arXivSub Start free trial

ICLR 2024 Papers — Page 2

International Conference on Learning Representations · 2260 papers

Adaptive Sharpness-Aware Pruning for Robust Sparse Networks

Anna Bair (Carnegie Mellon University), Jose M. Alvarez (NVIDIA)

CompressionOptimizationConvolutional Neural NetworkImage

🎯 What it does: A three-step pruning framework named AdaSAP is proposed, which can simultaneously enhance the model's sparsity and robustness against unseen input perturbations during the pruning process.

Adaptive Stochastic Gradient Algorithm for Black-box Multi-Objective Learning

Feiyang Ye (Southern University of Science and Technology), Ivor Tsang

OptimizationGaussian SplattingTabular

🎯 What it does: An adaptive stochastic gradient algorithm (ASMG) for black-box multi-objective optimization is proposed, capable of solving Pareto optimal or Pareto stable solutions in scenarios where only function values can be queried.

Adaptive Window Pruning for Efficient Local Motion Deblurring

Haoying Li (Nanyang Technological University), Chen Change Loy (Nanyang Technological University)

RestorationComputational EfficiencyTransformerImage

🎯 What it does: A Vision Transformer based on adaptive window cropping (LMD-ViT) is proposed for local motion deblurring, focusing attention calculations only on blurred areas while preserving clear regions.

ADDP: Learning General Representations for Image Recognition and Generation with Alternating Denoising Diffusion Process

Changyao Tian (Chinese University of Hong Kong), Xizhou Zhu (Tsinghua University)

RecognitionObject DetectionSegmentationGenerationTransformerDiffusion modelImage

🎯 What it does: This paper proposes an Alternating Denoising Diffusion Process (ADDP) that learns a unified representation for both image generation and image recognition within a single model.

Addressing Loss of Plasticity and Catastrophic Forgetting in Continual Learning

Mohamed Elsayed (University of Alberta), A. Rupam Mahmood (University of Alberta)

OptimizationReinforcement LearningImage

🎯 What it does: Proposes Utility-based Perturbed Gradient Descent (UPGD), which addresses catastrophic forgetting and loss of plasticity in continual learning by utilizing weight importance gated gradients and injecting noise into unimportant weights.

Addressing Signal Delay in Deep Reinforcement Learning

Wei Wang (Western University), Dongsheng Li (Microsoft Research Asia)

Robotic IntelligenceRecurrent Neural NetworkTransformerReinforcement LearningSequential

🎯 What it does: This paper studies the performance degradation problem caused by signal delay in deep reinforcement learning, proposing a Delay Observed Markov Decision Process (DOMDP) framework, and based on this, designs improved strategies for the Critic and Actor.

AdjointDPM: Adjoint Sensitivity Method for Gradient Backpropagation of Diffusion Probabilistic Models

Jiachun Pan (National University of Singapore), Hanshu Yan (ByteDance)

GenerationOptimizationDiffusion modelImageOrdinary Differential Equation

🎯 What it does: A gradient backpropagation framework for diffusion generative models based on the adjoint sensitivity method, AdjointDPM, is proposed, which can simultaneously optimize model weights, conditional text, and noise states.

ADOPD: A Large-Scale Document Page Decomposition Dataset

Jiuxiang Gu (Adobe Research), Tong Sun (Adobe Research)

ClassificationObject DetectionSegmentationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringImageText

🎯 What it does: A large-scale document page decomposition dataset, ADOPD, has been constructed and released, and four downstream tasks (Doc2Mask, Doc2Box, Doc2Seq, Doc2Tag) have been defined and evaluated on this dataset.

Advancing Pose-Guided Image Synthesis with Progressive Conditional Diffusion Models

Fei Shen (Tencent AI Lab), Yang Wei

GenerationData SynthesisPose EstimationTransformerDiffusion modelImage

🎯 What it does: A three-stage progressive conditional diffusion model (PCDMs) is proposed for pose-guided portrait image synthesis.

Advancing the Lower Bounds: an Accelerated, Stochastic, Second-order Method with Optimal Adaptation to Inexactness

Artem Agafonov (Moscow Institute of Physics and Technology), Martin Takáč (Laboratory for Information and Inference Systems IEM STI EPFL)

OptimizationTabular

🎯 What it does: An accelerated stochastic second-order optimization method is proposed, which achieves optimal convergence speed even when the gradient and Hessian matrix are noisy (imprecise);

Adversarial Adaptive Sampling: Unify PINN and Optimal Transport for the Approximation of PDEs

Kejun Tang (Peking University), Chao Yang (Peking University)

OptimizationFlow-based ModelGenerative Adversarial NetworkPhysics Related

🎯 What it does: A new Adversarial Adaptive Sampling (AAS) framework is proposed to simultaneously optimize the solver network and sampling distribution in Physics-Informed Neural Networks (PINN), thereby reducing statistical errors and improving the accuracy of PDE approximations.

Adversarial Attacks on Fairness of Graph Neural Networks

Binchi Zhang (University of Virginia), Jundong Li (University of Virginia)

Adversarial AttackGraph Neural NetworkGraph

🎯 What it does: A new attack framework for the fairness of graph neural networks, G‑FairAttack, is proposed, which can significantly weaken the model's fairness while maintaining almost unchanged predictive performance.

Adversarial AutoMixup

Huafeng Qin (Chongqing Technology and Business University), Xinbo Gao (Chongqing University of Posts and Telecommunications)

ClassificationData SynthesisAdversarial AttackConvolutional Neural NetworkTransformerAuto EncoderGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes AdAutoMixup, an automatic mixup data augmentation framework that simultaneously trains a mixed sample generator and classifier through adversarial learning to enhance the generalization, calibration, and robustness of image classification.

Adversarial Causal Bayesian Optimization

Scott Sussex (ETH Zürich), Andreas Krause (ETH Zürich)

OptimizationReinforcement LearningTabular

🎯 What it does: This paper proposes a framework for causal Bayesian optimization with adversarial intervention (ACBO) and presents the CBO-MW algorithm, which combines multiplicative weights with self-calibrating causal models. It derives an asymptotic lossless upper bound based on graph structures and designs a scalable distributed version, D-CBO-MW, on this basis.

Adversarial Feature Map Pruning for Backdoor

Dong HUANG, Qingwen Bu (Shanghai Jiao Tong University)

ClassificationAdversarial AttackConvolutional Neural NetworkSupervised Fine-TuningImage

🎯 What it does: A backdoor defense method based on feature map pruning, FMP, is proposed, which can detect and remove backdoor feature maps in the model without relying on trigger reproduction, and restore model performance through fine-tuning.

Adversarial Imitation Learning via Boosting

Jonathan Daniel Chang, Wen Sun (Cornell University)

Adversarial AttackReinforcement LearningSequential

🎯 What it does: A new adversarial imitation learning algorithm called AILBoost is proposed, which optimizes the imitation learning process through a boosting framework.

Adversarial Supervision Makes Layout-to-Image Diffusion Models Thrive

Yumeng Li (Bosch Center for Artificial Intelligence), Anna Khoreva (Bosch Center for Artificial Intelligence)

Image TranslationSegmentationGenerationDiffusion modelGenerative Adversarial NetworkImage

🎯 What it does: This paper proposes an Adversarial Supervised Layout-to-Image Diffusion Model (ALDM) that enhances the conformity to input layouts while maintaining text controllability.

Adversarial Training on Purification (AToP): Advancing Both Robustness and Generalization

Guang Lin (Tokyo University of Agriculture and Technology), Qibin Zhao (RIKEN Center for Advanced Intelligence Project)

OptimizationAdversarial AttackGenerative Adversarial NetworkImage

🎯 What it does: By combining adversarial training with adversarial purification, the AToP method is proposed to construct a robust purification model.

Adversarial Training Should Be Cast as a Non-Zero-Sum Game

Alexander Robey (University of Pennsylvania), Volkan Cevher (École Polytechnique Fédérale de Lausanne)

OptimizationAdversarial AttackConvolutional Neural NetworkImage

🎯 What it does: A non-zero-sum bi-level optimization framework is proposed, separating the objectives of the attacker and defender, and developing the BETA attack and BETA-AT adversarial training algorithms.

AffineQuant: Affine Transformation Quantization for Large Language Models

Yuexiao Ma (ByteDance Inc), Rongrong Ji (Xiamen University)

TransformerLarge Language ModelText

🎯 What it does: A post-training quantization (AffineQuant) scheme based on reversible affine transformations is proposed for low-bit quantization of large-scale language models, significantly reducing quantization error and maintaining output consistency.

AgentBench: Evaluating LLMs as Agents

Xiao Liu (Tsinghua University), Jie Tang (Tsinghua University)

Large Language ModelAgentic AITextBenchmarkChain-of-Thought

🎯 What it does: Constructed and released AgentBench—a large-scale LLM-as-Agent evaluation benchmark covering 8 different environments (code, games, web), and systematically evaluated 29 types of LLMs (API and OSS).

AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors

Weize Chen (Tsinghua University), Jie Zhou (Tencent Inc.)

TransformerLarge Language ModelAgentic AIText

🎯 What it does: This paper proposes the AGENTVERSE multi-agent collaboration framework, which simulates four stages of human team dynamics: expert recruitment, negotiation decision-making, execution, and evaluation, enhancing the collaborative capabilities of LLM agents in multi-task scenarios.

AGILE3D: Attention Guided Interactive Multi-object 3D Segmentation

Yuanwen Yue (ETH Zurich), Theodora Kontogianni (ETH Zurich)

Object DetectionSegmentationDomain AdaptationConvolutional Neural NetworkPoint Cloud

🎯 What it does: Proposes AGILE3D, which supports interactive multi-object segmentation in 3D point clouds in a single session.

AirPhyNet: Harnessing Physics-Guided Neural Networks for Air Quality Prediction

Kethmi Hirushini Hettige (Nanyang Technological University), Jingyuan Wang (Beihang University)

Graph Neural NetworkTime SeriesPhysics RelatedOrdinary Differential Equation

🎯 What it does: A hybrid physics-data-driven model called AirPhyNet is proposed, which integrates diffusion and transport equations into a graph neural network for predicting urban PM2.5 air quality.

ALAM: Averaged Low-Precision Activation for Memory-Efficient Training of Transformer Models

Sunghyeon Woo (Seoul National University), Dongsuk Jeon (Seoul National University)

CompressionComputational EfficiencyTransformerLarge Language ModelText

🎯 What it does: A framework called ALAM based on average quantization and gradient norm variance is proposed to significantly compress the activation memory during Transformer training.

Algorithms for Caching and MTS with reduced number of predictions

Karim Ahmed Abdel Sadek, Marek Elias (Bocconi University)

Recommendation SystemOptimizationTabularTime Series

🎯 What it does: Designed to enhance the performance of caching and Metric Task System (MTS) algorithms in scenarios with limited query counts or restricted query frequencies, using action predictions, and proposed the Follower-Robust combination algorithm.

Alice Benchmarks: Connecting Real World Re-Identification with the Synthetic

Xiaoxiao Sun (Australian National University), Liang Zheng (Australian National University)

RecognitionDomain AdaptationGenerative Adversarial NetworkImageBenchmark

🎯 What it does: This paper proposes the Alice Benchmark, providing a domain adaptation benchmark for object re-identification (person/vehicle) tasks, which includes editable synthetic source data and newly collected real target data.

Align With Purpose: Optimize Desired Properties in CTC Models with a General Plug-and-Play Framework

Eliya Segev (Orcam), Tal Rosenwein (Orcam)

RecognitionOptimizationConvolutional Neural NetworkTransformerAudio

🎯 What it does: A Plug-and-Play framework (AWP) is proposed, which adds an additional loss to CTC training that can target any alignment attribute (such as low latency, minimum WER) to enhance that attribute.

AlignDiff: Aligning Diverse Human Preferences via Behavior-Customisable Diffusion Model

Zibin Dong (Tianjin University), Zhipeng Hu (Netease)

Robotic IntelligenceReinforcement Learning from Human FeedbackTransformerReinforcement LearningDiffusion modelSequential

🎯 What it does: The AlignDiff framework is proposed, which quantifies human preferences through RLHF and utilizes an attribute conditional diffusion model to achieve zero-shot behavior customization.

Aligning Relational Learning with Lipschitz Fairness

Yaning Jia (Hubei University of Technology), Soroush Vosoughi (Dartmouth College)

Recommendation SystemGraph Neural NetworkGraph

🎯 What it does: This paper proposes to constrain the model output using the Lipschitz upper bound of GNNs to enhance individual fairness based on ranking in graph neural networks.

Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps

Mingxiao Li (KU Leuven), Marie-Francine Moens (KU Leuven)

GenerationData SynthesisDiffusion modelImage

🎯 What it does: This study investigates the exposure bias problem that arises during the inference process of diffusion models (DPM) and proposes a Time-Shift sampler that alleviates this bias without the need to retrain the model.

AlpaGasus: Training a Better Alpaca with Fewer Data

Lichang Chen (University of Maryland), Hongxia Jin (Samsung Research America)

Large Language ModelSupervised Fine-TuningPrompt EngineeringText

🎯 What it does: By using an automated LLM evaluator to filter out low-quality instruction/response pairs, only high-quality data is retained, and then instruction fine-tuning is performed on a smaller dataset to obtain a better LLM;

Alt-Text with Context: Improving Accessibility for Images on Twitter

Nikita Srivatsan (Carnegie Mellon University), Taylor Berg-Kirkpatrick (University of California San Diego)

GenerationData SynthesisTransformerLarge Language ModelVision Language ModelImageTextMultimodality

🎯 What it does: A model for generating alt-text based on multimodal input (image features + tweet text) is proposed, and a Twitter dataset consisting of 371k images, tweet texts, and alt-text is publicly released.

AMAGO: Scalable In-Context Reinforcement Learning for Adaptive Agents

Jake Grigsby (University of Texas at Austin), Yuke Zhu (NVIDIA Research)

Robotic IntelligenceTransformerReinforcement LearningAgentic AISequential

🎯 What it does: AMAGO is proposed, an offline in-context reinforcement learning agent based on Transformer, capable of adaptive decision-making in long sequence contexts and supporting multi-step goal instructions.

Amortized Network Intervention to Steer the Excitatory Point Processes

Zitao Song (Chinese University of Hong Kong), Shuang Li (Chinese University of Hong Kong)

OptimizationGraph Neural NetworkReinforcement LearningGraphTime SeriesOrdinary Differential Equation

🎯 What it does: A framework for Amortizable Network Interventions (ANI) is proposed, which guides the event flow of incentivized point processes by modifying edges on dynamic networks.

AmortizedPeriod: Attention-based Amortized Inference for Periodicity Identification

Hang Yu (Ant Group), Xinzhe Wang (Ant Group)

Anomaly DetectionTransformerAuto EncoderTime SeriesFinance Related

🎯 What it does: A periodic detection method based on variational inference and attention networks, AmortizedPeriod, is proposed, which can simultaneously identify cycles, trends, noise, anomalies, and missing data.

Amortizing intractable inference in large language models

Edward J Hu, Nikolay Malkin (Mila - Quebec AI Institute, Universite de Montreal)

TransformerLarge Language ModelReinforcement LearningFlow-based ModelText

🎯 What it does: This paper proposes the use of Generative Flow Networks (GFlowNet) to fine-tune large language models, enabling them to sample intractable posterior distributions, thereby achieving complex tasks such as chain reasoning and text completion.

An Agnostic View on the Cost of Overfitting in (Kernel) Ridge Regression

Lijia Zhou (University of Chicago), Nathan Srebro (TTI-Chicago)

🎯 What it does: This study investigates the ratio of test errors between the overfitting (ridgeless) model and the optimal regularization model in noise kernel ridge regression (KRR), defining and analyzing the cost of overfitting.

An Analytical Solution to Gauss-Newton Loss for Direct Image Alignment

Sergei Solonets (Technical University of Munich), Daniel Cremers (Technical University of Munich)

OptimizationSimultaneous Localization and MappingImage

🎯 What it does: Derived and implemented a closed-form solution for the Gauss-Newton loss to dynamically control the convergence domain of direct image alignment, enhancing robustness to initialization.

An Efficient Membership Inference Attack for the Diffusion Model by Proximal Initialization

Fei Kong (University of Electronic Science and Technology of China), Kaidi Xu (Drexel University)

Computational EfficiencyAdversarial AttackDiffusion modelImageStochastic Differential EquationAudio

🎯 What it does: An efficient membership inference attack (PIA/PIAN) for diffusion models is proposed, which only requires two queries to determine whether a sample belongs to the training set.

An Efficient Tester-Learner for Halfspaces

Aravind Gollakota (Apple), Arsen Vasilyan (MIT)

Optimization

🎯 What it does: Under the framework of testable learning, an efficient half-space learning algorithm is proposed, achieving approximately optimal error rates under both Massart noise and agnostic noise models.

An Emulator for Fine-tuning Large Language Models using Small Language Models

Eric Mitchell (Stanford University), Christopher D Manning

Knowledge DistillationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText

🎯 What it does: A framework called 'Emulated Fine-Tuning' (EFT) is proposed, which combines the benchmark log probabilities of the pre-trained model with the behavior delta of the fine-tuned model to achieve an independent combination of pre-training and fine-tuning knowledge at different scales; at the same time, the framework enables model up-scaling and reward interpolation during testing.

An Extensible Framework for Open Heterogeneous Collaborative Perception

Yifan Lu (Shanghai Jiao Tong University), Siheng Chen (Shanghai Jiao Tong University)

Object DetectionAutonomous DrivingSafty and PrivacyComputational EfficiencyMultimodality

🎯 What it does: The HEAL framework is proposed to achieve open heterogeneous collaborative perception, supporting the low-cost integration of new types of sensors or models.

An Image Is Worth 1000 Lies: Transferability of Adversarial Images across Prompts on Vision-Language Models

Haochen Luo (University of Oxford), Philip Torr (University of Oxford)

Adversarial AttackPrompt EngineeringVision Language ModelImageTextMultimodality

🎯 What it does: This paper proposes the Cross-Prompt Attack (CroPA) and studies the adversarial transferability of visual-language models under different text prompts.

An improved analysis of per-sample and per-update clipping in federated learning

Bo Li (Technical University of Denmark), Sebastian U Stich

OptimizationFederated LearningSafty and PrivacyImage

🎯 What it does: This paper conducts a rigorous theoretical analysis of two common gradient clipping strategies in federated learning: sample-wise clipping and update-wise clipping, under the FedAvg algorithm, and provides convergence guarantees for arbitrary clipping thresholds.

An interpretable error correction method for enhancing code-to-code translation

Min Xue (Heidelberg University), Marla Leuther (Heidelberg University)

Explainability and InterpretabilityAI Code AssistantTransformerTextRetrieval-Augmented Generation

🎯 What it does: A kNN-based error correction method kNN‑ECD and kNN‑ECSm is proposed to improve the code translation of TransCoder‑ST and provide interpretable correction paths.

An Intuitive Multi-Frequency Feature Representation for SO(3)-Equivariant Networks

Dongwon Son (Korea Advanced Institute of Science and Technology), Beomjoon Kim (Korea Advanced Institute of Science and Technology)

ClassificationSegmentationCompressionGraph Neural NetworkPoint Cloud

🎯 What it does: A frequency-based SO(3)-equivariant feature representation (FER) is proposed, which maps 3D points to high-dimensional space and combines with Vector Neuron networks to enhance the ability to capture multi-frequency details.

An Investigation of Representation and Allocation Harms in Contrastive Learning

Subha Maity (University of Michigan), Yuekai Sun (University of Michigan)

Representation LearningContrastive LearningImageText

🎯 What it does: This study investigates how the under-representation of minority groups in Contrastive Learning (CL) leads to their feature representations clustering with similar majority groups, resulting in representation harm and allocation harm.

An LLM can Fool Itself: A Prompt-Based Adversarial Attack

Xilie Xu (National University of Singapore), Mohan Kankanhalli (National University of Singapore)

Adversarial AttackTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: A prompt-based adversarial attack called PromptAttack is proposed to evaluate the adversarial robustness of large language models.

An operator preconditioning perspective on training in physics-informed machine learning

Tim De Ryck (ETH Zurich), Emmanuel de Bezenac (ETH Zurich)

SequentialPhysics Related

🎯 What it does: This study investigates the convergence behavior of Physics-Informed Neural Networks (PINN) during the gradient descent training process, finding that training difficulties mainly stem from the condition number issues of the relevant operators, and proposes improvements through preprocessing.

An Unforgeable Publicly Verifiable Watermark for Large Language Models

Aiwei Liu (Tsinghua University), Philip S. Yu (University of Illinois at Chicago)

GenerationData SynthesisRecurrent Neural NetworkLarge Language ModelText

🎯 What it does: This paper proposes a publicly verifiable and unforgeable text watermarking algorithm for large language models, called UPV, which enables watermark detection without exposing the watermark generation key.

Analysis of Learning a Flow-based Generative Model from Limited Sample Complexity

Hugo Cui (École Polytechnique Fédérale de Lausanne), Lenka Zdeborova

GenerationData SynthesisFlow-based ModelAuto EncoderTabular

🎯 What it does: This paper studies the end-to-end analysis of learning and sampling high-dimensional Gaussian mixtures using a two-layer DAE parameterized flow model under finite sample conditions.

Analyzing and Improving Optimal-Transport-based Adversarial Networks

Jaemoo Choi (Seoul National University), Myungjoo Kang (Seoul National University)

GenerationData SynthesisOptimizationGenerative Adversarial NetworkImage

🎯 What it does: A unified framework is proposed to analyze OT-based GANs, and the UOTM is improved by scheduling the unbiased term, resulting in more stable and superior generation performance.

Analyzing and Mitigating Object Hallucination in Large Vision-Language Models

Yiyang Zhou (University of North Carolina at Chapel Hill), Huaxiu Yao (University of North Carolina at Chapel Hill)

Object DetectionGenerationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality

🎯 What it does: The LURE method is proposed, utilizing a post-processing mechanism to correct object misreporting in the text generated by large visual-language models.

Analyzing Feed-Forward Blocks in Transformers through the Lens of Attention Maps

Goro Kobayashi (Tohoku University), Kentaro Inui (MBZUAI)

TransformerLarge Language ModelText

🎯 What it does: This paper studies the impact of the feed-forward (FF) block and its surrounding residuals and layer normalization on input contextualization in Transformer layers by incorporating FF blocks into a refined analysis of attention maps, and proposes a method to decompose the nonlinear part of FF using Integrated Gradients (IG).

AnimateDiff: Animate Your Personalized Text-to-Image Diffusion Models without Specific Tuning

Yuwei Guo (Chinese University of Hong Kong), Bo Dai (Chinese University of Hong Kong)

GenerationDomain AdaptationTransformerDiffusion modelVideoText

🎯 What it does: Transform the existing personalized text-to-image diffusion model into an animation generator, supporting model-agnostic fine-tuning.

Annealing Self-Distillation Rectification Improves Adversarial Training

Yu-Yu Wu (National Taiwan University), Shang-Tse Chen (National Taiwan University)

Knowledge DistillationAdversarial AttackConvolutional Neural NetworkContrastive LearningImage

🎯 What it does: This paper proposes an ADR method that generates smooth labels through self-distillation EMA and temperature annealing during adversarial training, which can correct labels and enhance robustness without the need for a pre-trained model.

AnomalyCLIP: Object-agnostic Prompt Learning for Zero-shot Anomaly Detection

Qihang Zhou (Zhejiang University), Jiming Chen (Zhejiang University)

Anomaly DetectionTransformerPrompt EngineeringVision Language ModelImageMultimodalityBiomedical DataMagnetic Resonance ImagingComputed Tomography

🎯 What it does: AnomalyCLIP is proposed, a visual-language model (VLM) adaptation method for zero-shot anomaly detection (ZSAD);

AntGPT: Can Large Language Models Help Long-term Action Anticipation from Videos?

Qi Zhao (Brown University), Chen Sun (Brown University)

RecognitionComputational EfficiencyKnowledge DistillationTransformerLarge Language ModelPrompt EngineeringVideo

🎯 What it does: This paper proposes AntGPT, which achieves long-term action prediction by converting videos into action sequences and utilizing large language models for goal inference and temporal dynamic modeling.

AnyText: Multilingual Visual Text Generation and Editing

Yuxiang Tuo (Institute for Intelligent Computing Alibaba Group), Xuansong Xie (Institute for Intelligent Computing Alibaba Group)

GenerationDiffusion modelImageTextBenchmark

🎯 What it does: Developed AnyText, a diffusion model that supports multilingual visual text generation and editing.

Approximately Piecewise E(3) Equivariant Point Networks

Matan Atzmon (NVIDIA), Or Litany (Technion)

SegmentationPoint Cloud

🎯 What it does: The APEN (Approximate Piecewise E(3) Equivariant Point Networks) framework is proposed, which constructs an approximate piecewise E(3) equivariant point cloud network using a hierarchical and block-based approach to enhance the modeling capability for local rigid transformations.

Approximating Nash Equilibria in Normal-Form Games via Stochastic Optimization

Ian Gemp (DeepMind), Georgios Piliouras (DeepMind)

OptimizationReinforcement Learning

🎯 What it does: A loss function that can be estimated unbiasedly through Monte Carlo methods is proposed for approximating the Nash equilibrium of general normal-form games, and solved using stochastic gradient descent and the X-armed bandit method.

ArchLock: Locking DNN Transferability at the Architecture Level with a Zero-Cost Binary Predictor

Tong Zhou (Northeastern University), Xiaolin Xu (Northeastern University)

OptimizationNeural Architecture SearchImageBenchmark

🎯 What it does: This paper proposes ArchLock, which utilizes Neural Architecture Search (NAS) to reduce the transferability of DNNs at the architectural level, maintaining high performance on the source task while weakening the model's performance on potential target tasks.

Are Bert Family Good Instruction Followers? A Study on Their Potential And Limitations

yisheng xiao, Min Zhang (Soochow University)

GenerationTransformerLarge Language ModelSupervised Fine-TuningPrompt EngineeringText

🎯 What it does: Using the encoder-only structure of the BERT family, a non-autoregressive generation scheme with dynamic mixed attention and Mask-Predict was designed for instruction fine-tuning of XML-RXL, resulting in Instruct-XMLR, which was validated in evaluations across multiple tasks, languages, and translations.

Are Human-generated Demonstrations Necessary for In-context Learning?

Rui Li (University of Science and Technology of China), Jiwei Li (Zhejiang University)

GenerationData SynthesisTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought

🎯 What it does: Proposes a Self-Reflection Prompt (SEC) framework that allows large language models to first generate demonstration examples suitable for each test sample without the need for human examples, and then use these demonstrations for reasoning.

Are Models Biased on Text without Gender-related Language?

Catarina G Belém, Sameer Singh (University of California Irvine)

TransformerLarge Language ModelText

🎯 What it does: This paper studies the gender bias of large language models in sentences without gender-related vocabulary by designing the UnStereoEval evaluation framework.

Are Transformers with One Layer Self-Attention Using Low-Rank Weight Matrices Universal Approximators?

Tokio Kajitsuka (University of Tokyo), Issei Sato (University of Tokyo)

TransformerText

🎯 What it does: It is proven that a single layer of single-head, low-rank weight matrix self-attention can achieve context mapping, and its universal approximation capability is demonstrated in finite sample memory and continuous permutation equivalent functions.

ARGS: Alignment as Reward-Guided Search

Maxim Khanov (University of Wisconsin Madison), Yixuan Li (University of Wisconsin Madison)

Reinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningText

🎯 What it does: During the inference phase, a reward model is used to schedule the outputs of large language models to align with human preferences, avoiding the costly RLHF training.

ARM: Refining Multivariate Forecasting with Adaptive Temporal-Contextual Learning

Jiecheng Lu (Georgia Institute of Technology), Shihao Yang (Georgia Institute of Technology)

TransformerMixture of ExpertsTime Series

🎯 What it does: This paper studies a multivariate long-term time series forecasting method called ARM, which utilizes Adaptive Univariate Effect Learning, Random Dropping, and Multi-kernel Local Smoothing to enhance the performance of the Transformer.

ASID: Active Exploration for System Identification in Robotic Manipulation

Marius Memmel (University of Washington), Abhishek Gupta (University of Washington)

Robotic IntelligenceReinforcement LearningSequential

🎯 What it does: This paper proposes an active exploration-based system identification framework called ASID. It first learns an exploration strategy that maximizes Fisher information in simulation, then collects an exploration trajectory in the real environment to estimate system parameters (such as mass, friction, joint structure, etc.), and finally trains the target task policy on the updated high-precision simulation and directly deploys it to the real environment with zero samples.

ASMR: Activation-Sharing Multi-Resolution Coordinate Networks for Efficient Inference

Jason Chun Lok Li (University of Hong Kong), Ngai Wong (University of Hong Kong)

Data SynthesisComputational EfficiencyImageMultimodalityAudio

🎯 What it does: This study investigates low-cost implicit networks and proposes the Activation-Sharing Multi-Resolution (ASMR) coordinate network, which achieves extremely low inference costs through multi-resolution coordinate decomposition, hierarchical modulation, and activation sharing.

Assessing Uncertainty in Similarity Scoring: Performance & Fairness in Face Recognition

Jean-Rémy Conti (Idemia), Stephan Clémençon

RecognitionImage

🎯 What it does: The study quantifies uncertainty in the ROC curve of similarity scores and proposes a re-centered bootstrap method to construct confidence bands for assessing fairness in facial recognition.

Asymptotically Free Sketched Ridge Ensembles: Risks, Cross-Validation, and Tuning

Pratik Patil (University of California), Daniel LeJeune (Stanford University)

OptimizationHyperparameter SearchTextTabularBiomedical DataStochastic Differential Equation

🎯 What it does: This paper studies the consistency estimation of predictive risk using generalized cross-validation (GCV) when aggregating ridge regression models after high-dimensional random feature compression (sketch), and achieves hyperparameter tuning based on this.

At Which Training Stage Does Code Data Help LLMs Reasoning?

YINGWEI MA, Shanshan Li (National University of Defense Technology)

TransformerLarge Language ModelSupervised Fine-TuningTextChain-of-Thought

🎯 What it does: This study investigates the impact of incorporating code data at different training stages (pre-training and instruction fine-tuning) on the reasoning capabilities of large language models, evaluating its enhancement effects on general reasoning and code reasoning tasks.

Attention Satisfies: A Constraint-Satisfaction Lens on Factual Errors of Language Models

Mert Yuksekgonul (Stanford University), Besmira Nushi (Microsoft Research)

GenerationExplainability and InterpretabilityTransformerLarge Language ModelText

🎯 What it does: This study investigates the internal attention behavior of Transformer LLMs when generating factual errors and analyzes it by treating factual queries as constraint satisfaction problems.

Attention-based Iterative Decomposition for Tensor Product Representation

Taewon Park (Kyungpook National University), Minho Lee (Kyungpook National University)

Recurrent Neural NetworkTransformerText

🎯 What it does: Proposes the AID module to improve the decomposition and binding process of Tensor Product Representation (TPR), enhancing the system's generalization ability.

Attention-Guided Contrastive Role Representations for Multi-agent Reinforcement Learning

Zican Hu (Nanjing University), Zhi Wang (Nanjing University)

Recurrent Neural NetworkReinforcement LearningContrastive LearningSequential

🎯 What it does: This paper proposes the ACORM framework, which achieves behavioral heterogeneity and cooperation optimization in multi-agent reinforcement learning through role representation.

AttEXplore: Attribution for Explanation with model parameters eXploration

Zhiyu Zhu (University of Sydney), Flora D. Salim (University of New South Wales)

Explainability and InterpretabilityAdversarial AttackConvolutional Neural NetworkTransformerImage

🎯 What it does: A new interpretability method called AttEXplore is proposed, which explores model parameters during the interpretability process and utilizes transferable adversarial attacks to generate more representative samples, thereby more accurately locating the importance of input features.

AUC-CL: A Batchsize-Robust Framework for Self-Supervised Contrastive Representation Learning

Rohan Sharma (University at Buffalo), Changyou Chen (MBZUAI)

Representation LearningConvolutional Neural NetworkTransformerContrastive LearningImage

🎯 What it does: A contrastive learning framework based on AUC maximization, AUC-CL, has been designed and implemented, aiming to maintain or improve the performance of self-supervised representation learning under small batch sizes.

AuG-KD: Anchor-Based Mixup Generation for Out-of-Domain Knowledge Distillation

Zihao TANG, Kun Kuang (Zhejiang University)

Domain AdaptationKnowledge DistillationAuto EncoderGenerative Adversarial NetworkImage

🎯 What it does: Proposes the AuG-KD method to complete knowledge distillation in the absence of training data for the teacher model and in the presence of domain shift.

AUGCAL: Improving Sim2Real Adaptation by Uncertainty Calibration on Augmented Synthetic Images

Prithvijit Chattopadhyay (Georgia Tech), Judy Hoffman (Georgia Tech)

Object DetectionSegmentationDomain AdaptationImage

🎯 What it does: A training-time repair method called AUGCAL is proposed, which enhances SIM images significantly and incorporates calibration loss to improve the reliability and calibration of SIM2REAL adaptation models.

Augmented Bayesian Policy Search

Mahdi Kallel (University of Wrzburg), Carlo D'Eramo (University of Lille)

OptimizationRobotic IntelligenceReinforcement LearningSequential

🎯 What it does: A Bayesian optimization method (ABS) that uses the advantage function as the mean of the Gaussian process is designed, combined with an adaptive aggregation of multiple Q-function estimators to search for deterministic policy parameters.

Augmenting Transformers with Recursively Composed Multi-grained Representations

Xiang Hu (Ant Group), Wei Wu (Ant Group)

TransformerText

🎯 What it does: A recursive combination-enhanced Transformer model named ReCAT is proposed, which can explicitly model the hierarchical syntactic structure of the original text during learning and reasoning processes without relying on gold-standard trees.

AutoCast++: Enhancing World Event Prediction with Zero-shot Ranking-based Context Retrieval

Qi Yan (University of British Columbia), Tristan Sylvain (Borealis AI)

RetrievalRecommendation SystemTransformerLarge Language ModelTextRetrieval-Augmented Generation

🎯 What it does: Proposes AutoCast++, an event prediction framework based on zero-shot retrieval and abstract summarization.

AutoChunk: Automated Activation Chunk for Memory-Efficient Deep Learning Inference

Xuanlei Zhao (National University of Singapore), Yang You (HPC-AI Technology Inc)

OptimizationComputational EfficiencyNeural Architecture SearchProtein Structure PredictionTransformerImageTextSequentialBiomedical Data

🎯 What it does: The automated compiler system AutoChunk generates and applies chunk plans for long sequence inference, significantly reducing activation memory.

AutoDAN: Generating Stealthy Jailbreak Prompts on Aligned Large Language Models

Xiaogeng Liu (University of Wisconsin Madison), Chaowei Xiao (University of Wisconsin Madison)

OptimizationAdversarial AttackTransformerLarge Language ModelPrompt EngineeringText

🎯 What it does: Developed AutoDAN, a method for automatically generating covert jailbreak prompts based on a hierarchical genetic algorithm.

AutoLoRa: An Automated Robust Fine-Tuning Framework

Xilie Xu (National University of Singapore), Mohan Kankanhalli (National University of Singapore)

ClassificationKnowledge DistillationAdversarial AttackHyperparameter SearchConvolutional Neural NetworkTransformerSupervised Fine-TuningImage

🎯 What it does: The pre-trained feature extractor is fine-tuned adversarially through a low-rank decomposition (LoRa) branch, automatically decoupling adversarial and natural objectives, and achieving robust fine-tuning without manual hyperparameter tuning through automatic learning rate and loss weight scheduling.

Automatic Functional Differentiation in JAX

Min Lin (Sea AI Lab)

Optimization

🎯 What it does: This paper presents AutoFD, a system for automatic functional differentiation (AutoFD) implemented within the JAX framework;

AutomaTikZ: Text-Guided Synthesis of Scientific Vector Graphics with TikZ

Jonas Belouadi (Bielefeld University), Steffen Eger (University of Mannheim)

GenerationData SynthesisTransformerLarge Language ModelSupervised Fine-TuningImageText

🎯 What it does: A system called AutomaTiZ for automatically generating scientific graphics is proposed, utilizing TikZ as an intermediate language for generating text to vector graphics.

AutoVP: An Automated Visual Prompting Framework and Benchmark

Hsi-Ai Tsao (National Tsing Hua University), Tsung-Yi Ho (Michigan State University)

ClassificationHyperparameter SearchPrompt EngineeringImageBenchmark

🎯 What it does: This paper presents AutoVP, an end-to-end automated visual prompting framework designed to select the optimal input scaling, visual prompts, pre-trained models, and label mapping strategies to adapt to downstream image classification tasks.

Aux-NAS: Exploiting Auxiliary Labels with Negligibly Extra Inference Cost

Yuan Gao (Wuhan University), Jiayi Ma (South China University of Technology)

Object DetectionSegmentationDepth EstimationNeural Architecture SearchConvolutional Neural NetworkTransformerImage

🎯 What it does: Utilize auxiliary task labels to enhance the performance of the main task while maintaining the computational and parameter costs of single tasks during inference.

Backdoor Contrastive Learning via Bi-level Trigger Optimization

Weiyu Sun (Nanjing University), Lu Lin (Hong Kong University of Science and Technology)

OptimizationRepresentation LearningAdversarial AttackContrastive LearningImage

🎯 What it does: This study investigates efficient backdoor attacks in contrastive learning through dual-layer trigger optimization.

Backdoor Federated Learning by Poisoning Backdoor-Critical Layers

Haomin Zhuang (South China University of Technology), Xu Yuan (University of Delaware)

Federated LearningAdversarial AttackConvolutional Neural NetworkImage

🎯 What it does: Proposes the Layer Substitution Analysis method to identify backdoor-critical layers in FL and designs two types of hierarchical backdoor attacks, LP and LF, based on this.

Backdoor Secrets Unveiled: Identifying Backdoor Data with Optimized Scaled Prediction Consistency

Soumyadeep Pal (University of Alberta), Sijia Liu (Michigan State University)

Anomaly DetectionOptimizationImage

🎯 What it does: This paper proposes a Mask-Aware SPC method based on two-layer optimization for automatically identifying data contaminated by backdoors without the need for clean samples and thresholds.

BadChain: Backdoor Chain-of-Thought Prompting for Large Language Models

Zhen Xiang (University of Illinois Urbana-Champaign), Bo Li (University of Illinois Urbana-Champaign)

Adversarial AttackTransformerLarge Language ModelPrompt EngineeringTextChain-of-Thought

🎯 What it does: We propose BadChain, a backdoor attack targeting LLMs using chain-of-thought (COT) prompts, which can induce the model to produce malicious answers upon the appearance of trigger words without accessing the training set or model parameters.

BadEdit: Backdooring Large Language Models by Model Editing

Yanzhou Li (Nanyang Technological University), Yang Liu (Nanyang Technological University)

Adversarial AttackTransformerLarge Language ModelText

🎯 What it does: This paper proposes a problem of injecting backdoors into large language models (LLMs) as a lightweight model editing task, implanting backdoors by directly modifying a small number of parameters, avoiding the large-scale data and computational resources required by traditional weight/data poisoning methods.

BaDExpert: Extracting Backdoor Functionality for Accurate Backdoor Input Detection

Tinghao Xie (Princeton University), Prateek Mittal (Princeton University)

Anomaly DetectionConvolutional Neural NetworkTransformerSupervised Fine-TuningImage

🎯 What it does: A backend defense method is proposed, which fine-tunes the attacked model on a small number of mislabelled clean samples to extract its backdoor functionality and construct a backdoor expert model for detecting and filtering backdoor inputs.

Balancing Act: Constraining Disparate Impact in Sparse Models

Meraj Hashemizadeh (Mila), Jose Gallego-Posada (Mila)

OptimizationExplainability and InterpretabilityImage

🎯 What it does: A constraint optimization-based sparsification method CEAG is proposed, which directly controls the accuracy difference between the dense model and the sparse model across various subgroups, thereby alleviating the impact of differences caused by pruning.

Bandits Meet Mechanism Design to Combat Clickbait in Online Recommendation

Thomas Kleine Buening (Alan Turing Institute), Haifeng Xu (University of Chicago)

Recommendation SystemOptimizationReinforcement LearningTabular

🎯 What it does: A strategic click-based multi-armed bandit model is proposed, and the UCB-S mechanism is designed to balance learning and incentives;

Bandits with Replenishable Knapsacks: the Best of both Worlds

Martino Bernasconi (Bocconi University), Federico Fusco (Sapienza University)

OptimizationReinforcement LearningFinance Related

🎯 What it does: This paper studies an extended online decision model that allows for non-monotonic consumption of resources, meaning that resources can be replenished over time.