NeurIPS 2024 Papers — Page 6
Conference on Neural Information Processing Systems · 4035 papers
Bridge the Points: Graph-based Few-shot Segment Anything Semantically
Anqi Zhang (Beijing Institute of Technology), Yunchao Wei (Beijing Jiaotong University)
SegmentationGraph Neural NetworkTransformerContrastive LearningImage
🎯 What it does: A graph-based few-shot semantic segmentation framework is proposed, utilizing SAM to automatically generate masks and achieving efficient hyperparameter-free semantic segmentation through point-mask clustering and gating.
Bridge-IF: Learning Inverse Protein Folding with Markov Bridges
Yiheng Zhu (Zhejiang University), Jian Wu (Zhejiang University)
GenerationProtein Structure PredictionGraph Neural NetworkTransformerLarge Language ModelSupervised Fine-TuningGraph
🎯 What it does: A generative model based on Markov bridge, named Bridge-IF, is proposed for inverse protein folding (generating foldable protein sequences from a given backbone structure). It generates a deterministic prior sequence through a structure encoder, and then gradually refines this prior on the Markov bridge, ultimately obtaining high-quality sequences that meet structural constraints.
Bridging Gaps: Federated Multi-View Clustering in Heterogeneous Hybrid Views
Xinyue Chen (University of Electronic Science and Technology of China), Yang Yang (University of Electronic Science and Technology of China)
Federated LearningAuto EncoderContrastive LearningImage
🎯 What it does: The FMCSC framework is proposed in federated multi-view clustering to address the issues of client differences and view differences caused by the mixture of single-view and multi-view clients.
Bridging Geometric States via Geometric Diffusion Bridge
Shengjie Luo (Peking University), Liwei Wang (Peking University)
OptimizationDrug DiscoveryGraph Neural NetworkDiffusion modelGraphTabularStochastic Differential Equation
🎯 What it does: A framework based on equivariant diffusion bridges (GDB) is proposed to accurately bridge the initial and target geometric states.
Bridging Model-Based Optimization and Generative Modeling via Conservative Fine-Tuning of Diffusion Models
Masatoshi Uehara (Genentech), Tommaso Biancalani (Genentech)
GenerationOptimizationReinforcement LearningDiffusion modelImageBiomedical Data
🎯 What it does: In the context of offline data scenarios, a dual conservative fine-tuning method based on a pre-trained diffusion model (BRAID) is proposed. During the fine-tuning process, uncertainty penalties are added to the reward model and KL regularization is incorporated into the policy learning, thereby suppressing over-optimization and avoiding the generation of invalid designs.
Bridging Multicalibration and Out-of-distribution Generalization Beyond Covariate Shift
Jiayun Wu (Tsinghua University), Steven Wu
Domain AdaptationOptimizationTabularBenchmark
🎯 What it does: A new model-free optimization framework is proposed, which extends multicalibration with a joint grouping function and achieves robust learning for out-of-distribution (OOD) generalization. The MC-PseudoLabel algorithm is introduced, aligning multicalibration with invariance.
Bridging OOD Detection and Generalization: A Graph-Theoretic View
Han Wang (University of Illinois), Yixuan Li (University of Wisconsin)
Domain AdaptationAnomaly DetectionGraph Neural NetworkContrastive LearningImage
🎯 What it does: This paper proposes a graph-theory-based framework that jointly addresses the problems of OOD generalization and detection, utilizing the decomposition of the graph's adjacency matrix to obtain a closed-form representation.
Bridging semantics and pragmatics in information-theoretic emergent communication
Eleonora Gualdoni (Apple), Noga Zaslavsky (New York University)
Object DetectionOptimizationReinforcement LearningAuto EncoderImage
🎯 What it does: Under unsupervised local context interaction, artificial intelligence agents self-play through an information theory framework, co-evolving a shared vocabulary that is both semantic and pragmatic, and evaluating their human similarity using multi-objective loss.
Bridging the Divide: Reconsidering Softmax and Linear Attention
Dongchen Han (Tsinghua University), Gao Huang (Tsinghua University)
Object DetectionSegmentationTransformerContrastive LearningImage
🎯 What it does: This study investigates the differences between linear attention and Softmax attention, proposing two attributes for improvement: injectivity and local modeling.
Bridging The Gap between Low-rank and Orthogonal Adaptation via Householder Reflection Adaptation
Shen Yuan (Renmin University of China), Hongteng Xu (Renmin University of China)
GenerationDomain AdaptationOptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A parameter-efficient fine-tuning method based on Householder reflections (HRA) is proposed for adapting large models to various downstream tasks.
Buffer of Thoughts: Thought-Augmented Reasoning with Large Language Models
Ling Yang (Peking University), Bin CUI
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: The Buffer of Thoughts (BoT) framework is proposed, which establishes a meta-buffer to store high-level thinking templates, supporting large language models (LLMs) in efficient, accurate, and robust reasoning in multi-task scenarios.
Building a stable classifier with the inflated argmax
Jake A Soloff, Rebecca Willett (University of Chicago)
ClassificationImage
🎯 What it does: A hypothesis-free, sample perturbation robust multi-classifier is constructed by combining bagging with 'inflated argmax'.
Building on Efficient Foundations: Effective Training of LLMs with Structured Feedforward Layers
Xiuying Wei (École Polytechnique Fédérale de Lausanne), Caglar Gulcehre (École Polytechnique Fédérale de Lausanne)
TransformerLarge Language ModelText
🎯 What it does: The study investigates the use of structured linear layers (low-rank, block shuffling, block dense) in Transformer large language models as a replacement for traditional FFN, and validates its feasibility under custom training scales.
Byzantine Robustness and Partial Participation Can Be Achieved at Once: Just Clip Gradient Differences
Grigory Malinovsky (King Abdullah University of Science and Technology), Eduard Gorbunov (Mohamed bin Zayed University of Artificial Intelligence)
OptimizationFederated LearningConvolutional Neural NetworkReinforcement LearningImageTabular
🎯 What it does: A new distributed optimization algorithm, Byz-VR-MARINA-PP, is proposed to achieve robust training in scenarios with Byzantine attackers and partial client participation.
C-GAIL: Stabilizing Generative Adversarial Imitation Learning with Control Theory
Tianjiao Luo (Tsinghua University), Jun Zhu (Tsinghua University)
OptimizationReinforcement LearningGenerative Adversarial NetworkSequential
🎯 What it does: By modeling the training process of GAIL as a continuous dynamical system, this paper analyzes its non-convergence issues and designs a linear negative feedback controller based on control theory. Ultimately, it proposes a regularization (C-GAIL) that can be directly added to the discriminator loss to stabilize and accelerate the convergence of GAIL.
CA-SSLR: Condition-Aware Self-Supervised Learning Representation for Generalized Speech Processing
Yen-Ju Lu (Johns Hopkins University), Jesus Villalba (Johns Hopkins University)
RecognitionRepresentation LearningContrastive LearningAudio
🎯 What it does: Proposes Condition-Aware Self-Supervised Learning Representation (CA-SSLR), which inserts lightweight conditional adapters into the pre-trained SSL encoder, enabling the model to perceive language and speaker information, supporting multi-tasks such as LID, ASR, and SV without large-scale fine-tuning.
Cal-DPO: Calibrated Direct Preference Optimization for Language Model Alignment
Teng Xiao (Pennsylvania State University), Vasant G Honavar (Pennsylvania State University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningText
🎯 What it does: This paper proposes a language model alignment method called Cal-DPO, which calibrates the implicit rewards in contrastive preference learning to maintain the same scale as the true rewards, thereby avoiding the issue of rewards decreasing during training that is common in traditional methods.
CALANet: Cheap All-Layer Aggregation for Human Activity Recognition
Jaegyun Park (Chung Ang University), Jaesung Lee (Chung Ang University)
RecognitionConvolutional Neural NetworkTime Series
🎯 What it does: A lightweight convolutional network called CALANet is proposed, which utilizes feature aggregation from all layers to enhance human action recognition accuracy.
Calibrated Self-Rewarding Vision Language Models
Yiyang Zhou (University of North Carolina at Chapel Hill), Huaxiu Yao (National University of Singapore)
GenerationOptimizationTransformerReinforcement LearningVision Language ModelImageTextMultimodality
🎯 What it does: Calibrating self-reward for large-scale visual language models (LVLM) to make the model pay more attention to image information when generating text, thereby reducing hallucinations and enhancing cross-modal alignment.
Calibrating Reasoning in Language Models with Internal Consistency
Zhihui Xie (Shanghai Jiao Tong University), Shuai Li (Shanghai Jiao Tong University)
TransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper proposes Internal Consistency as a self-assessment and calibration metric for the reasoning process of large language models (LLMs), and utilizes this metric to improve the accuracy of Chain of Thought (CoT) reasoning.
CALVIN: Improved Contextual Video Captioning via Instruction Tuning
Gowthami Somepalli (University of Maryland), David W. Jacobs
GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelVideoTextMultimodalityAudio
🎯 What it does: Developed the CALVIN model, which utilizes video LLM combined with contextual information to generate more contextually appropriate scene descriptions/audio descriptions.
Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs
Shengbang Tong (New York University), Saining Xie (New York University)
RecognitionData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodalityBenchmark
🎯 What it does: A fully open-source, vision-centered multimodal large language model (MLLM) family called Cambrian-1 has been developed, systematically evaluating visual encoders, connector designs, instruction tuning data, and benchmarks, constructing a new vision-centered benchmark CV-Bench and releasing complete resources.
Can an AI Agent Safely Run a Government? Existence of Probably Approximately Aligned Policies
Frédéric Berdoz (ETH Zurich), Roger Wattenhofer (ETH Zurich)
Reinforcement LearningAgentic AI
🎯 What it does: The concept of quantifying AI alignment in social decision-making environments is proposed, along with theoretical guarantees for Possible Approximate Alignment (PAA) and safety policies.
Can Graph Learning Improve Planning in LLM-based Agents?
Xixi Wu (Fudan University), Dongsheng Li (Microsoft Research Asia)
Graph Neural NetworkTransformerLarge Language ModelTextGraph
🎯 What it does: This paper proposes modeling task planning as a graph decision problem and combines graph neural networks with large language models (LLM) to achieve subtask retrieval, significantly improving planning effectiveness.
Can Graph Neural Networks Expose Training Data Properties? An Efficient Risk Assessment Approach
Hanyang Yuan (Zhejiang University), Yang Yang (Finvolution Group)
Adversarial AttackGraph Neural NetworkGraph
🎯 What it does: This paper studies attribute inference attacks based on graph neural networks and proposes an efficient method to generate approximate shadow models using model approximation.
Can Language Models Learn to Skip Steps?
Tengxiao Liu (University of California Santa Barbara), Zheng Zhang (Amazon)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: Exploring and training language models to learn step skipping reasoning, enhancing reasoning efficiency and generalization ability.
Can Language Models Perform Robust Reasoning in Chain-of-thought Prompting with Noisy Rationales?
Zhanke Zhou (Hong Kong Baptist University), Bo Han (Hong Kong Baptist University)
Large Language ModelPrompt EngineeringContrastive LearningTextChain-of-Thought
🎯 What it does: This paper proposes a method for robustness assessment and enhancement against noisy reasoning (invalid or erroneous reasoning steps) in chain-of-thought prompts, constructing a dedicated NoRa dataset and designing a CD-CoT comparative denoising framework;
Can Large Language Model Agents Simulate Human Trust Behavior?
Chengxing Xie (King Abdullah University of Science and Technology), Guohao Li
TransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This study investigates whether large language model (LLM) agents can simulate human trust behavior by participating in trust games, analyzing their decision-making and reasoning processes.
Can large language models explore in-context?
Akshay Krishnamurthy (Microsoft Research), Aleksandrs Slivkins (Microsoft Research)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper systematically evaluates the native exploration capabilities of existing large language models (LLMs) without training interventions by fully describing a multi-armed bandit environment in the prompts, allowing the model to interact directly as a decision-making agent.
Can Learned Optimization Make Reinforcement Learning Less Difficult?
Alexander D. Goldie (University of Oxford), Jakob Nicolaus Foerster
OptimizationMeta LearningRecurrent Neural NetworkReinforcement LearningTabular
🎯 What it does: To address the challenges of non-stationarity, plasticity loss, and exploration in reinforcement learning, this paper proposes and trains a meta-learning optimizer called OPEN.
Can LLMs Implicitly Learn Numeric Parameter Constraints in Data Science APIs?
Yinlin Deng (University of Illinois Urbana-Champaign), LINGMING ZHANG
OptimizationAI Code AssistantTransformerLarge Language ModelTextBenchmark
🎯 What it does: This paper systematically studies the understanding of numerical parameter constraints by large language models (LLMs) when generating data science (DS) programs, particularly through an empirical analysis of 28 APIs from the PyTorch and NumPy libraries.
Can LLMs Learn by Teaching for Better Reasoning? A Preliminary Study
Xuefei Ning (Tsinghua University), Yu Wang (Tsinghua University)
TransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper explores whether large language models (LLMs) can enhance reasoning abilities through a method called 'Learning by Teaching' (LbT). It proposes three teacher-student interaction-based improvement methods, M1, M2, and M3, and conducts experiments on mathematical reasoning, code synthesis, and text classification tasks.
Can Models Learn Skill Composition from Examples?
Haoyu Zhao (Princeton University), Sanjeev Arora (Princeton University)
TransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringTextBenchmark
🎯 What it does: Fine-tuned small language models (LLaMA‑2‑13B‑Chat, Mistral‑7B‑Instruct‑v0.2) to combine multiple language skills in short texts.
Can neural operators always be continuously discretized?
Takashi Furuya (Shimane University), Maarten V. de Hoop (Rice University)
🎯 What it does: This paper studies the discretization problem of neural operators in infinite-dimensional Hilbert spaces and provides criteria for the feasibility of discretization.
Can Simple Averaging Defeat Modern Watermarks?
Pei Yang (National University of Singapore), Mike Zheng Shou (National University of Singapore)
Adversarial AttackDiffusion modelImageAudio
🎯 What it does: This paper studies the vulnerability of content-agnostic digital watermarks and proposes a steganalysis attack based on simple averaging, which can effectively remove or forge watermarks in both gray-box and black-box scenarios.
Can Transformers Smell Like Humans?
Farzaneh Taleb (KTH Royal Institute of Technology), Danica Kragic (KTH Royal Institute of Technology)
TransformerTabular
🎯 What it does: This study validates the potential of unsupervised models in predicting olfactory attributes by aligning the representations encoded by the pre-trained chemical structure Transformer (MoLFormer) with human olfactory perception.
Can We Leave Deepfake Data Behind in Training Deepfake Detector?
Jikang Cheng (Wuhan University), Chen Li (Tencent Inc)
ClassificationRecognitionConvolutional Neural NetworkGenerative Adversarial NetworkImageVideo
🎯 What it does: A deepfake detection method called ProDet is proposed, which organizes three types of samples: real, blendfake, and deepfake into the latent space in a progressive manner, and enhances detection performance through joint training.
Capturing the denoising effect of PCA via compression ratio
Chandra Sekhar Mukherjee (University of Southern California), Jiapeng Zhang (University of Southern California)
Anomaly DetectionTabularBiomedical Data
🎯 What it does: A compression ratio metric is proposed to quantify the denoising effect of PCA on high-dimensional noisy data with community structure, and a simple anomaly detection method is designed based on this metric.
Cardinality-Aware Set Prediction and Top-$k$ Classification
Corinna Cortes (Google Research), Yutao Zhong (Courant Institute)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: This paper proposes a Set prediction framework based on card number perception and designs two new cost-sensitive approximate losses;
Carrot and Stick: Eliciting Comparison Data and Beyond
Yiling Chen (Harvard University), Fang-Yi Yu (George Mason University)
GraphTabular
🎯 What it does: This paper designs a dual prediction mechanism based on Bonus-Penalty Payment (BPP) to induce participants to report comparative data and network data truthfully in an environment without verifiable true values.
Cascade of phase transitions in the training of energy-based models
Dimitrios Bachtis (Paris Sciences et Lettres), Beatriz Seoane (Complutense University of Madrid)
GenerationImageBiomedical DataPhysics Related
🎯 What it does: This study investigates the feature encoding mechanism of Restricted Boltzmann Machines (RBM) during the training process, finding that the singular value decomposition of the weight matrix exhibits a series of second-order phase transitions over time, gradually learning the main features of the data.
Cascade Speculative Drafting for Even Faster LLM Inference
Ziyi Chen (University of Illinois), Jie Huang (University of Illinois)
OptimizationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: A new inference acceleration algorithm is proposed - Cascade Speculative Drafting (CS Drafting), which achieves faster LLM inference through two levels of cascading (vertical cascading and horizontal cascading).
CAT: Coordinating Anatomical-Textual Prompts for Multi-Organ and Tumor Segmentation
Zhongzhen Huang (Shanghai Jiao Tong University), Xiaofan Zhang (Shanghai Jiao Tong University)
SegmentationTransformerPrompt EngineeringContrastive LearningImageBiomedical DataComputed Tomography
🎯 What it does: Designed and implemented the CAT model, which automatically completes the segmentation of multiple abdominal organs and tumors using 3D anatomical volume visual cues and medical text prompts.
CAT3D: Create Anything in 3D with Multi-View Diffusion Models
Ruiqi Gao (Google DeepMind), Ben Poole (Google DeepMind)
GenerationData SynthesisDiffusion modelNeural Radiance FieldImageVideo
🎯 What it does: Generate multiple consistent views through a multi-view diffusion model, and then use a robust 3D reconstruction pipeline to create real-time renderable 3D scenes, supporting the rapid creation of 3D content from a single or a few real/generated images or even text prompts.
Catastrophic Goodhart: regularizing RLHF with KL divergence does not mitigate heavy-tailed reward misspecification
Thomas Kwa (Independent / FAR Labs), Adrià Garriga-Alonso (FAR AI)
Reinforcement Learning from Human FeedbackReinforcement LearningText
🎯 What it does: This study investigates the impact of KL regularization on reward error in RLHF, proposing the phenomenon of 'catastrophic Goodhart'.
Categorical Flow Matching on Statistical Manifolds
Chaoran Cheng (University of Illinois Urbana-Champaign), Ge Liu (University of Illinois Urbana-Champaign)
GenerationData SynthesisFlow-based ModelTextTabularOrdinary Differential Equation
🎯 What it does: A framework for generation on statistical manifolds is proposed - Statistical Flow Matching (SFM), which is applied to the generation of discrete categorical data.
Causal Context Adjustment Loss for Learned Image Compression
Minghao Han (University of Electronic Science and Technology of China), Shuhang Gu (University of Electronic Science and Technology of China)
CompressionConvolutional Neural NetworkAuto EncoderImage
🎯 What it does: A loss function named Causal Context Adjustment Loss (CCA-loss) is proposed to explicitly adjust the causal context in learning-based image compression, allowing the autoencoder to carry more important information beneficial for subsequent predictions in the early stages of the autoregressive entropy model; at the same time, a CNN-based autoencoder and an autoregressive entropy model with uneven channel grouping are employed to enhance encoding speed and compression performance.
Causal Contrastive Learning for Counterfactual Regression Over Time
Mouad El Bouchattaoui (Paris-Saclay University), Paul-Henry Cournède (Saint-Gobain)
Recurrent Neural NetworkContrastive LearningTime SeriesSequentialBiomedical Data
🎯 What it does: A causal contrastive learning model based on RNN is proposed for inverse causal regression of time series, capable of predicting potential outcomes several steps ahead given historical records.
Causal Deciphering and Inpainting in Spatio-Temporal Dynamics via Diffusion Model
Yifan Duan (University of Science and Technology of China), Xuelong Li
GenerationExplainability and InterpretabilityTransformerDiffusion modelTime Series
🎯 What it does: This paper achieves causal region partitioning and completion of spatiotemporal data through causal identification and diffusion models, enhancing the generalization and interpretability of prediction models.
Causal Dependence Plots
Joshua R. Loftus (London School of Economics), Sakina Hansen (London School of Economics)
Explainability and InterpretabilityTabularBiomedical Data
🎯 What it does: Proposed and implemented Causal Dependence Plots (CDPs) for visualizing the causal dependence of machine learning models on individual features, and compared them with traditional non-causal explanation methods (such as PDP, ICE, etc.).
Causal Discovery from Event Sequences by Local Cause-Effect Attribution
Joscha Cüppers (CISPA Helmholtz Center for Information Security), Jilles Vreeken (CISPA Helmholtz Center for Information Security)
Time SeriesSequentialFinance Related
🎯 What it does: This study investigates how to discover causal structures from event sequences, proposing a causal model based on one-to-one event matching and introducing the CASCADE algorithm.
Causal discovery with endogenous context variables
Wiebke Günther (German Aerospace Center), Jakob Runge (German Aerospace Center)
Tabular
🎯 What it does: This paper proposes a constraint-based causal discovery method suitable for endogenous contextual variables (PC-AC), which can simultaneously recover the causal graph for each context and its joint graph.
Causal Effect Identification in a Sub-Population with Latent Variables
Amir Mohammad Abouei (École Polytechnique Fédérale de Lausanne), Matthias Grossglauser (École Polytechnique Fédérale de Lausanne)
🎯 What it does: This paper proposes a theoretical framework for identifying causal effects in subpopulations (S-ID) under the presence of latent variables, and provides conditions for determination and a recursive solving algorithm.
Causal Imitation for Markov Decision Processes: a Partial Identification Approach
Kangrui Ruan (Columbia University), Elias Bareinboim (Columbia University)
Autonomous DrivingOptimizationReinforcement LearningGenerative Adversarial NetworkTabularBiomedical Data
🎯 What it does: The study investigates how to achieve robust imitation learning using partial identification theory in the presence of unobserved confounding in Markov Decision Processes (MDP), and proposes two algorithms (CAIL-R and CAIL-T) to achieve expert performance when transitions or rewards are identifiable.
Causal Inference in the Closed-Loop: Marginal Structural Models for Sequential Excursion Effects
Alexander W. Levis (Carnegie Mellon University), Francisco Pereira (National Institutes of Health)
TabularTime Series
🎯 What it does: A non-parametric causal inference framework for closed-loop optogenetic experiments is proposed, utilizing the Historical Restricted Marginal Structure Model (HR-MSM) to estimate multi-step 'sequential sprint effects'.
Causal language modeling can elicit search and reasoning capabilities on logic puzzles
Kulin Shah (University of Texas at Austin), Rina Panigrahy (Google Research)
TransformerLarge Language ModelText
🎯 What it does: This paper investigates whether causal language models (Transformers) can solve logical puzzles (such as Sudoku and chess puzzles) by training solely on the next-word prediction task, and explores the impact of training order on the model's reasoning ability.
Causal Temporal Representation Learning with Nonstationary Sparse Transition
Xiangchen Song (Carnegie Mellon University), Kun Zhang (Carnegie Mellon University)
Representation LearningAuto EncoderVideoTime Series
🎯 What it does: This paper proposes a causal time representation learning framework called CtrlNS based on sparse transfer, which can identify potential causal processes and distribution transfers from non-stationary time series data without relying on domain variable priors or Markov assumptions.
Causal vs. Anticausal merging of predictors
Sergio Hernan Garrido Mejia (Max Planck Institute for Intelligent Systems), Dominik Janzing (Amazon)
ClassificationOptimizationTabular
🎯 What it does: The study examines the differences in predictive models generated by Causal Maximum Entropy (CMAXENT) when merging predictors in causal and counterfactual directions.
CausalDiff: Causality-Inspired Disentanglement via Diffusion Model for Adversarial Defense
Mingkun Zhang (Chinese Academy of Sciences), Xueqi Cheng (Chinese Academy of Sciences)
ClassificationAnomaly DetectionAdversarial AttackDiffusion modelImage
🎯 What it does: A CausalDiff framework based on causal diffusion models has been developed, utilizing diffusion models for conditional image generation, and separating causal features associated with labels from non-causal features during the generation process, thereby achieving robust classification in the face of unseen adversarial attacks.
CausalStock: Deep End-to-end Causal Discovery for News-driven Multi-stock Movement Prediction
Shuqi Li (Renmin University of China), Rui Yan (Renmin University of China)
Recommendation SystemOptimizationExplainability and InterpretabilityComputational EfficiencyRecurrent Neural NetworkTransformerLarge Language ModelTextTabularTime SeriesFinance Related
🎯 What it does: The CausalStock framework is proposed for predicting the price trends of multiple stocks based on news, capable of simultaneously learning the causal relationships between stocks and utilizing denoised news representations to achieve more accurate predictions.
CE-NAS: An End-to-End Carbon-Efficient Neural Architecture Search Framework
Yiyang Zhao (Worcester Polytechnic Institute), Tian Guo (Worcester Polytechnic Institute)
OptimizationComputational EfficiencyNeural Architecture SearchTransformerReinforcement LearningTime Series
🎯 What it does: The CE-NAS framework is proposed, which utilizes reinforcement learning to dynamically allocate GPU resources, prioritizing low-carbon one/shot NAS during high/low carbon periods and high-energy-consuming Vanilla NAS, thereby significantly reducing carbon emissions while ensuring search quality.
Cell ontology guided transcriptome foundation model
Xinyu Yuan (University of Montréal), Jian Tang (University of Montréal)
TransformerBiomedical Data
🎯 What it does: A basic model for single-cell transcriptomics, scCello, utilizing cell ontology knowledge is proposed, and unsupervised pre-training of gene expression is achieved through a triple loss function.
CemiFace: Center-based Semi-hard Synthetic Face Generation for Face Recognition
Zhonglin Sun (Queen Mary University of London), Georgios Tzimiropoulos (Queen Mary University of London)
RecognitionGenerationData SynthesisDiffusion modelImage
🎯 What it does: A facial synthesis method based on diffusion models, CemiFace, is proposed, which generates semi-hard samples with moderate similarity to the identity center using a similarity control factor, aimed at training more discriminative facial recognition models.
Certified Adversarial Robustness via Randomized $\alpha$-Smoothing for Regression Models
Aref Miri Rekavandi (University of Melbourne), Benjamin I. P. Rubinstein
OptimizationAdversarial AttackTabular
🎯 What it does: A randomized smoothing method based on α-trimmed smoothing is proposed, providing a probabilistic robustness proof against ℓp attacks for regression models and achieving certification in a black-box setting.
Certified Machine Unlearning via Noisy Stochastic Gradient Descent
Eli Chien (Georgia Institute of Technology), Pan Li (Georgia Institute of Technology)
OptimizationSafty and PrivacyImage
🎯 What it does: This paper proposes the use of Projected Noise Stochastic Gradient Descent (PNSGD) to achieve the unlearning functionality of machine learning models, providing an approximate no-learning guarantee under convex assumptions.
Certified Robustness for Deep Equilibrium Models via Serialized Random Smoothing
Weizhi Gao (North Carolina State University), Xiaorui Liu (North Carolina State University)
ClassificationComputational EfficiencyGaussian SplattingImage
🎯 What it does: This paper is the first to apply randomized smoothing methods to deep equilibrium models (DEQ) and proposes Serialized Randomized Smoothing (SRS) to significantly accelerate the provable robustness evaluation of DEQ.
Chain of Agents: Large Language Models Collaborating on Long-Context Tasks
Yusen Zhang (Penn State University), Sercan O Arik
TransformerLarge Language ModelAgentic AIText
🎯 What it does: Proposes the Chain-of-Agents (CoA) framework, which utilizes multi-agent LLMs to collaboratively handle long-context tasks by reading in segments and gradually passing information, ultimately generating answers by a manager.
Chain of Preference Optimization: Improving Chain-of-Thought Reasoning in LLMs
Xuan Zhang (Singapore Management University), Min Lin (Sea AI Lab)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought
🎯 What it does: When performing chain reasoning with large language models, we first use Tree-of-Thought (ToT) search to obtain multiple lines of thought, and then pair the 'preferred' and 'deteriorated' thoughts generated at each step of the search tree, utilizing this preference data to fine-tune the model during the training phase (Chain of Preference Optimization, CPO).
Chain of Thoughtlessness? An Analysis of CoT in Planning
Kaya Stechly (Arizona State University), Subbarao Kambhampati (Arizona State University)
TransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought
🎯 What it does: This paper systematically evaluates the effectiveness of Chain of Thought (CoT) in solving classic planning tasks (Blocksworld) using large language models (LLMs) and explores the impact of prompt granularity on reasoning performance.
Chain-of-Thought Reasoning Without Prompting
Xuezhi Wang (Google DeepMind), Denny Zhou (Google DeepMind)
TransformerLarge Language ModelTextChain-of-Thought
🎯 What it does: This paper proposes a new reasoning method—CoT-decoding, which explores reasoning paths using the top-k candidate words from a pre-trained large language model in the first decoding step, without requiring any prompts.
Challenges of Generating Structurally Diverse Graphs
Fedor Velikonivtsev (Higher School of Economics University), Liudmila Prokhorenkova (Yandex Research)
GenerationData SynthesisOptimizationGraph Neural NetworkGraph
🎯 What it does: Research and implementation of various methods for generating structurally diverse graph collections to meet the evaluation needs of graph algorithms and graph neural networks.
Changing the Training Data Distribution to Reduce Simplicity Bias Improves In-distribution Generalization
Dang Nguyen, Baharan Mirzasoleiman (University of California Los Angeles)
ClassificationOptimizationConvolutional Neural NetworkImage
🎯 What it does: A technique based on changes in training data distribution (USEFUL) is proposed, which clusters and separates easily learnable feature samples early in training, followed by a single up-sampling of the remaining samples and re-training, thereby reducing the 'simplicity bias' of gradient descent and improving ID generalization performance on the original distribution; a theoretical analysis of the learning dynamics of SAM and GD in a two-layer CNN is also conducted.
CHASE: Learning Convex Hull Adaptive Shift for Skeleton-based Multi-Entity Action Recognition
Yuhang Wen (Sun Yat-sen University), Beichen Ding (Sun Yat-sen University)
RecognitionPose EstimationGraph Neural NetworkTransformerVideoMultimodality
🎯 What it does: A skeleton multi-entity action recognition framework based on Convex Hull Adaptive Translation (CHASE) is proposed, utilizing sample-adaptive coordinate origin relocation to alleviate distribution differences between different entities.
Chat-Scene: Bridging 3D Scene and Large Language Models with Object Identifiers
Haifeng Huang (Zhejiang University), Zhou Zhao (Zhejiang University)
Object DetectionRepresentation LearningTransformerLarge Language ModelVision Language ModelMultimodalityPoint Cloud
🎯 What it does: This study investigates 3D multimodal large language models, proposing object identifiers and object-level representations, unifying scene-language tasks into a question-answer format.
ChatCam: Empowering Camera Control through Conversational AI
Xinhang Liu (Hong Kong University of Science and Technology), Chi-Keung Tang (Dartmouth College)
GenerationOptimizationTransformerLarge Language ModelNeural Radiance FieldImageText
🎯 What it does: The ChatCam system is proposed, allowing users to directly control camera trajectories through natural language using conversational AI.
ChatQA: Surpassing GPT-4 on Conversational QA and RAG
Zihan Liu (NVIDIA), Bryan Catanzaro (NVIDIA)
GenerationRetrievalTransformerLarge Language ModelSupervised Fine-TuningTextBenchmarkRetrieval-Augmented Generation
🎯 What it does: This paper proposes the ChatQA model, which utilizes two-stage instruction fine-tuning and a specialized dense retriever to achieve performance exceeding that of GPT-4 on conversational question answering and retrieval-augmented generation (RAG) tasks, and creates the CHATRAG BENCH benchmark; it also publicly releases the model weights, training data, and retriever.
ChatTracker: Enhancing Visual Tracking Performance via Chatting with Multimodal Large Language Model
Yiming Sun (East China Normal University), Changbo Wang (East China Normal University)
Object TrackingTransformerLarge Language ModelPrompt EngineeringVision Language ModelTextMultimodality
🎯 What it does: Combining multimodal large language models to generate high-quality target descriptions, and optimizing iterative improvements through reflective prompts to enhance visual tracking performance.
Cherry on Top: Parameter Heterogeneity and Quantization in Large Language Models
Wanyun Cui (Shanghai University of Finance and Economics), Qianle Wang (Shanghai University of Finance and Economics)
OptimizationTransformerLarge Language ModelTextBenchmark
🎯 What it does: This study investigates the heterogeneity of parameters in large language models (LLMs) and proposes a mixed-precision quantization method called CherryQ, which retains a small number of key parameters (the 'cherry' parameters) at high precision while quantizing the remaining parameters to low bits.
Chimera: Effectively Modeling Multivariate Time Series with 2-Dimensional State Space Models
Ali Behrouz (Cornell University), Ramin Zabih (Cornell University)
Anomaly DetectionTransformerTime SeriesSequentialElectrocardiogramAudio
🎯 What it does: A three-headed 2-dimensional state space model (Chimera) is proposed, capable of simultaneously modeling long-term trends, seasonality, and cross-variation dependencies along the time and variation axes, achieving efficient training and inference through learnable discretization and 2-D selection scanning.
ChronoEpilogi: Scalable Time Series Selection with Multiple Solutions
Etienne Vareille (CY Cergy Paris Universite), Vassilis Christophides
Time Series
🎯 What it does: A scalable time series variable selection method called ChronoEpilogi is proposed, which can simultaneously identify all minimal predictive subsets (Markov Boundary) and provide an equivalent class representation of interchangeable variables.
CIFD: Controlled Information Flow to Enhance Knowledge Distillation
Yashas Malur Saidutta (Samsung Research America), Hongxia Jin (Samsung Research America)
Knowledge DistillationConvolutional Neural NetworkContrastive LearningImage
🎯 What it does: A knowledge distillation framework for controlling information flow, CIFD, is proposed, which replaces the expensive teacher assistant with teacher embeddings and an information compression module, and adds an information bottleneck module on the student side to prevent overfitting.
CigTime: Corrective Instruction Generation Through Inverse Motion Editing
Qihang Fang (University of Hong Kong), Yanchao Yang (University of Hong Kong)
GenerationData SynthesisRobotic IntelligenceTransformerLarge Language ModelSupervised Fine-TuningDiffusion modelText
🎯 What it does: This study investigates the task of generating corrective instructions from source actions to target actions, utilizing a motion editing process to automatically generate large-scale triplets and fine-tuning with a large language model to achieve the mapping from action differences to textual instructions.
CLAP4CLIP: Continual Learning with Probabilistic Finetuning for Vision-Language Models
Saurav Jha (University of New South Wales), Lina Yao (CSIRO Data61)
ClassificationKnowledge DistillationRepresentation LearningTransformerMixture of ExpertsVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a CLIP-based probabilistic fine-tuning framework called CLAP for continual learning, which can learn task-specific distributions in a multi-task flow while retaining pre-trained knowledge.
Class Distribution Shifts in Zero-Shot Learning: Learning Robust Representations
Yuli Slavutsky (Hebrew University of Jerusalem), Yuval Benjamini (Hebrew University of Jerusalem)
Domain AdaptationRepresentation LearningContrastive LearningImage
🎯 What it does: This study investigates the issue of class distribution shift caused by unknown attributes in zero-shot learning and proposes a robust representation learning algorithm based on synthetic environments and environment balancing.
Classification Diffusion Models: Revitalizing Density Ratio Estimation
Shahar Yadin (Technion Israel Institute of Technology), Tomer Michaeli (Technion Israel Institute of Technology)
GenerationData SynthesisDiffusion modelImage
🎯 What it does: A generative model based on density ratio estimation is proposed - the Classification Diffusion Model (CDM), which achieves pixel-level denoising and image generation by training a noise level classifier.
Classification Done Right for Vision-Language Pre-Training
Zilong Huang (ByteDance Research), Haoqi Fan (ByteDance Research)
ClassificationRetrievalTransformerVision Language ModelContrastive LearningImageTextMultimodality
🎯 What it does: SuperClass is proposed, a simple method that directly uses the raw text tokens from image-text data as classification labels for pre-training;
Classifier Clustering and Feature Alignment for Federated Learning under Distributed Concept Drift
Junbao Chen (Beijing Institute of Technology), Lu Huang (Beijing Institute of Technology)
Federated LearningContrastive LearningImage
🎯 What it does: This paper proposes the FedCCFA framework, which combines classifier clustering and feature alignment to address the issues of distributed concept drift and data heterogeneity in federated learning.
Classifier-guided Gradient Modulation for Enhanced Multimodal Learning
Zirun Guo (Zhejiang University), Zhou Zhao (Zhejiang University)
ClassificationRecognitionSegmentationTransformerMultimodalityMagnetic Resonance Imaging
🎯 What it does: A classifier-guided gradient modulation method (CGGM) is proposed to balance the multimodal learning process.
ClavaDDPM: Multi-relational Data Synthesis with Cluster-guided Diffusion Models
Wei Pang (University of Waterloo), Xi He (University of Waterloo)
Data SynthesisDiffusion modelTabular
🎯 What it does: A synthetic data generation framework for multi-relational databases, ClavaDDPM, is proposed, which can synthesize multi-table data while preserving primary and foreign key relationships.
CLIP in Mirror: Disentangling text from visual images through reflection
Tiancheng Wang (Beihang University), Baochang Zhang
ClassificationRecognitionAdversarial AttackTransformerVision Language ModelContrastive LearningImageText
🎯 What it does: A zero-shot dual-stream framework called MirrorCLIP is proposed, which inputs the original image and its horizontally flipped version into the CLIP encoder. It generates separate masks to achieve decoupling of text and visual features by utilizing the invariance of visual features to horizontal flipping and the difference in invariance of text features.
CLIPAway: Harmonizing focused embeddings for removing objects via diffusion models
Yiğit Ekin (Bilkent University), Aysegul Dundar (Bilkent University)
Image HarmonizationRestorationGenerationDiffusion modelImage
🎯 What it does: This paper studies a plug-and-play module called CLIPAway, which utilizes AlphaCLIP embeddings to achieve an image editing method for object removal that fills in the background without generating hallucinations.
CLIPCEIL: Domain Generalization through CLIP via Channel rEfinement and Image-text aLignment
Xi Yu (Brookhaven National Laboratory), Yuewei Lin (Brookhaven National Laboratory)
Domain AdaptationTransformerVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: A lightweight adapter is added to the CLIP pre-trained model, utilizing channel refinement and image-text alignment to achieve domain-invariant and class-related feature extraction, thereby enhancing cross-domain generalization performance.
CLIPLoss and Norm-Based Data Selection Methods for Multimodal Contrastive Learning
Yiping Wang (University of Washington), Simon Shaolei Du (University of Washington)
RetrievalRepresentation LearningData-Centric LearningContrastive LearningImageTextMultimodality
🎯 What it does: Two new data selection metrics are proposed: negCLIPLoss (using the negative value of the CLIP training loss as a quality score) and NormSim (using the p-norm similarity of visual embeddings to compare the target task data) to enhance data selection effectiveness in multimodal contrastive learning.
Closed-Loop Visuomotor Control with Generative Expectation for Robotic Manipulation
Qingwen Bu (Shanghai AI Lab), Hongyang Li (Shanghai AI Lab)
Robotic IntelligenceReinforcement LearningDiffusion modelOptical FlowVideoMultimodality
🎯 What it does: This paper presents CLOVER, a closed-loop visual-motor control framework designed for long-sequence robotic operations.
Cloud Object Detector Adaptation by Integrating Different Source Knowledge
Shuaifeng Li (University of Electronic Science and Technology of China), Xiatian Zhu (University of Surrey)
Object DetectionDomain AdaptationKnowledge DistillationConvolutional Neural NetworkPrompt EngineeringVision Language ModelContrastive LearningImage
🎯 What it does: Proposes the Cloud Object Detector Adaptation (CODA) task, which utilizes the detection results from large cloud models and integrates knowledge from the open-source vision-language model CLIP to train detectors in an unlabeled target domain.
CLUES: Collaborative Private-domain High-quality Data Selection for LLMs via Training Dynamics
Wanru Zhao (University of Cambridge), Nicholas Donald Lane
Federated LearningData-Centric LearningTransformerLarge Language ModelSupervised Fine-TuningTextTabularBiomedical DataFinance Related
🎯 What it does: This paper proposes CLUES, an automated high-quality data screening process suitable for collaborative fine-tuning of large language models without sharing private data.
Cluster-Learngene: Inheriting Adaptive Clusters for Vision Transformers
Qiufeng Wang (Southeast University), Xin Geng (Southeast University)
ClassificationRecognitionTransformerImage
🎯 What it does: This paper proposes a method that performs adaptive clustering of the attention heads and FFN of large pre-trained visual Transformers, using them as learning genes for inherited initialization, thereby quickly obtaining sub-models under different scales and resource constraints.
Cluster-wise Graph Transformer with Dual-granularity Kernelized Attention
Siyuan Huang (Shanghai Jiao Tong University), Zhouhan Lin (Shanghai Jiao Tong University)
ClassificationGraph Neural NetworkTransformerGraphBiomedical Data
🎯 What it does: A graph Transformer architecture based on node clustering, Cluster-GT, is proposed, using node sets as tokens and achieving dual-granularity attention transfer between node sets and nodes through Node-to-Cluster Attention (N2C-Attn).
Clustering in Causal Attention Masking
Nikita Karagodin (Massachusetts Institute of Technology), Philippe Rigollet (Massachusetts Institute of Technology)
TransformerOrdinary Differential Equation
🎯 What it does: This paper modifies the dynamical model of Transformer self-attention, specifically studying self-attention under causal masking (where only preceding tokens are visible), and treats it as an interacting particle system on a sphere, analyzing the final clustering behavior of the particles.
Clustering then Propagation: Select Better Anchors for Knowledge Graph Embedding
KE LIANG, Xinwang Liu (National University of Defense Technology)
Computational EfficiencyKnowledge DistillationGraph Neural NetworkGraphBenchmark
🎯 What it does: A relationship clustering-based anchor point selection strategy, RecPiece, is proposed to enhance the efficiency and effectiveness of knowledge graph embedding.