NeurIPS 2024 Papers — Page 5
Conference on Neural Information Processing Systems · 4035 papers
BAdam: A Memory Efficient Full Parameter Optimization Method for Large Language Models
Qijun Luo (Chinese University of Hong Kong), Xiao Li (Chinese University of Hong Kong)
OptimizationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A BAdam method based on the Block Coordinate Descent (BCD) framework and Adam update rule is proposed for full parameter fine-tuning of large language models under memory-constrained conditions.
BAKU: An Efficient Transformer for Multi-Task Policy Learning
Siddhant Haldar (New York University), Lerrel Pinto (New York University)
Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerMultimodality
🎯 What it does: This work presents BAKU, a concise Transformer architecture for multi-task robotic policy learning, significantly improving multi-task learning efficiency.
Balancing Context Length and Mixing Times for Reinforcement Learning at Scale
Matthew Riemer (IBM Research), Sarath Chandar (Mila)
TransformerReinforcement LearningTabular
🎯 What it does: This study investigates the trade-off between the context length of policies and the mixing time in large, partially observable reinforcement learning environments, providing new theoretical bounds that are subsequently validated through experiments.
BAM! Just Like That: Simple and Efficient Parameter Upcycling for Mixture of Experts
Qizhen Zhang (University of Oxford), Acyr Locatelli (Cohere)
TransformerLarge Language ModelMixture of ExpertsText
🎯 What it does: Transfer the pre-trained dense language model (including FFN and attention parameters) to the Mixture of Experts (MoE) framework to construct the BMA model;
BAN: Detecting Backdoors Activated by Adversarial Neuron Noise
Xiaoyun Xu (Radboud University), Stjepan Picek (Radboud University)
Adversarial AttackSupervised Fine-TuningImage
🎯 What it does: This paper proposes a method using Bait Adversarial Neuron Noise (BAN) to detect and defend against backdoor attacks in deep learning models, combining feature space masking to separate positive and negative features.
Banded Square Root Matrix Factorization for Differentially Private Model Training
Nikita Kalinin (Institute of Science and Technology), Christoph H. Lampert
OptimizationSafty and PrivacyComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: A new Banded Square Root Matrix Factorization (BSR) method is proposed for differential privacy SGD training, which can efficiently perform matrix factorization on large-scale problems.
Bandit-Feedback Online Multiclass Classification: Variants and Tradeoffs
Yuval Filmus (Technion Institute of Technology), Shay Moran (Google)
ClassificationOptimization
🎯 What it does: This paper studies the impact of bandit feedback and full information feedback, adaptive adversaries, and randomized learners on the error bounds in adversarial online multi-class learning.
Bandits with Abstention under Expert Advice
Stephen Pasteris (Alan Turing Institute), Mark Herbster (University College London)
Recommendation SystemOptimizationReinforcement Learning from Human FeedbackReinforcement LearningGraph
🎯 What it does: A new Confidence-Based Abstraction (CBA) algorithm is proposed to address the bandwidth feedback prediction problem with zero-reward abandonment actions.
Bandits with Preference Feedback: A Stackelberg Game Perspective
Barna Pásztor (ETH Zurich), Andreas Krause (ETH Zurich)
Recommendation SystemOptimizationReinforcement LearningTabular
🎯 What it does: This paper proposes a new algorithm MAXMINLCB for uncertain optimization under preference feedback. The algorithm selects action pairs by constructing a zero-sum Stackelberg game, thus achieving a balance between exploration and exploitation.
Bandits with Ranking Feedback
Davide Maran (Politecnico di Milano), Marcello Restelli (Politecnico di Milano)
Recommendation SystemReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper proposes a new variant of the multi-armed bandit, called the bandit with ranking feedback. Unlike traditional bandits, this variant provides feedback that allows learners to rank the arms based on previous pulls without needing to quantify performance differences.
Barely Random Algorithms and Collective Metrical Task Systems
Romain Cosson (Inria), Laurent Massoulié (Inria)
Optimization
🎯 What it does: It is proven that any fully random metric task system (MTS) algorithm can be transformed into a 'quasi-random' algorithm that uses only O(log n) bits of randomness, while maintaining a competitive ratio with at most a factor of 2 gain.
Base of RoPE Bounds Context Length
Mingyu Xu (Baichuan Inc), weipeng chen
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This study proves that the cardinality of RoPE has an absolute lower limit on the context length that large language models can handle, revealing the property of long-distance attention capability decaying with relative distance, and validating this theory during both the Fine-Tuning and Pre-Training phases.
Batched Energy-Entropy acquisition for Bayesian Optimization
Felix Teufel (Novo Nordisk), Jesper Ferkinghoff-Borg (Novo Nordisk)
OptimizationTabular
🎯 What it does: A statistical physics-based energy-entropy (BEEBO) acquisition function is proposed for high-dimensional large-batch Bayesian optimization, supporting Gaussian processes and heteroscedastic noise, and can be directly optimized for batch points using gradient methods.
Bayes-optimal learning of an extensive-width neural network from quadratically many samples
Antoine Maillard (ETH Zurich), Lenka Zdeborova
Tabular
🎯 What it does: This study investigates the Bayes optimal learning performance of single hidden layer quadratic activation neural networks in the case where both the input dimension and network width tend to infinity, and the sample size is a power of the dimension, providing a closed-form expression for the test error.
Bayesian Adaptive Calibration and Optimal Design
Rafael Oliveira (CSIRO Data61), Edwin V. Bonilla (CSIRO Data61)
OptimizationTabular
🎯 What it does: BACON is proposed, a Bayesian Adaptive Calibration and Optimal Design method that uses information gain to guide simulation design for efficient calibration of computational models.
Bayesian Domain Adaptation with Gaussian Mixture Domain-Indexing
Yanfang Ling (Sun Yat-sen University), Shangsong Liang (Sun Yat-sen University)
Domain AdaptationTabular
🎯 What it does: A domain index inference algorithm GMDI based on Gaussian Mixture Models is proposed for domain adaptation in the absence of domain index information.
Bayesian Nonparametrics Meets Data-Driven Distributionally Robust Optimization
Nicola Bariletto (University of Texas at Austin), Nhat Ho (University of Texas at Austin)
OptimizationTabular
🎯 What it does: A data-driven distributionally robust optimization criterion is proposed, combining Bayesian nonparametrics (Dirichlet process) with smooth ambiguity aversion theory, and a computable approximate form is provided.
Bayesian Online Natural Gradient (BONG)
Matt Jones (University of Colorado), Kevin Patrick Murphy
OptimizationConvolutional Neural NetworkImage
🎯 What it does: An online Bayesian natural gradient (BONG) method is proposed, which updates the expected likelihood using a single-step natural gradient, eliminating KL regularization, and is suitable for sequential inference in neural networks.
Bayesian Optimisation with Unknown Hyperparameters: Regret Bounds Logarithmically Closer to Optimal
Juliusz Ziomek (University of Oxford), Michael A Osborne
OptimizationHyperparameter SearchTabular
🎯 What it does: To address the challenge of unknown length scales in Bayesian optimization, a length scale balancing algorithm LB-GP-UCB is proposed, improving upon the previous A-GP-UCB and providing a tighter cumulative regret upper bound.
Bayesian Optimization of Functions over Node Subsets in Graphs
Huidong Liang (University of Oxford), Xiaowen Dong (University of Oxford)
OptimizationGraph Neural NetworkGraph
🎯 What it does: A framework called GraphComBO based on Bayesian optimization is proposed to find the optimal subset of nodes in a graph for a black-box set function.
Bayesian Strategic Classification
Lee Cohen (Stanford University), Juba Ziani (Georgia Institute of Technology)
ClassificationOptimization
🎯 What it does: Under the traditional strategic classification assumption that agents have complete knowledge of the learner's decision-maker, this paper introduces a Bayesian perspective: agents only know the prior distribution of the classifier used by the learner, and the learner can actively disclose partial information (i.e., a subset containing the true classifier) to influence the agent's strategy and improve its classification accuracy.
Bayesian-guided Label Mapping for Visual Reprogramming
Chengyi Cai (University of Melbourne), Feng Liu (Singapore University of Technology and Design)
ClassificationRecognitionConvolutional Neural NetworkVision Language ModelImage
🎯 What it does: A probability label mapping method based on Bayesian inference is proposed for the output label mapping of visual reprogramming.
Be Confident in What You Know: Bayesian Parameter Efficient Fine-Tuning of Vision Foundation Models
Deep Shankar Pandey (Rochester Institute of Technology), Qi Yu (Rochester Institute of Technology)
ClassificationOptimizationTransformerSupervised Fine-TuningImage
🎯 What it does: This study investigates the issue of insufficient confidence in parameter-efficient fine-tuning of the Transformer base model under few-shot tasks and proposes the Bayesian-PEFT framework to enhance model accuracy and confidence calibration.
Be like a Goldfish, Don't Memorize! Mitigating Memorization in Generative LLMs
Abhimanyu Hans (University of Maryland), Tom Goldstein (University of Maryland)
GenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: A training objective called Goldfish Loss is proposed, which randomly masks a portion of tokens in each training sample, preventing the model from fully memorizing the training data and thereby reducing the probability of verbatim copying during generation.
Beating Adversarial Low-Rank MDPs with Unknown Transition and Bandit Feedback
Haolin Liu (University of Virginia), Julian Zimmert (Google Research)
OptimizationReinforcement Learning
🎯 What it does: Minimizing returns in low-rank MDPs with unknown transitions and adversarial losses, and proposing new algorithms and improved theoretical bounds.
BECAUSE: Bilinear Causal Representation for Generalizable Offline Model-based Reinforcement Learning
Haohong Lin (Carnegie Mellon University), Ding Zhao (University of Chicago)
Robotic IntelligenceReinforcement Learning
🎯 What it does: This study investigates how to address the target mismatch problem in offline model-based reinforcement learning, proposing the BECAUSE framework, which utilizes causal representation learning and low-rank bilinear structures to improve the alignment between model learning and policy planning.
BehaviorGPT: Smart Agent Simulation for Autonomous Driving with Next-Patch Prediction
Zikang Zhou (City University of Hong Kong), Chun Jason Xue (Mohamed bin Zayed University of Artificial Intelligence)
Autonomous DrivingTransformerAgentic AITime Series
🎯 What it does: A fully autoregressive Transformer model called BehaviorGPT is proposed for multi-agent traffic simulation, enhancing long-term interactive reasoning through the Next Patch Prediction Paradigm (NP3).
Belief-State Query Policies for User-Aligned POMDPs
Daniel Richard Bramblett (Arizona State University), Siddharth Srivastava (Arizona State University)
OptimizationReinforcement Learning from Human FeedbackReinforcement LearningAgentic AI
🎯 What it does: A strategy framework based on Belief State Query (BSQ) is proposed to express user constraints and preferences regarding agent behavior in partially observable POMDPs, and to align user task objectives through parameterized BSQ strategies. The theoretical analysis of this framework is provided along with a partition-based optimization algorithm called Partition Refinement Search (PRS).
BELM: Bidirectional Explicit Linear Multi-step Sampler for Exact Inversion in Diffusion Models
Fangyikang Wang (Zhejiang University), Chen Li (Tencent)
GenerationData SynthesisDiffusion modelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: This paper studies the precise inversion problem of diffusion model sampling, proposing a general bidirectional explicit linear multi-step sampling framework (BELM), and based on this, designs an optimal sampler (O-BELM) to achieve training-independent precise inversion and high-quality sampling.
BendVLM: Test-Time Debiasing of Vision-Language Embeddings
Walter Gerych (Massachusetts Institute of Technology), Marzyeh Ghassemi (Massachusetts Institute of Technology)
ClassificationRecognitionData-Centric LearningTransformerLarge Language ModelVision Language ModelImageMultimodality
🎯 What it does: This paper proposes a testing-time, non-fine-tuning VLM debiasing method called BEND-VLM, which removes gender/racial bias in visual language models by orthogonalizing the local attribute subspace of each query and equalizing reference images.
Benign overfitting in leaky ReLU networks with moderate input dimension
Kedar Karhadkar (University of California Los Angeles), Deanna Needell (University of California Los Angeles)
ClassificationOptimizationTabular
🎯 What it does: This study investigates the overfitting behavior of a two-layer leaky ReLU network when trained with hinge loss on a binary classification task with label noise, and provides theoretical conditions under which beneficial (benign) or harmful (non-benign) overfitting occurs.
BERTs are Generative In-Context Learners
David Samuel (University of Oslo)
GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: Using a simple reasoning method, the author directly transforms the pre-trained masked language model DeBERTa into a generative and ranking model, demonstrating that masked models can also achieve in-context learning.
Better by default: Strong pre-tuned MLPs and boosted trees on tabular data
David Holzmüller (Inria), Ingo Steinwart
Hyperparameter SearchMeta LearningTabularBenchmark
🎯 What it does: An improved version of the multilayer perceptron, RealMLP, and tuned default parameters for GBDT are proposed, and their performance is evaluated on large-scale meta-training/meta-testing benchmarks.
BetterDepth: Plug-and-Play Diffusion Refiner for Zero-Shot Monocular Depth Estimation
Xiang Zhang (ETH Zürich), Christopher Schroers (ETH Zürich)
Depth EstimationDiffusion modelImage
🎯 What it does: By combining a pre-trained zero-shot monocular depth estimation model with a conditional diffusion model, BetterDepth is constructed to maintain global geometric priors while achieving high-quality detail refinement.
Beware of Road Markings: A New Adversarial Patch Attack to Monocular Depth Estimation
Hangcheng Liu (Nanyang Technological University), Tianwei Zhang (Nanyang Technological University)
Depth EstimationAutonomous DrivingAdversarial AttackConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes an attack method (AdvRM) that detaches adversarial patches from the physical location of obstacles, disguises them as ordinary road signs on the road, and misleads monocular depth estimation models.
Beyond Accuracy: Ensuring Correct Predictions With Correct Rationales
Tang Li (University of Delaware), Xi Peng (University of Delaware)
RetrievalOptimizationExplainability and InterpretabilityTransformerLarge Language ModelPrompt EngineeringContrastive LearningImage
🎯 What it does: A dual-positive prediction framework is proposed, which constructs a reason dataset with a hierarchical structure and designs a reason-based unsupervised optimization method to enhance the model's prediction accuracy and reason interpretability.
Beyond Accuracy: Tracking more like Human via Visual Search
Dailing Zhang (University of Chinese Academy of Sciences), Kaiqi Huang (Chinese Academy of Sciences)
Object TrackingConvolutional Neural NetworkTransformerVideoBenchmark
🎯 What it does: A target tracking algorithm called CPDTrack is proposed, which is based on the human visual search mechanism and can continuously locate targets and quickly recover in complex scenes.
Beyond Concept Bottleneck Models: How to Make Black Boxes Intervenable?
Sonia Laguna (ETH Zurich), Julia E Vogt
Supervised Fine-TuningVision Language ModelImageTabularBiomedical Data
🎯 What it does: Implement instance-level concept intervention on existing black-box neural networks and enhance intervention effectiveness through fine-tuning on a small validation set.
Beyond Efficiency: Molecular Data Pruning for Enhanced Generalization
Dingshuo Chen (Chinese Academy of Sciences), Liang Wang (Chinese Academy of Sciences)
Data-Centric LearningDrug DiscoveryGraph Neural NetworkSupervised Fine-TuningGraphBiomedical Data
🎯 What it does: A molecular data screening framework called MolPeg is proposed, which can significantly improve the generalization performance of downstream tasks and reduce training costs through dynamic pruning while only using pre-trained models.
Beyond Euclidean: Dual-Space Representation Learning for Weakly Supervised Video Violence Detection
Jiaxu Leng (Chongqing University of Posts and Telecommunications), Xinbo Gao (Chongqing University of Posts and Telecommunications)
RecognitionRepresentation LearningGraph Neural NetworkVideoMultimodalityAudio
🎯 What it does: A Dual-Space Representation Learning (DSRL) framework is proposed, which integrates Euclidean space and hyperbolic space to enhance the recognition ability of weakly supervised video violence detection for visually similar and ambiguous events.
Beyond Optimism: Exploration With Partially Observable Rewards
Simone Parisi (University of Alberta), Michael Bowling (University of Alberta)
OptimizationReinforcement LearningTabular
🎯 What it does: A novel exploration strategy driven by Successor Representation (SR) that does not rely on optimistic assumptions is proposed and its effectiveness is validated within the framework of Monitored MDP with partially observable rewards.
Beyond Primal-Dual Methods in Bandits with Stochastic and Adversarial Constraints
Martino Bernasconi (Bocconi University), Federico Fusco (Sapienza University of Rome)
OptimizationReinforcement Learning from Human FeedbackReinforcement Learning
🎯 What it does: This paper proposes an optimal dual-world algorithm for the Bandit problem with general long-term constraints, achieving optimal returns while satisfying the constraints.
Beyond Redundancy: Information-aware Unsupervised Multiplex Graph Structure Learning
Zhixiang Shen (University of Electronic Science and Technology of China), zhao kang
ClassificationRepresentation LearningGraph Neural NetworkContrastive LearningGraph
🎯 What it does: In unsupervised multi-layer graph structure learning, this paper proposes the InfoMGF framework, which first refines the graph structure for each view to remove irrelevant noise, and then merges all views to generate a global graph that contains both shared and unique task-related information for learning node representations.
Beyond Single Stationary Policies: Meta-Task Players as Naturally Superior Collaborators
Haoming Wang (Xi'an Jiaotong University), Zhongmin Cai (University of Notre Dame)
Robotic IntelligenceMeta LearningReinforcement Learning from Human FeedbackReinforcement LearningVideo
🎯 What it does: A human-machine collaboration framework named Collaborative Bayesian Policy Reuse (CBPR) has been designed and implemented, which can dynamically select the most suitable meta-task-specific strategy based on the non-stationary behavior of human partners, thereby achieving more efficient collaboration.
Beyond Slow Signs in High-fidelity Model Extraction
Hanna Foerster (University of Cambridge), Jamie Hayes (Google DeepMind)
Computational EfficiencyAdversarial AttackConvolutional Neural NetworkImage
🎯 What it does: This paper unifies previous cryptanalysis extraction methods to construct a complete codebase for end-to-end extraction of neural network parameters (weights and symbols), and conducts a systematic evaluation on standard benchmarks (MNIST, CIFAR-10), significantly improving extraction speed and robustness.
Beyond task diversity: provable representation transfer for sequential multitask linear bandits
Thang Duong (University of Arizona), Chicheng Zhang (University of Arizona)
OptimizationRepresentation LearningReinforcement LearningSequential
🎯 What it does: The study investigates the sequential multi-task linear bandit problem under the assumption of no task diversity and proposes the BOSS algorithm, which can learn and transfer low-rank representations.
Beyond the Doors of Perception: Vision Transformers Represent Relations Between Objects
Michael A. Lepori (Brown University), Ellie Pavlick (Brown University)
Explainability and InterpretabilityRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: A mechanism interpretability analysis of the Vision Transformer reveals that it employs a two-stage processing flow in same-different tasks, extracting object features during the perception stage and comparing objects during the relationship stage.
Bias Amplification in Language Model Evolution: An Iterated Learning Perspective
Yi Ren (University of British Columbia), Danica J. Sutherland (University of British Columbia)
TransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: This study explores the similarities between large language models (LLMs) and human cultural evolution during the process of self-iterative improvement. It utilizes a Bayesian Iterative Learning (IL) framework to explain the sampling and learning behaviors of LLMs, and experimentally verifies the bias amplification mechanism and the regulatory effects of the interaction phase.
Bias Detection via Signaling
Yiling Chen (Harvard University), Itai Shapira (Harvard University)
🎯 What it does: This paper proposes a signal scheme based on information design to detect whether agents update their beliefs according to Bayesian rules when faced with new evidence, or whether there is a deviation from the original prior.
Bias in Motion: Theoretical Insights into the Dynamics of Bias in SGD Training
Anchit Jain (University of Cambridge), Stefano Sarao Mannelli (University College London)
ClassificationOptimizationImageOrdinary Differential Equation
🎯 What it does: This study investigates the evolution of bias during the training of high-dimensional linear classifiers using stochastic gradient descent (SGD) under a teacher-mixed model, providing a complete analytical closed-form solution that reveals the non-monotonic changes of bias in three stages during the training process.
Bidirectional Recurrence for Cardiac Motion Tracking with Gaussian Process Latent Coding
Jiewen Yang (Hong Kong University of Science and Technology), Xiaomeng Li (Hong Kong University of Science and Technology)
Object TrackingSegmentationDomain AdaptationComputational EfficiencyRecurrent Neural NetworkTransformerAuto EncoderImageBiomedical DataMagnetic Resonance Imaging
🎯 What it does: A bidirectional recursive framework GPTrack based on Gaussian process latent encoding is proposed for accurate tracking of cardiac motion.
BiDM: Pushing the Limit of Quantization for Diffusion Models
Xingyu Zheng (Beihang University), Haotong Qin (ETH Zürich)
GenerationCompressionKnowledge DistillationDiffusion modelImage
🎯 What it does: This paper proposes BiDM, a fully binarized diffusion model (DM) compression method that addresses the high computational and storage overhead of DMs in resource-constrained scenarios.
Bigger, Regularized, Optimistic: scaling for compute and sample efficient continuous control
Michal Nauman (University of Warsaw), Marek Cygan (University of Warsaw)
Robotic IntelligenceReinforcement LearningBenchmark
🎯 What it does: The BRO (Bigger, Regularized, Optimistic) algorithm is proposed and implemented, significantly improving sample efficiency in continuous control tasks through a large-scale regularized critic network, BroNet structure, optimistic exploration, and non-pessimistic quantized Q approximation.
Bileve: Securing Text Provenance in Large Language Models Against Spoofing with Bi-level Signature
Tong Zhou (Northeastern University), Shaolei Ren (UC Riverside)
GenerationTransformerLarge Language ModelText
🎯 What it does: A dual-layer signature scheme called Bileve is proposed to achieve both robust watermarking and verifiable content integrity signatures in text generated by large language models, thereby preventing forgery attacks and providing detectable tampering evidence.
Binarized Diffusion Model for Image Super-Resolution
Zheng Chen (Shanghai Jiao Tong University), Yulun Zhang (Shanghai Jiao Tong University)
RestorationSuper ResolutionConvolutional Neural NetworkDiffusion modelImage
🎯 What it does: A binary diffusion model (BI-DiffSR) is proposed for image super-resolution tasks, aiming to significantly reduce memory and computational consumption.
Binary Search with Distributional Predictions
Michael Dinitz (Johns Hopkins University), Sergei Vassilvitskii (Google Research)
Time Series
🎯 What it does: This study investigates an algorithm that uses distributed predictions in binary search, proposing a robust algorithm that maintains low query complexity even when the prediction distribution is imprecise.
Binding in hippocampal-entorhinal circuits enables compositionality in cognitive maps
Christopher Kymn, Bruno Olshausen
🎯 What it does: A normative model based on the Residue Number System (RNS) and binding vectors is proposed to explain the role of the hippocampus-olfactory cortex in spatial representation and path integration, further achieving efficient and error-correcting spatial encoding through a modular attractor network.
Binocular-Guided 3D Gaussian Splatting with View Consistency for Sparse View Synthesis
Liang Han (Tsinghua University), Zhizhong Han (Wayne State University)
Data SynthesisDepth EstimationNeural Radiance FieldGaussian SplattingPoint Cloud
🎯 What it does: A sparse view synthesis method based on 3D Gaussian Splatting is proposed, utilizing self-supervised binocular disparity constraints without the need for external priors.
Biologically Inspired Learning Model for Instructed Vision
Roy Abel (Weizmann Institute of Science), Shimon Ullman (Weizmann Institute of Science)
ClassificationRecognitionImage
🎯 What it does: A biologically interpretable visual learning model that combines the upward (BU) and downward (TD) pathways is proposed, with TD feedback used simultaneously for visual guidance and learning.
BiScope: AI-generated Text Detection by Checking Memorization of Preceding Tokens
Hanxi Guo (Purdue University), Xiangyu Zhang (Purdue University)
ClassificationGenerationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: The BISCOPE method is proposed, which captures the forward and backward token information in the logits of LLM outputs through bidirectional cross-entropy loss for detecting AI-generated text.
Bisimulation Metrics are Optimal Transport Distances, and Can be Computed Efficiently
Sergio Calo (Universitat Pompeu Fabra), Javier Segovia-Aguas (Universitat Pompeu Fabra)
OptimizationComputational EfficiencyReinforcement LearningSequential
🎯 What it does: This paper studies the distance metric between finite Markov chains, proving that the bisimulation metric is equivalent to the optimal transport distance of bicausal, and provides a linear programming formulation based on occupancy coupling. Subsequently, the authors designed a new algorithm—Sinkhorn Value Iteration (SVI), which combines entropy-regularized Sinkhorn iteration and value iteration to efficiently solve this LP, thereby obtaining the optimal transport distance between Markov chains.
bit2bit: 1-bit quanta video reconstruction via self-supervised photon prediction
Yehe Liu (Case Western Reserve University), Michael W. Jenkins
RestorationConvolutional Neural NetworkSupervised Fine-TuningVideo
🎯 What it does: This paper proposes a self-supervised 1-bit quantum image reconstruction method called bit2bit, which directly recovers high spatiotemporal resolution videos from sparse binary SPAD data.
BitDelta: Your Fine-Tune May Only Be Worth One Bit
James Liu (Massachusetts Institute of Technology), Tianle Cai (Princeton University)
CompressionComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes the BitDelta method, which performs 1-bit binary compression on the weight increments generated by LLM fine-tuning, enabling multiple models to share the same base model.
BitsFusion: 1.99 bits Weight Quantization of Diffusion Model
Yang Sui (Snap Inc.), Jian Ren (Snap Inc.)
GenerationCompressionKnowledge DistillationDiffusion modelImageText
🎯 What it does: Quantized the weights of the UNet in Stable Diffusion v1.5 to 1.99 bits, compressing it to approximately 219 MB, while achieving significant model compression without compromising and even improving image generation quality.
Black-Box Forgetting
Yusuke Kuwana (Tokyo University of Science), Go Irie (Tokyo University of Science)
OptimizationPrompt EngineeringVision Language ModelImage
🎯 What it does: A framework for selective forgetting (Black-Box Forgetting) on black-box pre-trained models is proposed, which allows the model to forget specified categories without changing the model parameters.
BLAST: Block-Level Adaptive Structured Matrices for Efficient Deep Neural Network Inference
Changwoo Lee (University of Michigan), Hun-Seok Kim (University of Michigan)
CompressionOptimizationComputational EfficiencyTransformerLarge Language ModelImageText
🎯 What it does: This paper proposes a Block-Level Adaptive Structured Matrix (BLAST) to replace the dense weights in the linear layers of large models, achieving inference acceleration.
Blind Image Restoration via Fast Diffusion Inversion
Hamadi Chihaoui (University of Bern), Paolo Favaro (University of Bern)
RestorationDiffusion modelImage
🎯 What it does: A training-free blind image restoration method called BIRD is proposed, which utilizes a pre-trained DDIM model to achieve image restoration by optimizing the initial noise and degradation parameters, ensuring that the restored image always lies on the data manifold.
BLoB: Bayesian Low-Rank Adaptation by Backpropagation for Large Language Models
Yibin Wang (Rutgers University), Hao Wang (Rutgers University)
TransformerLarge Language ModelSupervised Fine-TuningText
🎯 What it does: This paper proposes a Bayesian low-rank adaptation framework BLoB that jointly estimates the weight mean and covariance through backpropagation during the fine-tuning process of LoRA-parameterized LLMs.
Block Sparse Bayesian Learning: A Diversified Scheme
Yanhao Zhang (Beihang University), Yong Xia (Beihang University)
OptimizationImageAudio
🎯 What it does: This paper proposes a Bayesian learning framework based on dispersed block sparse priors (DivSBL), which can adaptively estimate block size and position, significantly improving the recovery accuracy of block sparse signals.
Block Transformer: Global-to-Local Language Modeling for Fast Inference
Namgyu Ho (KAIST), Se-Young Yun (KAIST)
TransformerLarge Language ModelText
🎯 What it does: A block-based Transformer structure is designed, divided into block-level global attention and word-level local attention, to accelerate autoregressive language model inference.
BMRS: Bayesian Model Reduction for Structured Pruning
Dustin Wright (University of Copenhagen), Raghavendra Selvan (University of Copenhagen)
CompressionComputational EfficiencyConvolutional Neural NetworkTransformerImage
🎯 What it does: This paper proposes a Bayesian Model Reduced Structured Pruning method (BMRS) that achieves threshold-free network structure pruning, further improving computational and energy efficiency.
BOLD: Boolean Logic Deep Learning
Van Minh Nguyen (Huawei), Ba-Hien Tran (Huawei)
ClassificationSegmentationSuper ResolutionImage
🎯 What it does: A Boolean logic deep learning framework based on Boolean mutation is proposed, enabling training and inference within the Boolean domain.
BoNBoN Alignment for Large Language Models and the Sweetness of Best-of-n Sampling
Lin Gui (University of Chicago), Victor Veitch (University of Chicago)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningContrastive LearningText
🎯 What it does: This paper analyzes the relationship between Best n-sampling (BoN) and alignment methods, proving that BoN is nearly optimal in the trade-off between win rate and KL divergence, and proposes the BoNBoN alignment method to approximate the BoN distribution without significantly increasing the discreteness.
BoostAdapter: Improving Vision-Language Test-Time Adaptation via Regional Bootstrapping
Taolin Zhang (Tsinghua University), Shu-Tao Xia (Tsinghua University)
Domain AdaptationTransformerVision Language ModelImageMultimodality
🎯 What it does: This paper proposes BoostAdapter, which improves the adaptive capability of zero-training visual language models during testing by incorporating self-enhanced low-entropy samples into the cache.
Boosted Conformal Prediction Intervals
Ran Xie (Stanford University), Emmanuel Candes
Tabular
🎯 What it does: This paper proposes a post-training Boosted Conformal Procedure that improves the existing conformity score through gradient boosting, thereby obtaining conformal prediction intervals that meet specified coverage conditions or have shorter interval lengths.
Boosting Alignment for Post-Unlearning Text-to-Image Generative Models
Myeongseob Ko (Virginia Tech), Ruoxi Jia (Virginia Tech)
GenerationData SynthesisLarge Language ModelDiffusion modelImageText
🎯 What it does: A machine forgetting framework that combines gradient restriction and data diversification for text-to-image generation models has been designed and implemented, capable of maintaining semantic alignment between images and text while removing specified concepts (such as nudity and artistic styles).
Boosting Generalization in Parametric PDE Neural Solvers through Adaptive Conditioning
Armand Kassaï Koupaï (Sorbonne Université), Patrick Gallinari (Criteo AI Lab)
Time SeriesPhysics RelatedOrdinary Differential Equation
🎯 What it does: The GEPS framework is proposed, which achieves rapid generalization of neural PDE solvers in multi-parameter environments through low-rank first-order meta-learning and adaptive context parameters.
Boosting Graph Pooling with Persistent Homology
Chaolong Ying (Chinese University of Hong Kong), Tianshu Yu (Chinese University of Hong Kong)
ClassificationRepresentation LearningGraph Neural NetworkGraph
🎯 What it does: A Topology-Invariant Pooling method has been developed to inject persistent homology information into the graph pooling layer, enhancing the expressive power of graph networks during the pooling process.
Boosting Sample Efficiency and Generalization in Multi-agent Reinforcement Learning via Equivariance
Joshua McClellan (Johns Hopkins University Applied Physics Laboratory), Pratap Tokekar (University of Maryland)
Graph Neural NetworkReinforcement LearningGraph
🎯 What it does: This paper proposes an exploration-enhanced equivariant graph neural network (E2GN2) to address the early exploration bias problem in traditional equivariant graph neural networks (EGNN) within multi-agent reinforcement learning, demonstrating its ability to improve sample efficiency and generalization in multi-agent environments.
Boosting Semi-Supervised Scene Text Recognition via Viewing and Summarizing
Yadong Qu (University of Science and Technology of China), Yongdong Zhang (University of Science and Technology of China)
RecognitionTransformerImage
🎯 What it does: A semi-supervised scene text recognition framework based on 'watching + summarizing' is proposed, which significantly improves the recognition performance of artistic and distorted text through online generation strategies and character unidirectional alignment loss.
Boosting Text-to-Video Generative Model with MLLMs Feedback
Xun Wu (Microsoft Research Asia), Furu Wei (Microsoft Research Asia)
GenerationData SynthesisReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningVideoTextMultimodality
🎯 What it does: Construct a large video preference dataset VIDEOPREFER and train a general video reward model VIDEORM to improve the quality of text-to-video generation models and their consistency with text prompts.
Boosting the Potential of Large Language Models with an Intelligent Information Assistant
Yujia Zhou (Tsinghua University), Zhicheng Dou (Renmin University of China)
GenerationRetrievalReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningTextRetrieval-Augmented Generation
🎯 What it does: The ASSISTRAG framework is proposed, which combines a trainable intelligent information assistant with the main LLM to enhance the reasoning and accuracy of retrieval-augmented generation tasks.
Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning
Di Ming (Chongqing University of Technology), Xin Feng (Chongqing University of Technology)
Adversarial AttackTransformerImage
🎯 What it does: An Adaptive Token Tuning (ATT) attack method is proposed to enhance the transferability of adversarial attacks on Vision Transformers (ViT).
Boosting Transferability and Discriminability for Time Series Domain Adaptation
Mingyang Liu (Harbin Institute of Technology), Liqiang Nie (Harbin Institute of Technology)
Domain AdaptationKnowledge DistillationTime Series
🎯 What it does: Designed the ACON network for unsupervised domain adaptation in time series, combining multi-period frequency feature learning, time-frequency domain mutual learning, and time-frequency related subspace adversarial learning to enhance transferability and discriminability.
Boosting Vision-Language Models with Transduction
Maxime Zanella (University of Louvain), Ismail Ben Ayed (École de technologie supérieure Montréal)
ClassificationRecognitionOptimizationTransformerVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: A transductive method named TransCLIP is proposed, which enhances zero/few-shot inference performance on visual language models by leveraging the structure of unlabeled data.
Boosting Weakly Supervised Referring Image Segmentation via Progressive Comprehension
Zaiquan Yang (City University of Hong Kong), Rynson W. H. Lau (City University of Hong Kong)
Object DetectionSegmentationTransformerLarge Language ModelImage
🎯 What it does: A weakly supervised referential image segmentation method called PCNet is proposed, which achieves precise localization of target objects by gradually utilizing target-related phrases in the language description, and uses the localization results to drive a pre-trained segmenter to generate the final segmentation mask.
Bootstrapping Top-down Information for Self-modulating Slot Attention
Dongwon Kim (POSTECH), Suha Kwak (POSTECH)
Object DetectionSegmentationTransformerAuto EncoderImage
🎯 What it does: By incorporating a self-supervised top-down pathway in Slot Attention, semantic information automatically bootstrapped from Slot outputs is utilized to self-regulate Slot Attention, thereby enhancing object-centric learning performance in an unlabeled setting.
Boundary Decomposition for Nadir Objective Vector Estimation
Ruihao Zheng (Southern University of Science and Technology), Zhenkun Wang (Southern University of Science and Technology)
Optimization
🎯 What it does: A boundary decomposition-based bi-level optimization method (BDNE) is proposed for accurately estimating the nadir objective vector of multi-objective optimization problems (MOPs).
Boundary Matters: A Bi-Level Active Finetuning Method
Han Lu (Shanghai Jiao Tong University), Junchi Yan (Shanghai Jiao Tong University)
ClassificationObject DetectionSegmentationSupervised Fine-TuningImage
🎯 What it does: This paper proposes the BiLAF framework, which achieves efficient active fine-tuning through one-shot sample selection under the pre-training-fine-tuning paradigm.
Bounds for the smallest eigenvalue of the NTK for arbitrary spherical data of arbitrary dimension
Kedar Karhadkar (University of California Los Angeles), Guido Montufar
Recurrent Neural Network
🎯 What it does: This paper studies the high-probability upper and lower bounds of the minimum eigenvalue of the neural tangent kernel (NTK) at initialization for fully connected ReLU networks under arbitrary spherical distributions and dimensions, without requiring the input dimension to grow with the sample size.
BPQP: A Differentiable Convex Optimization Framework for Efficient End-to-End Learning
Jianming Pan (University of California), Jiang Bian (Microsoft Research Asia)
OptimizationComputational EfficiencyFinance Related
🎯 What it does: A differentiable convex optimization framework BPQP is proposed, which can efficiently solve the gradients of the optimization layer in end-to-end learning.
Brain-JEPA: Brain Dynamics Foundation Model with Gradient Positioning and Spatiotemporal Masking
Zijian Dong (National University of Singapore), Juan Helen Zhou (National University of Singapore)
ClassificationRepresentation LearningGraph Neural NetworkTransformerContrastive LearningTime SeriesBiomedical DataMagnetic Resonance ImagingAlzheimer's Disease
🎯 What it does: This paper proposes Brain-JEPA, a brain dynamics foundational model based on Joint-Embedding Predictive Architecture.
BrainBits: How Much of the Brain are Generative Reconstruction Methods Using?
David Mayo (Massachusetts Institute of Technology), Andrei Barbu (Massachusetts Institute of Technology)
GenerationData SynthesisDiffusion modelImageTextMagnetic Resonance Imaging
🎯 What it does: This study investigates the practical utilization of stimulus reconstruction methods for brain signals, proposing the BrainBits metric to quantify this utilization and provide random baselines and reconstruction upper limits.
Breaking Determinism: Fuzzy Modeling of Sequential Recommendation Using Discrete State Space Diffusion Model
Wenjia Xie (University of Science and Technology of China), Enhong Chen (University of Science and Technology of China)
Recommendation SystemTransformerDiffusion modelSequential
🎯 What it does: A sequence recommendation model DDSR based on discrete state space diffusion is proposed, which utilizes fuzzy sets to model user interests.
Breaking Long-Tailed Learning Bottlenecks: A Controllable Paradigm with Hypernetwork-Generated Diverse Experts
Zhe Zhao (University of Science and Technology of China), Yang Wang (University of Science and Technology of China)
ClassificationMixture of ExpertsImage
🎯 What it does: A controllable long-tail learning framework PRL is proposed, which utilizes a hypernetwork to generate diverse expert models, and dynamically adjusts the head and tail class weights through a preference vector during testing, achieving dual adaptation to distribution shifts and user needs.
Breaking Semantic Artifacts for Generalized AI-generated Image Detection
Chende Zheng (Xi'an Jiaotong University), Chao Shen (Xi'an Jiaotong University)
ClassificationObject DetectionConvolutional Neural NetworkDiffusion modelGenerative Adversarial NetworkImage
🎯 What it does: This paper proposes an AI-generated image detection method based on image patch shuffling and an end-to-end block-level classifier, aimed at eliminating the issue of insufficient cross-scene generalization caused by 'semantic artifacts'.
Breaking the curse of dimensionality in structured density estimation
Robert A. Vandermeulen (Berlin Institute for the Foundations of Learning and Data), Bryon Aragam (University of Chicago)
Graph Neural NetworkGraph
🎯 What it does: In high-dimensional nonparametric density estimation, a graph structure (Markov random field) is introduced, proposing a new graph measure—graph resilience—and providing theoretical results on controlling sample complexity on any undirected graph.
Breaking the False Sense of Security in Backdoor Defense through Re-Activation Attack
Mingli Zhu (Chinese University of Hong Kong), Baoyuan Wu (Chinese University of Hong Kong)
OptimizationAdversarial AttackImage
🎯 What it does: This study investigates the residual backdoors in post-training defense models and proposes backdoor reactivation attack schemes in white-box, black-box, and transfer scenarios.
BricksRL: A Platform for Democratizing Robotics and Reinforcement Learning Research and Education with LEGO
Sebastian Dittert (Universitat Pompeu Fabra), Gianni De Fabritiis (Universitat Pompeu Fabra)
Robotic IntelligenceReinforcement Learning
🎯 What it does: The BricksRL platform has been built, integrating LEGO hardware, Pybricks, and TorchRL, enabling the training of reinforcement learning (RL) algorithms on real LEGO robots, and providing examples of various tasks and robot models.
Bridge the Modality and Capability Gaps in Vision-Language Model Selection
Chao Yi (Nanjing University), Han-Jia Ye (Nanjing University)
ClassificationRecognitionRetrievalTransformerVision Language ModelImageTextMultimodality
🎯 What it does: The research selects the most suitable visual language model from the VLM repository using only the textual information of the target dataset, proposing to improve selection accuracy by bridging the modality gap and capability gap.