NeurIPS 2024 Papers — Page 4
Conference on Neural Information Processing Systems · 4035 papers
An Offline Adaptation Framework for Constrained Multi-Objective Reinforcement Learning
Qian Lin (Sun Yat-sen University), Chao Yu (Pengcheng Laboratory)
OptimizationReinforcement Learning
🎯 What it does: A framework called Preference Distribution Offline Adaptation (PDOA) is proposed, which can infer target multi-objective or safety constraint preferences and generate corresponding policies using only a small amount of demonstration data, without providing explicit preferences or safety thresholds.
ANAH-v2: Scaling Analytical Hallucination Annotation of Large Language Models
Yuzhe Gu (Shanghai Jiao Tong University), Kai Chen (Shanghai AI Laboratory)
TransformerLarge Language ModelSupervised Fine-TuningTextMultimodality
🎯 What it does: This paper proposes an EM-based iterative self-training framework that gradually expands the hallucination labeled dataset and improves annotator accuracy.
Analysing Multi-Task Regression via Random Matrix Theory with Application to Time Series Forecasting
Romain Ilbert (Huawei Noah's Ark Lab), Ievgen Redko (Huawei Noah's Ark Lab)
TransformerTime Series
🎯 What it does: Using random matrix theory to analyze high-dimensional multi-task regression models, we provide closed-form theoretical estimates of training and testing errors, and based on this, design optimal regularization parameters. This framework is then applied to multivariate time series forecasting, enhancing the performance of univariate models.
Analysing the Generalisation and Reliability of Steering Vectors
Daniel Chee Hian Tan (University College London), Robert Kirk (University College London)
Large Language ModelContrastive LearningText
🎯 What it does: Evaluated the reliability and generalization performance of Steering Vectors (SVs) used during inference both in-distribution and out-of-distribution.
Analysis of Corrected Graph Convolutions
Robert Wang (University of Waterloo), Kimon Fountoulakis
ClassificationGraph Neural NetworkGraph
🎯 What it does: The study and theoretical proof demonstrate that the 'corrected convolution' removing the principal eigenvector of the adjacency matrix can alleviate oversmoothing and improve binary and multi-class node classification performance under the context of the stochastic block model.
Analytically deriving Partial Information Decomposition for affine systems of stable and convolution-closed distributions
Chaitanya Goswami (Carnegie Mellon University), Amanda Merkley (Carnegie Mellon University)
Tabular
🎯 What it does: This paper provides a closed-form solution using partial information decomposition (PID) through theoretical derivation, further deriving the corresponding PID upper bound within the family of stable distributions and convolution-closed distributions, and performing numerical validation on the Poisson distribution.
Analyzing & Reducing the Need for Learning Rate Warmup in GPT Training
Atli Kosson (École Polytechnique Fédérale de Lausanne), Martin Jaggi (École Polytechnique Fédérale de Lausanne)
OptimizationTransformerLarge Language ModelText
🎯 What it does: Analyze the necessity of learning rate warm-up in GPT training and reduce or eliminate the need for warm-up by modifying the optimizer.
Animal-Bench: Benchmarking Multimodal Video Models for Animal-centric Video Understanding
Yinuo Jing (Beijing University of Posts and Telecommunications), Jun Guo (Beijing University of Posts and Telecommunications)
Large Language ModelDiffusion modelVideoMultimodalityBenchmark
🎯 What it does: This study constructed the animal-centered multimodal video evaluation benchmark Animal-Bench and evaluated the performance of several large models based on it.
Animate3D: Animating Any 3D Model with Multi-view Video Diffusion
Yanqin Jiang (Chinese Academy of Sciences), Jin Gao (Chinese Academy of Sciences)
GenerationData SynthesisOptimizationDiffusion modelGaussian SplattingVideoMesh
🎯 What it does: A framework named Animate3D is proposed, capable of generating animations for any static 3D model through a multi-view video diffusion model (MV-VDM) and 4D Gaussian splatting optimization (4DGS), and supports direct animation of Mesh.
Annealed Multiple Choice Learning: Overcoming limitations of Winner-takes-all with annealing
David Perera (Telecom Paris), Gaël Richard (Telecom Paris)
OptimizationSupervised Fine-TuningTabularAudio
🎯 What it does: Proposes Annealed Multiple Choice Learning (aMCL), which combines deterministic annealing with MCL training to address hypothesis collapse and local optima issues.
ANT: Adaptive Noise Schedule for Time Series Diffusion Models
Seunghan Lee (Yonsei University), Taeyoung Park (Yonsei University)
Diffusion modelTime Series
🎯 What it does: An adaptive noise scheduling method based on the non-stationarity statistics of time series (ANT) is proposed, which can automatically select appropriate noise schedules for each dataset before training the diffusion model;
Antigen-Specific Antibody Design via Direct Energy-based Preference Optimization
Xiangxin Zhou (University of Chinese Academy of Sciences), Quanquan Gu (ByteDance Research)
OptimizationDrug DiscoveryProtein Structure PredictionDiffusion modelBiomedical Data
🎯 What it does: A direct preference optimization method based on energy, ABDPO, was developed for the sequence-structure co-design of antigen-specific antibodies.
Any2Graph: Deep End-To-End Supervised Graph Prediction With An Optimal Transport Loss
Paul KRZAKALA, Matthieu Labeau (Telecom Paris)
OptimizationGraph Neural NetworkTransformerSupervised Fine-TuningImageTextGraph
🎯 What it does: Proposes the Any2Graph framework, which supports arbitrary inputs (images, text, vectors, etc.) to directly predict graph structures with arbitrary size and unordered nodes through an end-to-end neural network.
Any2Policy: Learning Visuomotor Policy with Any-Modality
Yichen Zhu (Midea Group), Jian Tang (Midea Group)
Robotic IntelligenceTransformerReinforcement LearningVision-Language-Action ModelImageVideoTextMultimodalityPoint Cloud
🎯 What it does: A robot policy framework (Any2Policy) capable of handling arbitrary modalities (text, speech, images, video, point clouds) is proposed, and its generality and performance are validated in real and simulated environments.
AnyFit: Controllable Virtual Try-on for Any Combination of Attire Across Any Scenario
Yuhan Li (Shanghai Jiao Tong University), Bingbing Ni
Image TranslationGenerationDiffusion modelImage
🎯 What it does: A controllable virtual fitting system named AnyFit has been designed, capable of achieving high-resolution and high-fidelity try-on effects for any combination of garments in any scenario.
AP-Adapter: Improving Generalization of Automatic Prompts on Unseen Text-to-Image Diffusion Models
Yuchen Fu (Nanjing University), Qing Gu (Nanjing University)
GenerationDomain AdaptationTransformerLarge Language ModelPrompt EngineeringDiffusion modelImageTextMultimodality
🎯 What it does: A new task is proposed - Automatic Prompt Optimization for Model Generalization (MGAPO), along with a two-stage AP-Adapter method;
Apathetic or Empathetic? Evaluating LLMs' Emotional Alignments with Humans
Jen-tse Huang (Chinese University of Hong Kong), Michael Lyu
TransformerLarge Language ModelPrompt EngineeringTextBenchmark
🎯 What it does: By constructing the EmotionBench framework, experiments were conducted on 428 emotion-inducing scenarios to measure the changes in emotional states of LLMs when faced with negative emotional situations, and these results were compared with the questionnaire responses of 1,266 human subjects.
Applying Guidance in a Limited Interval Improves Sample and Distribution Quality in Diffusion Models
Tuomas Kynkäänniemi (Aalto University), Jaakko Lehtinen (NVIDIA)
GenerationData SynthesisComputational EfficiencyDiffusion modelImageOrdinary Differential Equation
🎯 What it does: This paper proposes to limit the Classifier-Free Guidance (CFG) during the sampling process of diffusion models to a specific range of noise levels, using guidance only in the middle noise range, thereby improving generation quality and speed.
Approaching Human-Level Forecasting with Language Models
Danny Halawi (University of California Berkeley), Jacob Steinhardt (University of California Berkeley)
TransformerLarge Language ModelSupervised Fine-TuningTextTime SeriesRetrieval-Augmented Generation
🎯 What it does: A retrieval-augmented language model system has been constructed to automatically search for relevant information, generate predictions, and aggregate results, achieving automated forecasting of binary events that approaches the level of human forecasters.
Approximated Orthogonal Projection Unit: Stabilizing Regression Network Training Using Natural Gradient
ShaoQi Wang, Siwei Lou (Zhejiang University)
TabularTime Series
🎯 What it does: This paper proposes a novel regression neural network called Approximated Orthogonal Projection Unit (AOPU), aimed at enhancing the training stability and robustness of soft sensors during online deployment.
Approximately Equivariant Neural Processes
Matthew Ashman (University of Cambridge), Richard E. Turner (Microsoft Research AI for Science)
Time Series
🎯 What it does: By incorporating fixed inputs into an existing equivariant neural process (NP) decoder, a model that can adaptively deviate from strict equivariance is constructed.
Approximately Pareto-optimal Solutions for Bi-Objective k-Clustering
Anna Arutyunova (Heinrich Heine University Düsseldorf), Julian Wargalla (Heinrich Heine University Düsseldorf)
OptimizationTabularTime Series
🎯 What it does: This paper studies the bi-objective clustering problem, proposing algorithms to approximate the Pareto front and exploring combinations of multi-objective clustering based on metrics such as k-center, k-diameter, k-median, k-means, and k-separation.
Approximating mutual information of high-dimensional variables using learned representations
Gokul Gowri (Harvard University), Peng Yin (Harvard University)
Representation LearningDrug DiscoveryAuto EncoderTabularBiomedical Data
🎯 What it does: Proposed and implemented the LMI (latent mutual information) approximation method, which uses low-dimensional learning representations and employs a non-parametric MI estimator in this space, addressing the challenge of MI estimation for high-dimensional variables.
Approximating the Top Eigenvector in Random Order Streams
Praneeth Kacham (Google Research), David Woodruff
🎯 What it does: This paper studies a streaming algorithm for approximating the principal eigenvector of the matrix A A^T (i.e., the principal right singular vector of A) under a random-order stream, and provides an algorithm that can output a correlation coefficient with the true principal eigenvector of at least 1 - O(1/√R) in space O(h·d·polylog(d)) (where h is the number of 'heavy rows' in the matrix).
Approximation Rate of the Transformer Architecture for Sequence Modeling
Haotian Jiang (CNRS@CREATE LTD), Qianxiao Li (National University of Singapore)
Recurrent Neural NetworkTransformerTime SeriesSequential
🎯 What it does: This paper constructs a representation theorem and complexity measure for the target space, providing the Jackson-type approximation rate for a single-layer, single-head Transformer model, and proves that low-rank temporal coupling structures can significantly enhance approximation performance.
Approximation-Aware Bayesian Optimization
Natalie Maus (University of Pennsylvania), Jacob R. Gardner (University of Pennsylvania)
OptimizationDrug DiscoveryTabular
🎯 What it does: This paper proposes a joint optimization of the posterior approximation of Sparse Variational Gaussian Processes (SVGP) and the sampling decisions of Bayesian Optimization (BO), forming the Expected Utility Lower Bound (EULBO) to enhance the data acquisition efficiency of high-dimensional large-budget BO.
AR-Pro: Counterfactual Explanations for Anomaly Repair with Formal Properties
Xiayan Ji (University of Pennsylvania), Insup Lee (University of Pennsylvania)
Anomaly DetectionExplainability and InterpretabilityDiffusion modelImageTime Series
🎯 What it does: Proposes the AR-Pro framework, which uses linearly decomposable anomaly detectors to generate counterfactuals for anomalous samples, serving as an explanation for interpretability.
ARC: A Generalist Graph Anomaly Detector with In-Context Learning
Yixin Liu (Griffith University), Shirui Pan (Hong Kong Polytechnic University)
Anomaly DetectionGraph Neural NetworkGraph
🎯 What it does: A 'one-for-all' graph anomaly detection model ARC is proposed, which can instantly detect anomalous nodes in the graph using a small amount of contextual information from normal nodes without retraining or fine-tuning the target dataset.
Architect: Generating Vivid and Interactive 3D Scenes with Hierarchical 2D Inpainting
Yian Wang (University of Massachusetts Amherst), Chuang Gan (University of Massachusetts Amherst)
GenerationData SynthesisDepth EstimationLarge Language ModelDiffusion modelImageText
🎯 What it does: The paper proposes the ARCHITECT framework, which utilizes a 2D image diffusion model for inpainting to generate multi-scale, interactive 3D scenes. It achieves the automated generation of complete indoor layouts from blank rooms through hierarchical iterative inpainting and visual perception.
Are Graph Neural Networks Optimal Approximation Algorithms?
Morris Yau (Massachusetts Institute of Technology), Stefanie Jegelka (Technical University of Munich and Massachusetts Institute of Technology)
OptimizationGraph Neural NetworkGraph
🎯 What it does: This paper designs and implements a graph neural network architecture named OptGNN, which can capture information from optimal approximation algorithms (based on semidefinite programming) and learn high-quality approximate solutions through unsupervised training.
Are High-Degree Representations Really Unnecessary in Equivariant Graph Neural Networks?
Jiacheng Cen (Renmin University of China), Wenbing Huang (Renmin University of China)
Graph Neural NetworkGraph
🎯 What it does: This paper challenges the assumption of the necessity of higher-order transferable vectors in equivariant graph neural networks, proposing the HEGNN model, which enhances expressive power by introducing higher-order transferable vectors while maintaining the efficiency of EGNN.
Are Language Models Actually Useful for Time Series Forecasting?
Mingtian Tan (University of Virginia), Thomas Hartvigsen (University of Virginia)
TransformerLarge Language ModelMultimodalityTime SeriesFinance Related
🎯 What it does: This study investigates the practical value of large language models (LLM) in time series forecasting, using ablation experiments to test whether LLM components truly enhance predictive performance.
Are Large-scale Soft Labels Necessary for Large-scale Dataset Distillation?
Lingao Xiao (Agency for Science Technology and Research), Yang He (Agency for Science Technology and Research)
CompressionKnowledge DistillationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a method called LPLD (Label Pruning for Large-scale Distillation), which first batches similar samples during the image synthesis phase and incorporates class-level BN supervision to enhance the intra-class diversity of the synthesized images. Subsequently, it randomly crops the soft labels and resamples from an improved label pool, achieving soft label compression and performance enhancement.
Are More LLM Calls All You Need? Towards the Scaling Properties of Compound AI Systems
Lingjiao Chen (Stanford University), James Zou (Stanford University)
Large Language ModelTextMultimodalityPhysics Related
🎯 What it does: This study investigates the non-monotonic characteristics of the performance of composite systems that involve multiple calls to language models, followed by voting or filtered voting, as the number of calls varies across different language tasks. A theoretical explanation centered on query difficulty and a predictable scale model are proposed.
Are Multiple Instance Learning Algorithms Learnable for Instances?
Jaeseok Jang (Seoul National University of Science and Technology), HYUK-YOON KWON
ClassificationData SynthesisRecurrent Neural NetworkImageTime Series
🎯 What it does: This paper proposes a theoretical framework based on PAC learning theory to verify whether deep multiple instance learning (Deep MIL) algorithms are learnable at the instance level, and provides necessary and sufficient conditions.
Are nuclear masks all you need for improved out-of-domain generalisation? A closer look at cancer classification in histopathology
Dhananjay Tomar (University of Oslo), Andreas Kleppe (Oslo University Hospital)
ClassificationDomain AdaptationConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: This paper proposes a method that utilizes nuclear segmentation masks during the training phase to guide the model's focus on nuclear morphology and arrangement features, thereby enhancing the out-of-domain generalization ability on data from different hospitals;
Are Self-Attentions Effective for Time Series Forecasting?
Dongbin Kim (Seoul National University), Hoki Kim (Chung-Ang University)
TransformerTime Series
🎯 What it does: This paper proposes a time series Transformer called CATS that uses only cross-attention, eliminating self-attention and focusing on future moment predictions.
Are Uncertainty Quantification Capabilities of Evidential Deep Learning a Mirage?
Maohao Shen (Massachusetts Institute of Technology), Gregory W. Wornell (Massachusetts Institute of Technology)
Anomaly DetectionKnowledge DistillationImage
🎯 What it does: This paper systematically analyzes and reveals the fundamental limitations of existing evidence deep learning (EDL) methods in uncertainty quantification, and proposes an improved scheme based on bootstrap sampling called Bootstrap-Distill.
Are We on the Right Way for Evaluating Large Vision-Language Models?
Lin Chen (University of Science and Technology of China), Feng Zhao (University of Science and Technology of China)
Large Language ModelVision Language ModelImageMultimodalityBenchmark
🎯 What it does: The MMStar benchmark is proposed to specifically address the issues of unnecessary visual content and data leakage in existing multimodal evaluations, constructing 1,500 manually reviewed samples that are visually essential and minimize leakage.
Are Your Models Still Fair? Fairness Attacks on Graph Neural Networks via Node Injections
Zihan Luo (Huazhong University of Science and Technology), Hai Jin (Huazhong University of Science and Technology)
Adversarial AttackGraph Neural NetworkGraph
🎯 What it does: A fairness attack method based on node injection, NIFA, is proposed to reduce the fairness of GNN models.
ArkVale: Efficient Generative LLM Inference with Recallable Key-Value Eviction
Renze Chen (Peking University), Yun Liang (Peking University)
GenerationRetrievalComputational EfficiencyTransformerLarge Language ModelTextRetrieval-Augmented Generation
🎯 What it does: ARKVALE dynamically identifies and recalls important pages by dividing the KV cache into pages, asynchronously backing them up to external memory, and using bounding volume compressed summaries, supporting long-context LLM inference.
AROMA: Preserving Spatial Structure for Latent PDE Modeling with Local Neural Fields
Louis Serrano (Sorbonne Université), Patrick Gallinari (Criteo AI Lab)
TransformerDiffusion modelAuto EncoderTime SeriesPhysics Related
🎯 What it does: The AROMA framework is proposed, which compresses spatial information through fixed-size latent markers on arbitrary geometries and allows querying of prediction results at any location.
ART: Automatic Red-teaming for Text-to-Image Models to Protect Benign Users
Guanlin Li (Nanyang Technological University), Tianwei Zhang (Nanyang Technological University)
GenerationTransformerLarge Language ModelPrompt EngineeringVision Language ModelImageText
🎯 What it does: An automatic red team framework named ART is proposed to discover security risks in text-to-image models under the premise of safe text prompts.
Artemis: Towards Referential Understanding in Complex Videos
Jihao Qiu (University of Chinese Academy of Sciences), Yunjie Tian (University at Buffalo)
Object DetectionObject TrackingGenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodality
🎯 What it does: This paper proposes the Artemis multimodal large language model, specifically designed to locate targets in given frames of videos and generate complete action descriptions.
Articulate your NeRF: Unsupervised articulated object modeling via conditional view synthesis
Jianning Deng (University of Edinburgh), Hakan Bilen (University of Edinburgh)
SegmentationData SynthesisPose EstimationNeural Radiance FieldPoint Cloud
🎯 What it does: An unsupervised perspective-synthesis-based neural radiance field method is proposed for learning part segmentation and pose of articulated objects.
Artificial Generational Intelligence: Cultural Accumulation in Reinforcement Learning
Jonathan Cook (University of Oxford), Jakob Nicolaus Foerster
Meta LearningReinforcement LearningSequential
🎯 What it does: This paper studies two mechanisms for achieving cultural accumulation in reinforcement learning: in-context rapid adaptation and in-weights accumulation, demonstrating that both methods can continuously improve performance across multiple generations of training.
AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation
Anil Kag (Snap Inc), Jian Ren (Snap Inc)
ClassificationSegmentationGenerationData SynthesisConvolutional Neural NetworkTransformerDiffusion modelImageText
🎯 What it does: This paper designs an AsCAN (Asymmetric Convolution-Attention Network) hybrid architecture, which distributes convolution blocks (FusedMBConv) and Transformer blocks (vanilla self-attention) asymmetrically across different stages, enabling simultaneous application to multi-tasks such as image recognition, semantic segmentation, class-conditional generation, and text-to-image (T2I) generation.
Ask, Attend, Attack: An Effective Decision-Based Black-Box Targeted Attack for Image-to-Text Models
Qingyuan Zeng (Xiamen University), Min Jiang (Xiamen University)
GenerationAdversarial AttackRecurrent Neural NetworkTransformerVision Language ModelImageText
🎯 What it does: A decision-based black-box target attack framework called Ask, Attend, Attack (AAA) is proposed, which can implement semantically consistent attacks on image-to-text models with only access to the model's output text.
Assembly Fuzzy Representation on Hypergraph for Open-Set 3D Object Retrieval
Yang Xu (Tsinghua University), Yue Gao (Tsinghua University)
RetrievalGraph Neural NetworkAuto EncoderPoint Cloud
🎯 What it does: For open-set (unseen category) 3D object retrieval, a hypergraph fuzzy representation framework (HAFR) based on part assembly is proposed, achieving high-quality object embedding through equivalent smoothing of part-level features, assembly autoencoding, and fuzzy reconstruction.
Association of Objects May Engender Stereotypes: Mitigating Association-Engendered Stereotypes in Text-to-Image Generation
Junlei Zhou (Southern University of Science and Technology), Xuetao Wei (Southern University of Science and Technology)
GenerationTransformerPrompt EngineeringDiffusion modelImageText
🎯 What it does: This paper studies the stereotypes generated by text-to-image (T2I) models in multi-object association scenarios and proposes a framework called MAS to eliminate these stereotypes caused by associations.
Association Pattern-aware Fusion for Biological Entity Relationship Prediction
Lingxiang Jia (Zhejiang University), Mingli Song (Zhejiang University)
Drug DiscoveryGraph Neural NetworkTransformerBiomedical Data
🎯 What it does: A method for predicting biological entity relationships based on association pattern-aware fusion, called Pattern-BERP, is proposed.
Assouad, Fano, and Le Cam with Interaction: A Unifying Lower Bound Framework and Characterization for Bandit Learnability
Fan Chen (Massachusetts Institute of Technology), Yunbei Xu (National University of Singapore)
Reinforcement Learning
🎯 What it does: A new interactive Fano method is proposed, unifying the three lower bound techniques of Assouad, Fano, and Le Cam under traditional statistical estimation, and combining it with the DEC framework in interactive decision-making; this method introduces the Fractional Covering Number as a new complexity measure, providing lower and upper bounds for problems such as structured multi-armed bandits and contextual bandits, narrowing the logarithmic factor gap between previous DEC lower and upper bounds.
Asymptotics of Alpha-Divergence Variational Inference Algorithms with Exponential Families
François Bertholom (SAMOVAR, Télécom Sud-Paris Institut Polytechnique de Paris), François Roueff (LTCI, Télécom Paris Institut Polytechnique de Paris)
GenerationOptimizationAuto EncoderImage
🎯 What it does: This paper studies the asymptotic properties of variational inference algorithms based on α-divergence and provides theoretical guarantees for their convergence and convergence rates.
AsyncDiff: Parallelizing Diffusion Models by Asynchronous Denoising
Zigeng Chen (National University of Singapore), Xinchao Wang (National University of Singapore)
GenerationData SynthesisDiffusion modelImageVideo
🎯 What it does: A distributed acceleration framework named AsyncDiff is proposed, which achieves asynchronous denoising by splitting the denoising network into several sub-modules and executing them in parallel on multiple GPUs, significantly reducing inference latency.
Asynchronous Perception Machine for Efficient Test Time Training
Rajat Modi (University of Central Florida), Yogesh S Rawat
ClassificationComputational EfficiencyKnowledge DistillationConvolutional Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A new Asynchronous Perceptron Machine (APM) architecture is proposed for fast adaptive training (test-time training) on a single test sample during inference, supporting learning using only the CLS vector extracted from the teacher model at once.
Atlas3D: Physically Constrained Self-Supporting Text-to-3D for Simulation and Fabrication
Yunuo Chen (University of California, Los Angeles), Chenfanfu Jiang (University of California, Los Angeles)
GenerationData SynthesisRobotic IntelligenceDiffusion modelScore-based ModelTextMesh
🎯 What it does: In the process of generating text to 3D, we incorporate differentiable rigid body simulation and stability constraints to generate self-supporting 3D models that can maintain self-support in simulation and can be directly 3D printed.
Attack-Aware Noise Calibration for Differential Privacy
Bogdan Kulynych (Lausanne University Hospital), Carmela Troncoso (École Polytechnique Fédérale de Lausanne)
Safty and PrivacyTransformerLarge Language ModelImageText
🎯 What it does: A method is proposed to directly calibrate the scale of differential privacy noise against attack risks (such as the advantages of membership inference attacks, FPR/FNR), skipping the traditional ε, δ parameter intermediary steps, thereby enhancing the practicality of the model.
Attack-Resilient Image Watermarking Using Stable Diffusion
Lijun Zhang (University of Massachusetts), Hui Guan (University of Massachusetts)
GenerationData SynthesisOptimizationDiffusion modelImage
🎯 What it does: Designed and implemented the ZoDiac watermarking framework, which embeds invisible watermarks in the latent space using a pre-trained Stable Diffusion model, and can still be reliably detected after the images undergo various attacks.
Attention boosted Individualized Regression
Guang Yang (City University of Hong Kong), Long Feng (University of Hong Kong)
TransformerImageBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: A personalized regression model based on the self-attention mechanism is proposed, which learns independent regression coefficients for each sample by utilizing the inter-block relationships within the samples.
Attention Temperature Matters in ViT-Based Cross-Domain Few-Shot Learning
Yixiong Zou (Huazhong University of Science and Technology), Ruixuan Li (Huazhong University of Science and Technology)
ClassificationDomain AdaptationTransformerImage
🎯 What it does: This paper proposes to improve the transfer performance in Cross-Domain Few-Shot Learning (CDFSL) by adjusting the attention temperature (even setting it to 0) in Vision Transformer (ViT), and provides corresponding source domain training and target domain inference strategies.
AttnDreamBooth: Towards Text-Aligned Personalized Text-to-Image Generation
Lianyu Pang (Sun Yat-sen University), Xudong Mao (Hong Kong Metropolitan University)
GenerationData SynthesisTransformerDiffusion modelImageText
🎯 What it does: A three-stage text-to-image personalization method called AttnDreamBooth is proposed, which achieves better text alignment while maintaining identity recognition.
Attractor Memory for Long-Term Time Series Forecasting: A Chaos Perspective
Jiaxi Hu (Hong Kong University of Science and Technology), Yuxuan Liang
OptimizationTime Series
🎯 What it does: A long-term time series forecasting model based on chaos theory, Attraos, is proposed, which captures chaotic attractors using phase space reconstruction, non-parametric embedding, and multi-scale dynamic memory units, and performs local evolution in the frequency domain.
AUC Maximization under Positive Distribution Shift
Atsutoshi Kumagai (NTT), Yasuhiro Fujiwara (NTT)
ClassificationOptimizationImageTabular
🎯 What it does: A method is proposed to maximize AUC under positive distribution shift, utilizing positive samples from the training set, unlabeled samples, and unlabeled samples from the test set for learning.
AUCSeg: AUC-oriented Pixel-level Long-tail Semantic Segmentation
Boyu Han (Institute of Computing Technology, Chinese Academy of Sciences), Qingming Huang (University of Chinese Academy of Sciences)
SegmentationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes AUCSeg, an AUC optimization method for pixel-level long-tail semantic segmentation, combined with a Tail-Class Memory Bank to address large batch memory issues.
Auditing Local Explanations is Hard
Robi Bhattacharjee (University of Tübingen and Tübingen AI Center), Ulrike von Luxburg (University of Tübingen and Tübingen AI Center)
Explainability and InterpretabilityTabular
🎯 What it does: This paper studies how a third-party auditing agency (or user collective) can assess the credibility of explanations through query model decisions and corresponding local explanations in an environment lacking mutual trust.
Auditing Privacy Mechanisms via Label Inference Attacks
Robert Istvan Busa-Fekete, Marika Swanberg (Boston University)
Safty and PrivacyTabular
🎯 What it does: This paper proposes a set of reconstruction advantage metrics (additive and multiplicative) for evaluating label privacy mechanisms, and uses these metrics to audit the security and accuracy of privacy-enhancing techniques such as Random Response (RR) and Label Aggregation (LLP).
Autobidder's Dilemma: Why More Sophisticated Autobidders Lead to Worse Auction Efficiency
Yuan Deng (Google Research), Song Zuo (Google Research)
Optimization
🎯 What it does: This paper studies how the price disorder (PoA) of a system changes when value-maximizing automated bidders use non-uniform bidding ratios in a first-price auction;
Autoformalize Mathematical Statements by Symbolic Equivalence and Semantic Consistency
Zenan Li (Nanjing University), Xiaoxing Ma (Nanjing University)
TransformerLarge Language ModelText
🎯 What it does: This paper designs an automatic formalization framework that scores and filters multiple candidates generated by LLMs using two self-consistency methods: symbolic equivalence and semantic consistency, in order to improve the accuracy of automatic formalization of mathematical expressions.
AutoGuide: Automated Generation and Selection of Context-Aware Guidelines for Large Language Model Agents
Yao Fu (University of Michigan), Honglak Lee (University of Michigan)
GenerationRecommendation SystemOptimizationTransformerLarge Language ModelPrompt EngineeringTextMultimodalityRetrieval-Augmented Generation
🎯 What it does: This study proposes the AUTOGUIDE framework, which automatically generates concise natural language guides that match the context using offline trajectories, and dynamically retrieves the corresponding guides during the testing phase to assist LLM agent decision-making.
AutoManual: Constructing Instruction Manuals by LLM Agents via Interactive Environmental Learning
Minghao Chen (Hangzhou Dianzi University), Xiaofei He (Zhejiang University)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Designed and implemented the AutoManual framework, allowing large language model agents to self-construct environment manuals through planning and rule updates in interactive environments, achieving adaptive learning.
Automated Efficient Estimation using Monte Carlo Efficient Influence Functions
Raj Agrawal (Basis Research Institute, Broad Institute), Eli Bingham (Basis Research Institute, Broad Institute)
OptimizationTabularFinance Related
🎯 What it does: Proposes the Monte Carlo Efficient Influence Function (MC‑EIF) technique for the automated calculation of efficient influence functions under high-dimensional parameter models, and based on this, implements efficient estimators such as first-order correction, double machine learning, and TMLE.
Automated Label Unification for Multi-Dataset Semantic Segmentation with GNNs
Rong Ma (Fudan University), Jian Pu (Fudan University)
Object DetectionSegmentationGraph Neural NetworkLarge Language ModelImage
🎯 What it does: A unified label space for multiple datasets is automatically constructed using Graph Neural Networks (GNN), enabling the semantic segmentation model to train simultaneously on seven different datasets and output a unified label.
Automated Multi-level Preference for MLLMs
Mengxi Zhang (Baidu Inc.), Yifan Sun (Chinese Academy of Science)
Recommendation SystemOptimizationTransformerLarge Language ModelReinforcement LearningVision Language ModelTextMultimodality
🎯 What it does: An Automated Multi-level Preference (AMP) framework has been developed, which significantly reduces hallucinations in multimodal large language models by generating a multi-level preference dataset without human annotations and using the MDPO algorithm.
Automated Multi-Task Learning for Joint Disease Prediction on Electronic Health Records
Suhan Cui (Pennsylvania State University), Prasenjit Mitra (Pennsylvania State University)
OptimizationMeta LearningTabularBiomedical DataElectronic Health Records
🎯 What it does: AutoDP is proposed, an automated multi-task learning framework for jointly predicting multiple diseases on electronic health records (EHR).
Automatic Outlier Rectification via Optimal Transport
Jose Blanchet (Stanford University), Greg Zanotti (Stanford University)
Anomaly DetectionOptimizationTabularTime SeriesFinance Related
🎯 What it does: Proposes a correction set based on optimal transport (concave cost) to achieve an integrated framework for automatic outlier correction and estimation within the probability distribution space.
Automatically Learning Hybrid Digital Twins of Dynamical Systems
Samuel Holt (University of Cambridge), Mihaela van der Schaar (University of Cambridge)
Large Language ModelTime SeriesBiomedical Data
🎯 What it does: Automatically generate and optimize hybrid digital twins (HDTwin) driven by LLM-based evolutionary algorithms, while integrating mechanistic equations and neural networks to achieve precise modeling of dynamic systems.
Automating Data Annotation under Strategic Human Agents: Risks and Potential Solutions
Tian Xie (Ohio State University), Xueru Zhang (Ohio State University)
OptimizationData-Centric LearningTabularFinance Related
🎯 What it does: This paper studies the strategic interaction between agents and the long-term dynamics of models when machine learning models are periodically retrained and use model-labeled and human-labeled samples. It theoretically derives the evolution laws of acceptance rate, qualification rate, and classifier bias; further, it proposes a refined retraining strategy using probabilistic samplers to stabilize the system and analyzes its long-term impact on algorithmic fairness.
AutoMix: Automatically Mixing Language Models
Pranjal Aggarwal (Carnegie Mellon University), Mausam .
Large Language ModelPrompt EngineeringText
🎯 What it does: AutoMix has been developed, a framework for automatically mixing LLMs of different scales, achieving a trade-off between cost and performance through self-validation and POMDP routing.
Autonomous Agents for Collaborative Task under Information Asymmetry
Wei Liu (Tsinghua University), Chen Qian (Tsinghua University)
Large Language ModelAgentic AITextBenchmark
🎯 What it does: Proposed the iAgents framework to address collaborative tasks of multi-agent systems under information asymmetry, and constructed the first benchmark for this scenario - InformativeBench.
Autonomous Driving with Spiking Neural Networks
Rui-Jie Zhu (University of California), Jason Eshraghian
Autonomous DrivingSpiking Neural NetworkPoint Cloud
🎯 What it does: This paper proposes SAD, an end-to-end spiking neural network for perception, prediction, and planning in autonomous driving.
AutoPSV: Automated Process-Supervised Verifier
Jianqiao Lu (University of Hong Kong), Zhijiang Guo (Chinese University of Hong Kong)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The AutoPSV method is proposed, which enhances the reasoning ability of LLMs by training a result supervision validator and automatically generating process labels for each step based on the model's own confidence changes.
Autoregressive Image Diffusion: Generation of Image Sequence and Application in MRI
Guanxiong Luo (University Medical Center Göttingen), Martin Uecker (Graz University of Technology)
RestorationGenerationDiffusion modelScore-based ModelImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A self-regressive image diffusion model (AID) is proposed for generating image sequences and serving as a generative prior in accelerated MRI reconstruction.
Autoregressive Image Generation without Vector Quantization
Tianhong Li (Massachusetts Institute of Technology), Kaiming He (Massachusetts Institute of Technology)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: A new Diffusion Loss is proposed, allowing autoregressive image generation models to directly use continuous value tokens, eliminating the vector quantization step.
Autoregressive Policy Optimization for Constrained Allocation Tasks
David Winkel (Munich Center for Machine Learning), Matthias Schubert (Munich Center for Machine Learning)
OptimizationReinforcement LearningTabularTime SeriesFinance Related
🎯 What it does: This paper proposes an autoregressive strategy called PASPO, which directly generates valid actions for allocation tasks that satisfy linear hard constraints.
AutoSurvey: Large Language Models Can Automatically Write Surveys
Yidong Wang (Squirrel AI), Yue Zhang (Westlake University)
Large Language ModelTextReview/Survey PaperRetrieval-Augmented Generation
🎯 What it does: This paper presents AutoSurvey, a complete process for automatically generating large-scale academic reviews using large language models.
AutoTimes: Autoregressive Time Series Forecasters via Large Language Models
Yong Liu (Tsinghua University), Mingsheng Long (Tsinghua University)
TransformerLarge Language ModelPrompt EngineeringTime Series
🎯 What it does: AutoTimes is proposed, transforming off-the-shelf decoder-based large language models (LLMs) into autoregressive time series predictors by embedding time series segments in the hidden space of the LLM and generating future time steps using the autoregressive capabilities of the LLM. It supports predictions of arbitrary lengths and introduces position information based on text timestamps and context prompts (in-context forecasting).
AV-Cloud: Spatial Audio Rendering Through Audio-Visual Cloud Splatting
Mingfei Chen (University of Washington), Eli Shlizerman (University of Washington)
TransformerGaussian SplattingPoint CloudAudio
🎯 What it does: A point cloud-based audio-visual cloud rendering framework called AV-Cloud is proposed, which can synchronously generate high-quality spatial audio from any viewpoint without the need for pre-rendered images.
AV-GS: Learning Material and Geometry Aware Priors for Novel View Acoustic Synthesis
Swapnil Bhosale (University of Surrey), Xiatian Zhu (University of Surrey)
GenerationData SynthesisGaussian SplattingAudio
🎯 What it does: Proposes the AV-GS model, which utilizes 3D Gaussian splatting to learn scene geometry and implicit material information, and achieves new perspective binaural audio synthesis through audio-guided parameters;
AvaTaR: Optimizing LLM Agents for Tool Usage via Contrastive Reasoning
Shirley Wu (Stanford University), James Zou (Stanford University)
RetrievalOptimizationTransformerLarge Language ModelPrompt EngineeringContrastive LearningText
🎯 What it does: Proposes AVATAR, a framework for automating the optimization of large language model agents to enhance tool usage efficiency.
Average gradient outer product as a mechanism for deep neural collapse
Daniel Beaglehole (University of California San Diego), Mikhail Belkin (University of California San Diego)
Representation LearningConvolutional Neural NetworkSupervised Fine-TuningImage
🎯 What it does: The phenomenon of Deep Neural Collapse (DNC) in deep neural networks is studied and verified through the Average Gradient Outer Product (AGOP) mechanism.
AverNet: All-in-one Video Restoration for Time-varying Unknown Degradations
Haiyu Zhao (Sichuan University), Xi Peng (Sichuan University)
RestorationPrompt EngineeringOptical FlowVideo
🎯 What it does: This paper proposes the All-in-one Video Restoration Network AverNet, specifically designed for recovering from time-varying unknown degradations (TUD).
Avoiding Undesired Future with Minimal Cost in Non-Stationary Environments
Wen-Bo Du (Nanjing University), Zhi-Hua Zhou (Nanjing University)
OptimizationReinforcement LearningTime SeriesSequential
🎯 What it does: An AUF-MICNS algorithm is proposed, which maintains influence relationships through sequences in non-stationary environments and minimizes costs to avoid expected adverse outcomes in multi-round decision-making.
AWT: Transferring Vision-Language Models via Augmentation, Weighting, and Transportation
Yuhan Zhu (Nanjing University), Limin Wang (Nanjing University)
ClassificationDomain AdaptationTransformerLarge Language ModelVision Language ModelImageVideoTextMultimodality
🎯 What it does: Proposes the AWT (Augment-Weight-Transport) framework, which enhances the zero/few-shot adaptation performance of pre-trained vision-language models using visual and textual augmentation, entropy weighting, and optimal transport.
Axioms for AI Alignment from Human Feedback
Luise Ge (Washington University in St. Louis), Junlin Wu (Washington University in St. Louis)
Recommendation SystemOptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: In RLHF, learning the reward function is viewed as a social choice problem, constructing a linear social choice model and designing new aggregation rules;
B-ary Tree Push-Pull Method is Provably Efficient for Distributed Learning on Heterogeneous Data
Runze You (Chinese University of Hong Kong), Shi Pu (Chinese University of Hong Kong)
Federated LearningComputational EfficiencyConvolutional Neural NetworkImageTabularOrdinary Differential Equation
🎯 What it does: Proposes the B-ary Tree Push-Pull (BTPP) distributed learning algorithm, which efficiently propagates parameters and gradients using two tree structures;
B-cosification: Transforming Deep Neural Networks to be Inherently Interpretable
Shreyash Arya (Max Planck Institute for Informatics), Bernt Schiele (Max Planck Institute for Informatics)
Explainability and InterpretabilityComputational EfficiencyConvolutional Neural NetworkTransformerSupervised Fine-TuningImage
🎯 What it does: Transform the pre-trained deep neural network into a B-cos network through fine-tuning to achieve intrinsic interpretability.
B'MOJO: Hybrid State Space Realizations of Foundation Models with Eidetic and Fading Memory
Luca Zancato (Amazon Web Services AI Labs), Stefano Soatto (Amazon Web Services AI Labs)
TransformerLarge Language ModelTextSequential
🎯 What it does: A hybrid state space and attention architecture named B'MOJO is designed to achieve adaptive memory expansion under limited resources, supporting offline retrieval and online reasoning.
Back to the Continuous Attractor
Ábel Ságodi (Champalimaud Foundation), Il Memming Park (Champalimaud Foundation)
Recurrent Neural NetworkSequential
🎯 What it does: This paper explores the approximate stability and memory performance of continuous attractors under perturbations through theoretical analysis and RNN experiments with task training, demonstrating that systems approximating continuous attractors still exhibit functional robustness on behavioral time scales.
BackdoorAlign: Mitigating Fine-tuning based Jailbreak Attack with Backdoor Enhanced Safety Alignment
Jiongxiao Wang (University of Wisconsin-Madison), Chaowei Xiao (University of Wisconsin-Madison)
Safty and PrivacyAdversarial AttackTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A Backdoor-enhanced security alignment method is proposed, which utilizes a random secret prefix (similar to a backdoor trigger) added before secure examples, allowing the prefix to be added only during inference after fine-tuning to restore security alignment and defend against fine-tuning-based jailbreak attacks.
BackTime: Backdoor Attacks on Multivariate Time Series Forecasting
Xiao Lin (University of Illinois Urbana-Champaign), Hanghang Tong (University of Illinois Urbana-Champaign)
OptimizationAdversarial AttackGraph Neural NetworkGenerative Adversarial NetworkTime Series
🎯 What it does: This paper studies backdoor attacks on multivariate time series prediction models and proposes an attack method named BACKTIME.