Solving the Identifying Code Set Problem with Grouped Independent Support
Anna L.D. Latour (National University of Singapore), Kuldeep S. Meel (National University of Singapore)
CodeOptimizationGraph
🎯 What it does: Studied the generalized identifying code set (GICS) problem for sensor placement in networks, and proposed a method to reduce it to the independent support problem of Boolean formulas through grouped independent support (GIS), thereby solving the minimal sensor set.
🎯 What it does: Proposed a spatiotemporal self-attention mechanism (STSA) that maintains asynchronous characteristics and a relative position information bias (STRPB), constructing an STS-Transformer model based on these modules for event-driven spiking neural networks (SNNs);
Specifying and Testing k-Safety Properties for Machine-Learning Models
Maria Christakis (MPI-SWS), Valentin Wüstholz (ConsenSys)
CodeSafty and PrivacyImageTextTabularAudio
🎯 What it does: This paper proposes a new specification language called NOMOS for specifying and testing k-safety properties of machine learning models, and demonstrates its broad applicability across multiple domains.
Speeding Up Multi-Objective Hyperparameter Optimization by Task Similarity-Based Meta-Learning for the Tree-Structured Parzen Estimator
Shuhei Watanabe (University of Freiburg), Frank Hutter (University of Freiburg)
CodeHyperparameter SearchMeta LearningBenchmark
🎯 What it does: Propose a meta-learning method for multi-objective tree-structured Parzen estimator (MO-TPE), which weights high-quality configurations within tasks based on task similarity to accelerate hyperparameter optimization.
Sph2Pob: Boosting Object Detection on Spherical Images with Planar Oriented Boxes Methods
Xinyuan Liu (Key Laboratory of Intelligent Information Processing of Chinese Academy of Sciences, Institute of Computing Technology, Chinese Academy of Sciences), Feng Dai (Key Laboratory of Intelligent Information Processing of Chinese Academy of Sciences, Institute of Computing Technology, Chinese Academy of Sciences)
CodeObject DetectionImage
🎯 What it does: The paper proposes a spherical-to-planar box transformation method called Sph2Pob, and based on this, introduces differentiable Sph2Pob-IoU and flexible, scalable Sph2Pob-Loss, significantly improving the performance of object detection on spherical images.
🎯 What it does: Proposes an SNN training framework based on Output Spike Count Maximization (SCM), combining structural risk minimization and a specially designed spike count loss to form a two-stage iterative training algorithm.
🎯 What it does: Proposed a news-driven quantitative trading framework based on reinforcement learning called SpotlightTrader, which generates continuous and flexible trading decisions using a trajectory optimization model, and introduces illumination news screening and state trajectory modeling, combined with a training pipeline of offline pre-training and online fine-tuning.
🎯 What it does: Constructed a large-scale, multi-accent naturally recorded speech reading comprehension dataset named SQuAD-SRC, and conducted experiments on question answering tasks involving multi-accent speech questions and textual context.
StackFLOW: Monocular Human-Object Reconstruction by Stacked Normalizing Flow with Offset
Chaofan Huo (ShanghaiTech University), Jingya Wang (ShanghaiTech University)
CodePose EstimationFlow-based ModelImage
🎯 What it does: This paper proposes a spatial relationship encoding based on Human-Object Offset, and utilizes Stacked Normalizing Flow to infer human-object relative poses from a single image. Subsequently, monocular 3D reconstruction of human-object interactions is achieved through optimization.
Stochastic Feature Averaging for Learning with Long-Tailed Noisy Labels
Hao-Tian Li (Southeast University), Min-Ling Zhang (Southeast University)
CodeClassificationImage
🎯 What it does: This paper proposes a framework based on Stochastic Feature Averaging (SFA) to simultaneously address learning problems under long-tailed distributions and label noise.
Yuning Cui (Technical University of Munich), Alois Knoll (Technical University of Munich)
CodeRestorationConvolutional Neural NetworkImage
🎯 What it does: Proposed a Strip Attention Network (SANet) that achieves efficient image restoration through horizontal and vertical local attention mechanisms, replacing traditional global self-attention.
🎯 What it does: This paper proposes a method for achieving finer-grained text-video retrieval by decomposing video and text features into multi-dimensional latent concepts and performing adaptive pooling.
TITAN : Task-oriented Dialogues with Mixed-Initiative Interactions
Sitong Yan (Xidian University), Guangneng Hu (Xidian University)
CodeGenerationTransformerLarge Language ModelSupervised Fine-TuningTextBenchmark
🎯 What it does: In this study, the authors constructed a multi-domain task-oriented dialogue dataset called TITAN, which includes system-side proactive interaction strategies, and evaluated multiple baseline models for response generation and dialogue behavior prediction on it.
🎯 What it does: Propose a fully dynamic hypergraph neural network (TDHNN) that can learn hyperedge feature distribution during training and dynamically adjust the number of hyperedges based on this distribution to achieve a more suitable hypergraph structure.
Toward Convex Manifolds: A Geometric Perspective for Deep Graph Clustering of Single-cell RNA-seq Data
Nairouz Mrabah (University of Quebec at Montreal), Abdoulaye Banire Diallo (University of Quebec at Montreal)
CodeRepresentation LearningGraph Neural NetworkAuto EncoderGenerative Adversarial NetworkBiomedical Data
🎯 What it does: Designed and implemented a deep graph clustering model called scTCM for single-cell RNA-seq data, improving the geometry of the latent space by incorporating local flattening and global convexification mechanisms during pre-training and clustering stages.
Towards a Better Understanding of Learning with Multiagent Teams
David Radke (University of Waterloo), Kyle Tilbury (University of Waterloo)
CodeReinforcement Learning
🎯 What it does: This paper studies the impact of multi-agent team structure on individual learning processes, combining theoretical analysis and experiments to evaluate the optimal team size.
Towards Accurate Video Text Spotting with Text-wise Semantic Reasoning
Xinyan Zu (Fudan University), Xiangyang Xue (Fudan University)
CodeRecognitionObject DetectionObject TrackingSuper ResolutionTransformerLarge Language ModelVision Language ModelContrastive LearningVideo
🎯 What it does: Proposed a video text recognition framework called VLSpotter, which combines text super-resolution, language models, and inter-text semantic reasoning to achieve end-to-end video text detection, tracking, and recognition.
🎯 What it does: This study introduces the Collaborative Plan Acquisition task for the first time on the existing MindCraft benchmark, aiming to enable agents to predict and complete missing task knowledge through multimodal dialogue and visual information when facing partial plans from themselves and their partners, while exploring the integration of dialogue actions and Theory of Mind (ToM) modeling into this process.
Towards Generalizable Reinforcement Learning for Trade Execution
Chuheng Zhang (Microsoft Research), Li Zhao (Microsoft Research)
CodeReinforcement LearningTime SeriesFinance Related
🎯 What it does: This paper proposes a framework named Offline Reinforcement Learning with Dynamic Context (ORDC) to address the generalization challenges in learning trading execution strategies from limited offline market data, and provides theoretical generalization bounds. Based on this framework, two context aggregation methods (CASH, handcrafted statistical guidance, and CATE, end-to-end learning) and a high-fidelity simulator based on historical order book (LOB) data are designed. Experiments on simplified tasks and real stock LOB data validate the effectiveness of these methods in reducing overfitting, improving transaction cost performance, and enhancing generalization capabilities.
🎯 What it does: Propose a Director-Actor Hierarchical Conversation Recommendation framework (DAHCR), which significantly improves the effectiveness of multi-round dialogue recommendations by separating inquiry and recommendation decisions, dynamically modeling user preferences with hypergraphs, and guiding weak supervision through intrinsic motivation.
🎯 What it does: Propose an adaptive reward reallocation algorithm named DTRD to address the performance degradation of Decision Transformer in long-term delayed reward environments.
Towards Robust Scene Text Image Super-resolution via Explicit Location Enhancement
Hang Guo (Tsinghua Shenzhen International Graduate School, Tsinghua University), Shu-Tao Xia (Tsinghua Shenzhen International Graduate School, Tsinghua University)
CodeRecognitionSuper ResolutionTransformerVision Language ModelImage
🎯 What it does: Propose a scene text image super-resolution method named LEMMA, which distinguishes text from the background by leveraging explicit character position enhancement, bidirectional multimodal alignment, and adaptive fusion to improve recognition performance.
🎯 What it does: Propose TPS++, an attention-enhanced Thin-Plate Spline (TPS) transformation for preprocessing in scene text recognition, improving traditional TPS's control point regression and content-agnostic parameter estimation.
🎯 What it does: Proposes an unsupervised domain adaptation multi-object tracking framework (DA-Tracker) that can transfer across different ant species and experimental environments, and first constructs a large-scale ant tracking dataset tailored for this task.
🎯 What it does: Proposed U-Match, an attention graph neural network for two-view matching learning, which can implicitly capture local context at multiple levels and fuse global information;
Unbiased Risk Estimator to Multi-Labeled Complementary Label Learning
Yi Gao (Southeast University), Min-Ling Zhang (Southeast University)
CodeClassificationOptimization
🎯 What it does: This paper proposes an unbiased risk estimator for multi-label complementary label learning (MLCLL) and provides an upper bound on the estimation error.
Understanding the Generalization Ability of Deep Learning Algorithms: A Kernelized Rényi's Entropy Perspective
Yuxin Dong (Xi'an Jiaotong University), Chen Li (Xi'an Jiaotong University)
CodeOptimizationExplainability and InterpretabilityImage
🎯 What it does: Propose a new information measure called kernel-based Rényi entropy, and use it to perform computable theoretical analysis of the generalization error in SGD and SGLD algorithms in deep learning.
🎯 What it does: Proposed and implemented a unified method (UADA) that adaptively updates data augmentation parameters during training based on the target model's gradient.
Unreliable Partial Label Learning with Recursive Separation
Yu Shi (Southeast University), Xin Geng (Southeast University)
CodeClassificationData-Centric LearningImageText
🎯 What it does: This paper proposes a two-stage framework for learning with partially unreliable labels called UPLLRS. It first uses adaptive recursive separation to divide the data into reliable and unreliable subsets. Then, it performs label disambiguation on the reliable subset and employs semi-supervised learning on the unreliable subset to enhance model performance.
CodeExplainability and InterpretabilityRepresentation LearningSequential
🎯 What it does: By conducting concept probing and global explanation on the Stockfish NNUE evaluation model, revealing its internally learned chess concepts and comparing them with traditional handcrafted evaluation models.
VecoCare: Visit Sequences-Clinical Notes Joint Learning for Diagnosis Prediction in Healthcare Data
Yongxin Xu (Key Laboratory of High Confidence Software Technologies Ministry of Education), Bing Xie (Key Laboratory of High Confidence Software Technologies Ministry of Education)
CodeTransformerLarge Language ModelContrastive LearningTextSequentialElectronic Health Records
🎯 What it does: This paper proposes a joint learning method called VecoCare to fuse access sequences and clinical notes in medical data, thereby improving the accuracy of diagnostic predictions.
🎯 What it does: Propose a locally-global context-guided autoregressive video diffusion model (LGC-VD) for video prediction, interpolation, and unconditional generation.
🎯 What it does: This work proposes an inter-frame interpolation method based on dense bidirectional correlation, achieving high-quality video frame interpolation by modeling correlation between two frames at a single high resolution, combined with motion generation and refinement modules, as well as a lightweight synthesis network.
🎯 What it does: Designed the Violin framework, which utilizes learning-generated virtual bridges (VOs) to associate unlabelled nodes with labelled nodes, thereby expanding the receptive field of GNNs and enhancing their representation capability.
🎯 What it does: Developed a reversible neural flow network called WBFlow for white balancing sRGB images, achieving fast generalization across multiple cameras through few-shot learning.
Why Rumors Spread Fast in Social Networks, and How to Stop It
Ahad N. Zehmakan (Australian National University), Sajjad Hesamipour Khelejan (Trinity College Dublin)
CodeAnomaly DetectionGraph
🎯 What it does: This paper proposes a rumor propagation model integrating IC, Push-Pull, and SIR, incorporating trust weights based on Jaccard similarity and a time-decaying forgetting mechanism. It analyzes the impact of graph structure (expansiveness and community connectivity) on propagation speed and range from both theoretical and experimental perspectives, while designing and evaluating six countermeasures to suppress rumors.