incorporating known features of neural computation into task-optimization for building better models of visual cortex.

The study of cells and their responses to genetic or chemical perturbations promises to accelerate the discovery of therapeutics targets. However, designing adequate and insightful models for such data is difficult because the response of a cell to perturbations essentially depends on contextual covariates (e.g., genetic background or type of the cell). There is therefore a need for models that can identify interactions between drugs and contextual covariates. This is crucial for discovering therapeutics targets, as such interactions may reveal drugs that affect certain cell types but not others. We tackle this problem with a novel Factorized Causal Representation (FCR) learning method, an identifiable deep generative model that reveals causal structure in single-cell perturbation data from several cell lines. FCR learns multiple cellular representations that are disentangled, comprised of covariate-specific (Z_x), treatment-specific (Z_t) and interaction-specific (Z_tx) representations. Based on recent advances of non-linear ICA theory, we prove the component-wise identifiability of Z_tx and block-wise identifiability of Z_t and Z_x. Then, we present our implementation of FCR, and empirically demonstrate that FCR outperforms state-of-the-art baselines in various tasks across four single-cell datasets.

Standard reinforcement learning (RL) agents never intelligently explore like a human (i.e. taking into account complex domain priors and adapting quickly based on previous exploration). Across episodes, RL agents struggle to perform even simple exploration strategies, for example systematic search that avoids exploring the same location multiple times. This poor exploration limits performance on challenging domains. Meta-RL is a potential solution, as unlike standard RL, meta-RL can *learn* to explore, and potentially learn highly complex strategies far beyond those of standard RL, strategies such as experimenting in early episodes to learn new skills, or conducting experiments to learn about the current environment. Traditional meta-RL focuses on the problem of learning to optimally balance exploration and exploitation to maximize the *cumulative reward* of the episode sequence (e.g., aiming to maximize the total wins in a tournament -- while also improving as a player). We identify a new challenge with state-of-the-art cumulative-reward meta-RL methods. When optimal behavior requires exploration that sacrifices immediate reward to enable higher subsequent reward, existing state-of-the-art cumulative-reward meta-RL methods become stuck on the local optimum of failing to explore. Our method, First-Explore, overcomes this limitation by learning two policies: one to solely explore, and one to solely exploit. When exploring requires forgoing early-episode reward, First-Explore significantly outperforms existing cumulative meta-RL methods. By identifying and solving the previously unrecognized problem of forgoing reward in early episodes, First-Explore represents a significant step towards developing meta-RL algorithms capable of human-like exploration on a broader range of domains.